The motivation is more than convenience. Repeated vaginal examinations remain the standard method for assessing labor progress, yet they are uncomfortable, carry an infection risk, and vary from one examiner to the next. Qualitative work has found that many women associate the procedure with pain, fear, and embarrassment, and a regional study in Latin America reported that nearly one in three women underwent five or more examinations during labor, a frequency considered excessive by current guidelines. Meanwhile, dystocia, a failure of the normal mechanisms of cervical dilation or fetal descent, drives an estimated 60 to 80 percent of cesarean sections performed for labor arrest. An objective, continuous, and painless way to gauge dilation could therefore reshape how clinicians detect abnormal labor before mother or baby is in danger.

The researchers turned to electrohysterography, or EHG, a technique that records the electrical activity of the uterine muscle through electrodes placed on the abdominal surface, much as a cardiogram listens to the heart. Between 2017 and 2019, the team assembled 72 ten-minute EHG recordings from women in low-risk term or moderate preterm labor at two Mexican hospitals: the Maternal and Childhood Research Center in Mexico City and the Mónica Pretelini-Sáenz Maternal-Perinatal Hospital in Toluca. Signals were captured with a Monica AN24 trans-abdominal monitor at 900 Hz using disposable electrodes, then exported through a single bipolar channel down-sampled to 20 Hz and band-limited to 0.2 to 1 Hz. Each participant contributed one recording, paired with a cervical dilation value documented by digital vaginal examination shortly beforehand.

The engineering challenge lay in translating raw electrical traces into numbers a regression model could digest. The team split each recording into three frequency bands and computed seven linear and nonlinear metrics in each: root-mean-square amplitude, the area under the rectified envelope, zero-crossing rate, mean and median frequencies, sample entropy, and bubble entropy. These 21 base metrics were calculated across nine overlapping 120-second windows within each recording, and eight temporal descriptors, including mean, slope, maximum, standard deviation, steepest positive and negative changes, total variation, and accumulated area, were then derived to capture how each metric evolved over the ten minutes. Combined with counts of low- and high-intensity contractions, maternal age, and gestational age, the pipeline produced a 172-feature vector per recording. Five recordings with sparsely represented dilation values outside the 4 to 8 cm range were excluded, leaving 67 records: 50 for model development and 17 held back for internal testing.

Model development followed a deliberately sequential blueprint. Four feature-ranking algorithms, minimum redundancy maximum relevance, the F-test, neighborhood component analysis, and regression-tree importance, were compared, and a composite ranking identified the five strongest candidates. Twenty-eight regression algorithms available in MATLAB’s Regression Learner app were then benchmarked with fivefold cross-validation. Their baseline errors clustered tightly between 1.324 and 1.331 cm, with coefficients of determination near zero or negative, indicating that the top-ranked features alone explained little of the variability in dilation. The team retained the squared-exponential Gaussian process regression model, or SE-GPR, for further optimization because it produces a smooth, continuous, nonlinear fit and can explicitly account for observation uncertainty, an important consideration when the reference values come from manual exams reported mostly in whole centimeters.

Because simple sequential feature selection proved insufficient, the researchers unleashed a genetic algorithm on the 172-dimensional search space, an optimization strategy inspired by evolution in which candidate feature subsets are encoded as binary chromosomes, scored by a composite fitness function balancing prediction error, correlation, and feature count, and iteratively refined through selection, crossover, and mutation. After 98 generations, the algorithm converged on a lean subset of five EHG-derived temporal features: the maximum difference in mean frequency from band one, the maximum median frequency and zero-crossing statistics from band two, the steepest negative change in zero-crossing rate from band two, and the accumulated area of the lower-frequency envelope from band three. Notably, none of the clinical variables survived the selection, suggesting the electrical signal itself carried the most informative, if fragile, relationship to dilation.

The optimized model’s performance told a two-part story. Within the development set, fivefold cross-validation yielded a root-mean-square error of 0.97 cm, a Pearson correlation of 0.67, and an R-squared of 0.45, a respectable showing for a continuous centimeter-scale prediction. But when the finalized model faced the 17 recordings it had never seen, performance collapsed: the error rose to 1.45 cm, the correlation fell to 0.23, and the R-squared turned negative at minus 0.31, meaning the model did worse than simply guessing the average dilation. The authors are candid about this gap. The small held-out sample inflates uncertainty, and the quantized, one-centimeter resolution of the clinical reference values amplifies the effect of small prediction deviations. Still, the negative R-squared signals that the moderate association observed during development did not generalize to unseen recordings.

Several physiological threads nonetheless emerge from the selected features. The chosen predictors describe the magnitude and temporal variation of the EHG spectrum, changes in signal oscillation within the 0.34 to 1 Hz fast-wave band, and the accumulated magnitude of the lower-frequency component. These align with prior observations that uterine electrical activity becomes more organized as labor advances, with entropy falling and burst frequencies rising from roughly 0.41 Hz below 3 cm dilation to about 0.52 Hz at 6 cm or more. The authors caution, however, that because the features were mined from a small, high-dimensional dataset, they should be treated as hypotheses rather than confirmed biomarkers of cervical dilation.

The study’s limitations are clearly drawn. The analysis rests on 67 recordings from just two Mexican hospitals, covers only the 4 to 8 cm dilation range, relies on a single bipolar EHG lead, and assigns one dilation value per ten-minute recording rather than tracking continuous change. Dystocia was never labeled or modeled, so no screening performance can be claimed, and the model produces one estimate per recording, not a second-by-second readout. Yet the strengths are equally real: an openly shared dataset on Zenodo, a systematic model-selection pipeline, rigorous separation of development and test data at the recording level, and a refusal to overstate the results. The authors call for larger prospective multicenter cohorts spanning the full dilation spectrum, multichannel EHG arrays, combinations with ultrasound elastography, and explainable-AI methods to clarify each predictor’s contribution.

If those validation efforts succeed, the payoff could be substantial: a wearable, electrode-based adjunct that flags abnormal labor progression early, prompts timely clinical reassessment, and reduces the number of invasive examinations women endure, lowering infection risk and discomfort in the process. For now, the message of this exploratory study is measured. The electrical signature of the laboring uterus does appear to carry information about cervical dilation, and modern machine learning can begin to decode it, but the road from proof of concept to clinical tool runs through many more patients, hospitals, and hours of data.

Subject of Research: Non-invasive machine-learning estimation of cervical dilation during labor from electrohysterography and maternal clinical data

Article Title: Exploratory Non-invasive Estimation of Cervical Dilation from Electrohysterography and Maternal Data: A Machine-Learning Proof-of-Concept Study

Article References: Portillo-Rodríguez, O., Escalante-Gaytán, J., Sandoval-González, O. O., Soria, P. R., Mendieta-Zerón, H., Echeverría, J. C., Peña-Castillo, M. Á., Abarca-Castro, E. A., & Reyes-Lagos, J. J. (2026). Exploratory Non-invasive Estimation of Cervical Dilation from Electrohysterography and Maternal Data: A Machine-Learning Proof-of-Concept Study. Annals of Biomedical Engineering. https://doi.org/10.1007/s10439-026-04349-6

Image Credits: AI Generated

DOI: 10.1007/s10439-026-04349-6

Keywords: electrohysterography, cervical dilation, machine learning, labor monitoring, obstetrics, dystocia, Gaussian process regression, genetic algorithm, uterine electromyography, non-invasive monitoring, vaginal examination, biomedical engineering

The research, led by Mingjing Lu, Zhongjian Xu, and Tang Hanfei, with senior authors Dehai Lang, Guo Daqiao, and Guofu Wang, took a deliberately dual approach. Rather than betting on a single machine learning paradigm, the investigators built and compared two complementary pipelines. The first was a classical radiomics pipeline, in which hand-engineered quantitative features—describing the shape, texture, and intensity distribution of the plaque—are extracted from manually defined regions of interest on CT angiography images. The second was a deep learning pipeline, in which convolutional neural networks learn discriminative image patterns directly from pixel data without explicit feature engineering. Both pipelines were trained to answer the same clinically critical question: is this plaque histologically stable or vulnerable, as confirmed by pathology after carotid endarterectomy?

The evidence base for the radiomics arm came from a retrospective multi-center cohort of 260 consecutive patients who underwent carotid endarterectomy, meaning that every plaque in the dataset had a gold-standard pathological diagnosis of stability or vulnerability. Of these, 200 patients were used for model development and internal validation, while 60 patients were held out entirely for external validation, a design choice that guards against the optimistic performance estimates that plague many machine learning studies in medicine. For the deep learning arm, the team assembled an independently annotated CT angiography dataset comprising 236 cases and a remarkable 7,394 individual regions of interest, providing the volume of labeled image data needed to train and test convolutional architectures meaningfully.

On the technical side, the radiomics workflow followed rigorous feature-selection discipline. Candidate radiomic features were screened using analysis of variance combined with Kruskal–Wallis testing to identify features that discriminated between stable and vulnerable plaques, followed by correlation filtering to remove redundant measures, and finally least absolute shrinkage and selection operator, or LASSO, regularization to compress the feature set to its most predictive and non-redundant core. Five classifier families were then trained on the selected features: Random Forest, Support Vector Machine, k-Nearest Neighbor, Naïve Bayes, and Logistic Regression. Performance was quantified using receiver operating characteristic analysis and the area under the curve, complemented by calibration analysis, which asks not merely whether a model ranks patients correctly but whether its predicted probabilities match observed reality—a property that matters enormously when model outputs are meant to inform surgical decisions.

The radiomics results were strikingly consistent. On the internal test set, the Random Forest model achieved an AUC of 0.858, followed closely by the Support Vector Machine at 0.857, Naïve Bayes at 0.855, k-Nearest Neighbor at 0.843, and Logistic Regression at 0.821, with small gaps between training and test performance that suggest the models were capturing genuine biological signal rather than memorizing noise. Crucially, this discrimination survived the move to entirely external data. In the 60-patient external validation cohort, the Support Vector Machine led with an AUC of 0.839, Logistic Regression reached 0.835, k-Nearest Neighbor 0.830, and both Naïve Bayes and Random Forest achieved 0.817. Calibration curves showed close agreement between predicted and observed probabilities across the model families, indicating that the classifiers were not only separating stable from vulnerable plaques but doing so with probabilities a clinician could reasonably act upon.

The deep learning arm told a complementary story. The authors trained a multilayer perceptron, a from-scratch convolutional neural network, and three ImageNet-pretrained transfer-learning backbones—VGG16, VGG19, and ResNet50—on the annotated regions of interest. Transfer learning, in which networks pre-trained on millions of natural images are fine-tuned on medical data, proved advantageous: VGG16 achieved the best test AUC of 0.767, ResNet50 followed at 0.754, and VGG19 reached 0.718, while the multilayer perceptron and the from-scratch CNN trailed at 0.669 and 0.654 respectively. External validation reproduced this ordering almost exactly, with VGG16 at 0.776, ResNet50 at 0.751, VGG19 at 0.719, and the simpler architectures in the high-0.68 range. Although the deep models did not surpass the radiomics classifiers in raw discrimination, their performance held up across centers, and they brought a distinct advantage: interpretability through attention mapping.

Using Gradient-weighted Class Activation Mapping, or Grad-CAM, the team visualized which parts of each image the networks attended to when making their predictions. The resulting heatmaps localized hyperattenuating regions—denser, brighter areas within the plaque that correspond to features such as calcification or hemorrhage—and these localizations were consistent with the pathological findings, most clearly for VGG16 and ResNet50. This alignment between machine attention and histological ground truth is more than a technical curiosity. It provides a sanity check that the networks are not exploiting scanner-specific artifacts or incidental image features, but are genuinely looking at the plaque biology that pathologists confirm under the microscope. For a field often criticized for black-box opacity, such visual evidence of anatomically plausible reasoning is a meaningful step toward clinical trust.

The clinical implications are considerable. Current decision-making for carotid stenosis leans heavily on the degree of luminal narrowing, yet the literature has long shown that plaque composition and stability, not stenosis alone, determine rupture risk. A patient with moderate narrowing but a vulnerable plaque may face a higher stroke risk than a patient with severe narrowing and a stable lesion. A validated, automated tool that reads a routine CT angiography scan—the same scan already obtained during standard preoperative workup—and reports a calibrated probability of plaque vulnerability could therefore reshape preoperative risk stratification, help prioritize patients for carotid endarterectomy or stenting, and guide the intensity of medical therapy and surveillance for those managed conservatively. Because the radiomics models in this study used conventional machine learning classifiers on compact feature sets, they are also lightweight and deployable, requiring no exotic hardware and lending themselves to integration into existing picture archiving and communication systems.

The authors are appropriately measured about what their framework can and cannot yet do. The study is retrospective, and the deep learning models, while externally validated, did not reach the discrimination levels of the best radiomics classifiers, suggesting that hand-crafted quantitative descriptors of plaque texture and morphology still carry information that current convolutional architectures do not fully capture from cropped regions of interest. The natural next steps include prospective validation in consecutive clinical cohorts, testing across scanner vendors and imaging protocols, and exploration of hybrid models that fuse radiomic features with deep-learned representations. Funding for the work came from the Shaoxing City Science and Technology Plan Project 2023, and the study was conducted under the Declaration of Helsinki with institutional ethics approval and informed consent from all participants. Even with those caveats, the convergence of pathology-validated labels, multi-center external testing, strong calibration, and interpretable attention maps makes this one of the more convincing demonstrations that AI-assisted plaque stability assessment is moving from proof of concept toward a genuinely usable instrument in the fight against ischemic stroke.

Subject of Research: AI-based carotid plaque stability assessment on CT angiography for ischemic stroke risk prediction

Article Title: Integrated radiomics and image-based deep learning framework using carotid CT angiography ROI datasets for plaque stability assessment as a predictor of ischemic stroke risk

Article References: Lu, M., Xu, Z., Hanfei, T., Luo, P., Huang, F., Wang, G., Bi, L., Jiang, N., Lang, D., Daqiao, G., & Wang, G. (2026). Integrated radiomics and image-based deep learning framework using carotid CT angiography ROI datasets for plaque stability assessment as a predictor of ischemic stroke risk. BioMedical Engineering OnLine. https://doi.org/10.1186/s12938-026-01621-7

Image Credits: AI Generated

DOI: 10.1186/s12938-026-01621-7

Keywords: carotid CT angiography, radiomics, deep learning, plaque stability, ischemic stroke, stroke risk stratification, carotid endarterectomy, machine learning, VGG16, ResNet50, Grad-CAM, biomedical engineering

The new platform, described in the open-access journal HardwareX by Jerry Febrico and Basari of Universitas Indonesia, is built around two inexpensive, widely available integrated circuits: the STM32F407VGT6 microcontroller and the AD5933 impedance converter. EIT works by injecting a small, safe alternating current into a subject or phantom through an array of surface electrodes and measuring the faint voltages that appear at the boundary. Because different tissues conduct electricity differently, the pattern of these boundary voltages encodes information about the internal distribution of conductivity, which a computer algorithm can then reconstruct into a cross-sectional image. For the lungs, this is particularly powerful: as air flows in and out during breathing, the local impedance of lung tissue changes dramatically, allowing EIT to track regional ventilation at the bedside without a single X-ray. Diseases such as chronic obstructive pulmonary disease and pulmonary edema, which demand repeated monitoring rather than one-off snapshots, are precisely the conditions where this continuous, non-ionizing approach holds the greatest clinical promise.

What distinguishes the new design from earlier open-source efforts is its deliberately modular analog front-end. The heart of the system is the AD5933, a chip that combines a direct digital synthesizer signal generator and a 12-bit impedance converter on a single die. In this implementation, the AD5933 is configured to output a sinusoidal excitation signal at 50 kilohertz, which is fed into a voltage-controlled current source that injects a constant current of roughly 1.44 milliamperes peak-to-peak into the imaging domain. On the receive side, weak differential voltages picked up from the electrodes are amplified by an instrumentation amplifier and routed back into the AD5933, whose on-chip digital signal processor applies a discrete Fourier transform to extract the real and imaginary components of the measured signal. The STM32 microcontroller orchestrates the entire sequence over an I2C bus, computes impedance magnitudes, and streams the results over USB to a personal computer running Python-based reconstruction software.

The modularity is not an abstract design philosophy; it is physically built into the circuit board. The single custom PCB, measuring 177.8 by 155.83 millimeters, carries two alternative voltage-controlled current source topologies, a Load-in-the-Loop design and a Mirrored Howland current pump, alongside two instrumentation amplifier options, a discrete INA118P chip and an AD620 module. Manual DIP switches on the board let researchers route the signal through any combination of these circuits, enabling direct, empirical comparison of four distinct analog front-end configurations under identical operating conditions. The board is assembled with IC sockets, female headers, and through-hole resistors specifically so that users can swap components, replace damaged parts, and experiment with their own circuit modifications without fabricating a new board. Every component is listed in a bill of materials with purchase links from global suppliers such as DigiKey and AliExpress, and the complete design files, from EasyEDA schematics to PCB Gerber files, are released under CERN-OHL, MIT, and CC BY 4.0 open-source licenses.

Data acquisition follows the classic adjacent four-terminal sensing pattern used across the EIT field. Four 16-channel CD74HC4067 analog multiplexers, steered by the microcontroller, dynamically connect the current source and the measurement amplifier to a ring of 16 stainless steel electrodes mounted on a 3D-printed polylactic acid container. Current is injected through one adjacent electrode pair while differential voltages are measured sequentially across all remaining neighboring pairs; the injection pair then rotates around the ring, and the process repeats. One complete sweep yields 208 impedance measurements, which the firmware collects in just two seconds. The authors tested alternative injection patterns, including opposite and diagonal schemes that push current deeper into the domain, but found in their comparative experiments that the adjacent pattern best preserved the shape and location of target inclusions, while cross and opposite patterns introduced excessive noise that obscured objects entirely.

Validation of the hardware was thorough and refreshingly candid. Signal-to-noise ratio measurements, taken across 50 consecutive frames in a saline-filled phantom with a conductivity of 966 millisiemens per meter, ranged from 21.5 to 43.7 decibels depending on the front-end configuration. The Mirrored Howland paired with the AD620 achieved the highest peak SNR of 43.7 decibels, a figure the authors note is comparable to the clinically validated Sheffield Mk 3.5 system, which operated at roughly 40 decibels. The Load-in-the-Loop configurations delivered more uniform channel-to-channel performance, around 37 to 38 decibels, because this topology does not depend on precisely matched resistor networks, whereas the Howland pump’s high output impedance degrades rapidly when built with standard 1 percent tolerance components. Precision testing with standard resistors from 100 to 1000 ohms revealed exceptionally tight repeatability, with standard deviations between 0.54 and 6.33 ohms across all configurations and coefficients of determination reaching as high as 0.9997.

Perhaps the most instructive finding concerns absolute accuracy. The system exhibited a systematic positive offset, producing relative errors that ranged from 0.60 percent to as much as 37.82 percent for the lowest resistor values. In most measurement contexts, such errors would be disqualifying. But EIT, as practiced here, relies on time-difference imaging: a baseline reference dataset is captured in a homogeneous medium, and images are reconstructed from the changes that occur relative to that baseline. Static hardware offsets, parasitic capacitances, and multiplexer channel resistances are mathematically subtracted during this differential process, which means repeatability, not absolute accuracy, is the metric that matters. By this standard, the platform performs admirably, and the authors are transparent that the fixed additive error, traced through linear regression to positive y-intercepts between 11.2 and 99.4 ohms, is effectively cancelled by the reconstruction algorithm.

Imaging experiments confirmed the system’s practical capability. Using cylindrical phantom targets, one highly conductive stainless steel tube and one non-conductive plastic tube, placed at various positions inside the saline tank, the team reconstructed 2D images using three algorithms available in the open-source pyEIT library: Back-Projection, the Jacobian-based JAC method, and GREIT. Back-Projection was favored for its speed and low computational cost. Quantitative comparison against simulated ground-truth images, scored with the Structural Similarity Index Measure and Root Mean Square Error, showed strong reconstruction fidelity, with the HOW-INA118 configuration achieving a peak SSIM of 0.9242 and the HOW-AD620 configuration delivering the lowest RMSE of 28.29. The reconstructed inclusions showed the characteristic boundary smearing and spatial deformation inherent to adjacent-drive EIT with linear reconstruction, particularly near the center of the domain, but the system reliably detected and localized both conductive and non-conductive anomalies across all four hardware configurations and all tested positions.

The team is explicit that this prototype is intended strictly for education and non-human phantom experiments; it lacks the medical-grade patient isolation required by the IEC 60601-1 safety standard, and no commercial analog currently exists at this price point. Yet the significance of the work extends well beyond the laboratory bench. Existing open-source EIT systems that support standard 16-electrode arrays typically cost between 250 and 330 dollars, while cheaper designs sacrifice electrode count or capability. By delivering a full 16-electrode, 50-kilohertz system with an end-to-end open-source software pipeline, including a Tkinter-based graphical interface, STM32 firmware, 3D-printable container files, and even a step-by-step video calibration tutorial, for under 100 dollars, the researchers have lowered the barrier to entry for students and early-career researchers to a fraction of its previous level. Future plans include upgrading to next-generation impedance converters with multi-frequency capability, compressing acquisition below one second for true real-time respiration monitoring, boosting SNR with active-shielded cabling, and exploring generative deep-learning models such as conditional variational autoencoders and diffusion models to push image quality beyond the physical limits of conventional reconstruction. For a field whose clinical value has long been constrained by hardware cost and complexity, a 95-dollar, fully open imaging platform is a genuinely democratizing development.

Subject of Research: A portable, low-cost, open-source electrical impedance tomography system with a modular analog front-end based on an STM32 microcontroller and AD5933 impedance converter.

Article Title: A portable, low-cost, and open-source electrical impedance tomography system with a modular analog front-end based on STM32 and AD5933

Article References: Febrico, J., & Basari (2026). A portable, low-cost, and open-source electrical impedance tomography system with a modular analog front-end based on STM32 and AD5933. HardwareX, Article e00836. https://doi.org/10.1016/j.ohx.2026.e00836

Image Credits: AI Generated

DOI: 10.1016/j.ohx.2026.e00836

Keywords: electrical impedance tomography, open-source hardware, STM32, AD5933, lung imaging, biomedical engineering, analog front-end, pyEIT, medical imaging, respiratory monitoring, low-cost instrumentation, phantom experiments

The paper, written by Marco Viceconti of the Department of Industrial Engineering at the University of Bologna, introduces what the author calls the 7S Framework, a seven-step general procedure for evaluating the credibility of any quantitative estimate, whether it originates from a sensor, a statistical model, or a predictive simulation. The motivation is practical rather than purely philosophical. In fields such as in silico medicine, where computational models increasingly inform clinical decisions and regulatory submissions, a new generation of tools refuses to sit neatly within any one of the traditional categories. In silico-augmented clinical trials, physics-informed machine learning predictors, and machine learning models trained on synthetic datasets all blend measured data, statistical inference, and causal prediction into single estimators, leaving established credibility frameworks unable to cover them cleanly.

To build the framework, Viceconti begins with a conceptual scaffolding often described as the pyramid of knowledge. In this picture, observation lifts raw signals produced by a system of interest into data; annotation with metadata about who, what, where, and when lifts data into information; modelling correlations lifts information into tentative causal beliefs; and subjecting those beliefs to falsification experiments lifts them into actionable knowledge. Quantitative information, in this scheme, is an annotated set of values whose metadata specifies the domain of information, everything sender and receiver must know for the values to be meaningful, but without the causal “why” that would elevate it to knowledge. New information can be created by measurement, which converts signals into data; by inference, which derives new information from existing information; or by prediction, which uses causal knowledge to generate estimates of quantities that were never observed.

From these foundations, the author generalises a vocabulary that statistics normally reserves for inference. The target quantity is the estimand, the value produced is the estimate, and whatever produces it, be it a thermometer, a regression, or a finite element model, is an estimator. Credibility is then defined with a deliberately demanding definition: the minimum accuracy with which an estimator recovers the true value of the estimand across the entire information space, the bounded multidimensional region defined by all the observable quantities on which the quantity of interest depends. Accuracy itself borrows from metrology, where trueness captures systematic error and precision captures random error, combined into a normalised class of accuracy averaged over repeated estimations.

A crucial insight of the paper is that credibility expectations come in three levels, and that these levels are properties of the intended use rather than of the type of estimator. Level 1 credibility demands only that an estimate fall within a predefined uncertainty band around the true value, appropriate when knowing the order of magnitude suffices. Level 2 requires that the average of repeated estimates match the true value in a statistical sense, as when comparing central properties of populations. Level 3 demands local accuracy at every validation point, the standard for subject-specific models intended to predict individual outcomes. A predictive model can therefore be L1, L2, or L3 credible depending on whether it is meant to capture a scale, a population mean, or a person-specific value, and the same hierarchy applies to measurement and inference.

Because brute-force induction, measuring the error at an effectively infinite number of points, is practically impossible and not even theoretically guaranteed for all estimators, the framework follows the strategy shared by metrology, statistics, and VVUQ: decompose the estimation error into its sources and check that each component behaves as theory predicts for a well-behaved estimator. The seven steps formalise this logic. Step S1 defines the context of use and the acceptable error threshold, the maximum error that still leaves the information useful for the decision it must support, and sets this against the limits of validity imposed by the physics of the phenomenon. Step S2 establishes the source of true values, insisting on measurement chains at least an order of magnitude more accurate than the threshold. Step S3 quantifies estimation error through controlled experiments sampled across the solution space. Step S4 identifies the sources of error, which the paper groups into approximation, aleatoric, and epistemic contributions. Step S5 decomposes the overall error among these sources, sometimes requiring special experiments in which all but one error source is excluded. Step S6 critically reviews whether each error component is distributed as expected. Step S7 examines robustness to biases that could emerge in routine use, including applicability, the guarantee that real-world inputs never stray beyond the validity limits explored during assessment. Transparency throughout, particularly about which error sources are considered and how they are separated, is flagged as essential.

The paper demonstrates the framework on seven use cases drawn from the author’s research programme, three of which are summarised in detail. The first concerns strain gauge measurements of bone tissue deformation, used to validate finite element models that predict fracture. The context of use fixes an error threshold derived from the strain difference used to determine fracture, attenuated by two orders of magnitude to account for the chain of inference. True values come from beam-theory calculations on machined aluminium alloy specimens corrected for curvature error; trueness is computed as a root-mean-square error; normality tests confirm the expected distribution of random and systematic errors; and repeated tests on bone specimens establish applicability.

The second case applies the framework to the BBCT-Hip predictor, a biophysical model that estimates mechanical strains in a patient’s bone from a calibrated computed tomography scan. Here the error threshold is set at two percent of the cortical bone failure strain in compression, about 146 microstrain. True values come from strain gauge measurements on carefully preserved cadaveric femurs. The model’s predictions carry numerical, aleatoric, and epistemic errors, and the VVUQ procedure separates them, with the numerical component required to be negligible, the aleatoric component normally distributed with a near-zero mean, and the epistemic component showing a root-mean-square error close to zero. Applicability is probed by exhaustive experiments spanning inter-subject variability and all relevant loading conditions.

The third case is the most forward-looking: assessing a synthetic cohort inferred from a real clinical cohort of elderly patients at risk of hip fracture. The goal is to run in silico trials on virtual populations far larger than any experimentally collected cohort could be, comparing central properties such as means and medians of feature distributions and model predictions. The error thresholds are tied to the measurement and prediction accuracy of each quantity; epistemic error vanishes because the synthetic data are generated by inference, leaving aleatoric error from measurement uncertainty and numerical error from the interpolation functions, which must be shown negligible by sensitivity analysis. Applicability restricts use of the synthetic cohort to the portion of the information space actually sampled by the clinical data.

Viceconi is careful about scope. For estimators that fall squarely into the classical categories, he recommends continuing to use metrology, statistics, or VVUQ, which are more mature and widely accepted within their communities. The 7S Framework is positioned as a supplement for the growing class of hybrid estimators that do not fit anywhere: in silico-augmented trials that inject model predictions into Bayesian device trials, physics-informed neural networks that encode biomechanical law inside learned predictors, synthetic datasets generated to sidestep privacy constraints, and machine learning surrogates trained to replace expensive biophysical simulations. The framework was also applied to cases covering fracture-risk prediction and machine learning surrogates of computational models, and the author reports that it proved effective, sufficiently general, and sensitive to the subtle differences in what credibility means for each information type. Limitations acknowledged include the restriction of the exposition to single scalar quantities, although extension to vectors and time-dependent quantities requires only adding a norm, and the framework’s status as a generalisation rather than a replacement of existing practice. As computational medicine pushes further into regulatory territory, the stakes of getting credibility assessment right rise accordingly, and a shared epistemological vocabulary spanning measurement, inference, and prediction may prove to be exactly what regulators, developers, and clinicians need.

Subject of Research: A general seven-step framework for assessing the credibility of measured, inferred, and predicted quantitative information in computational medicine

Article Title: Assessing the Credibility of Quantitative Information: A General Framework

Article References: Viceconti, M. (2026). Assessing the Credibility of Quantitative Information: A General Framework. Annals of Biomedical Engineering. https://doi.org/10.1007/s10439-026-04367-4

Image Credits: AI Generated

DOI: 10.1007/s10439-026-04367-4

Keywords: credibility assessment, quantitative information, metrology, statistical inference, verification validation and uncertainty quantification, in silico medicine, machine learning predictors, synthetic data, 7S Framework, biomedical engineering, clinical decision-making, predictive models

The concept of a closed-loop therapeutic system is not new. People with type 1 diabetes have lived with versions of it for decades: continuous glucose monitors feed real-time data into an insulin pump, and a control algorithm decides how much hormone to deliver without any manual input. These artificial pancreas systems, first envisioned as servomechanisms in the early 1960s and now cleared by regulators worldwide, demonstrate that autonomous drug delivery is technically and clinically feasible. CLOOR systems aim to apply the same three-component architecture — sensing, actuation and control — to a very different and far more time-critical emergency. Instead of stabilizing a metabolic value over hours, the device must detect a lethal respiratory crisis and counteract it within a narrow window measured in minutes.

The sensing problem is the first and arguably hardest piece. Opioids kill by suppressing respiration, so the most direct overdose signature is a slowing or stopping of breathing. Fentanyl and its synthetic cousins act rapidly on mu-opioid receptors in the brainstem, including the Kölliker–Fuse and parabrachial complexes, sometimes driving breathing to a halt before a person even loses consciousness. Candidate detection modalities therefore include respiratory rate, blood oxygen saturation, chest wall movement, heart rate and cerebral oxygenation. Smartphone-based systems have already demonstrated that acoustic signals and radar-like sonar can capture the apnea characteristic of overdose, and consumer smartwatches have recently received regulatory attention for automated loss-of-pulse detection. Yet each sensing route carries technical liabilities: wearable optical sensors degrade with poor skin contact, pulse oximetry accuracy varies with skin pigmentation, and motion artifacts plague real-world wear. No single validated biomarker currently defines the moment an overdose becomes lethal, which is a stark contrast to the well-characterized glucose thresholds that anchor insulin closed-loop systems.

Once an overdose is detected, the device must act. Prototypes have explored a remarkable range of actuation strategies. Wearable injectors can fire a preloaded dose of naloxone intramuscularly on command, while implantable devices have been designed to sit quietly beneath the skin and respond autonomously to hypoxia. One autonomous implant described in Science Advances integrates sensing and drug delivery in a single unit intended to prevent death from overdose in high-risk individuals. Other designs favor minimally invasive microneedle arrays or patches that combine accelerometer-based respiration tracking with a stored antidote reservoir. The pharmacology matters as much as the hardware: naloxone’s short half-life means that fentanyl can outlast the antidote and reassert its respiratory suppression, a phenomenon that has pushed clinicians toward higher and repeated doses in the synthetic opioid era. Longer-acting antagonists such as nalmefene offer an alternative, though their adoption in community settings remains debated because prolonged reversal can also trigger withdrawal and complicate patient behavior.

Between sensor and actuator sits the control algorithm, the component the researchers identify as the least mature. The algorithm must fuse noisy physiological streams, distinguish a true overdose from sleep, exercise, sedation or sensor failure, and decide when the benefit of automatic naloxone delivery outweighs the risk of a false alarm. Machine learning approaches, including deep-learning respiratory rate detection and personalized Gaussian-process models of individual baselines, offer a path to robust decision-making under uncertainty. Federated learning could allow algorithms to improve across large user populations without centralizing sensitive health data. Still, the authors emphasize that unlike sepsis management or diabetes control, there is no consensus intervention threshold for overdose, no clinically validated definition of the physiological point of no return, and limited clinical evidence from which to train and validate decision systems. Whole-body physiology models that simulate fentanyl-induced respiratory depression and naloxone reversal are helping fill that gap computationally, but translational models cannot fully substitute for human evidence.

The regulatory landscape for CLOOR devices is as complicated as their engineering. Autonomous emergency intervention raises questions that existing frameworks were not designed to answer. The United States Food and Drug Administration has issued guidance on physiological closed-loop control technology and has convened joint public workshops with NIDA to define what evidence would justify approving a device that acts without a patient’s conscious participation. Precedents from automated insulin delivery, including the first regulatory clearance of an open-source automated insulin dosing algorithm, suggest a route is possible, but overdose reversal devices face a different evidentiary burden: their target event is rare, unpredictable and ethically impossible to reproduce in a controlled trial. De novo classifications and 510(k) clearances for related sensors and pulse-detection features hint at how regulators may decompose the problem, yet none of these pathways has yet produced an approved autonomous overdose-reversing device.

Even a technically flawless, regulator-approved device would fail if the people who need it do not wear it. Stigma, distrust and the realities of daily life shape adoption as powerfully as any engineering specification. Studies of people who use opioids in Philadelphia and elsewhere reveal meaningful willingness to use devices capable of detecting and reversing overdose, but also persistent concerns about privacy, involuntary data sharing, battery life, comfort and whether the device might summon police instead of medical help. Community-engaged design efforts, in which people with lived experience co-develop wearable biosensors, have emerged as a model for building the trust that purely technology-driven projects lack. The authors also point to a quieter problem shared by all wearables: attrition. Users abandon tracking devices at high rates, and a device worn only intermittently protects no one during an overdose.

Equity and distribution complete the translational puzzle. The populations at highest risk of fatal overdose — people who inject drugs, people experiencing homelessness, people in rural areas far from emergency services — are precisely those least likely to access expensive novel medical technology. A CLOOR system that costs more than the communities can absorb or that requires maintenance infrastructure unavailable outside clinical settings would widen the survival gap it was built to close. The researchers argue for affordability and distribution strategies engineered from the outset, drawing on lessons from mobile medical systems designed for equitable health care, and for modular architectures that could lower manufacturing costs and speed iterative improvement.

What emerges from the analysis is neither a dismissal of CLOOR technology nor a promise of imminent arrival, but a roadmap. The authors synthesize engineering, clinical, regulatory and public health perspectives into a sequence of priorities: validate overdose biomarkers through controlled human and computational studies; build and benchmark multi-sensor data fusion and control algorithms against realistic physiological variability; establish regulatory pathways that can evaluate autonomous emergency intervention responsibly; design devices around user needs identified through genuine community partnership; and construct distribution models that deliver the technology to those with the most to lose. The fentanyl era has compressed the time available for human rescue to almost nothing. Whether machines can be trusted to take those minutes back, reliably and equitably, is now one of the most consequential questions in bioengineering — and the answer, the researchers conclude, will depend as much on regulation, trust and access as on sensors and algorithms.

Subject of Research: Closed-loop opioid overdose reversal systems that autonomously detect respiratory depression and deliver naloxone without bystander intervention.

Article Title: Opioid overdose detection and reversal with closed-loop systems

Article References: Angelone, L. M., & Koustova, E. (2026). Opioid overdose detection and reversal with closed-loop systems. Nature Reviews Bioengineering. https://doi.org/10.1038/s44222-026-00492-w

Image Credits: AI Generated

DOI: 10.1038/s44222-026-00492-w

Keywords: opioid overdose, naloxone, closed-loop systems, wearable sensors, respiratory depression, drug delivery, FDA regulation, biosensors, fentanyl, public health, translational research, biomedical engineering

Researchers led by Mahbuba Ferdowsi, Ban-Hoe Kwan, Maw Pin Tan and Choon-Hian Goh, working across Universiti Tunku Abdul Rahman in Malaysia, the University of New South Wales in Australia, Universiti Malaya and Sunway University, have built a machine learning pipeline that trains on sensitive physiological data using differential privacy, produces explanations clinicians can inspect, and even runs its predictions on encrypted data. The work, published in BioMedical Engineering OnLine, analysed records from 137 participants who underwent head-up tilt testing, of whom 54 were classified as vasovagal syncope positive and 83 as negative. From electrocardiogram and blood pressure signals recorded during the tilt protocol, the team extracted 54 features capturing the haemodynamic and autonomic signatures of each patient’s response.

The privacy guarantee at the heart of the training process comes from differentially private stochastic gradient descent, or DP-SGD. In conventional training, a neural model’s gradients are computed from each patient example and used to update the model weights, which creates a risk that memorised details of individual patients could be extracted from the finished model. DP-SGD interrupts this leakage path in two ways. First, it clips the gradient computed from every individual example so that no single patient can exert an outsized influence on the weights. Second, it injects calibrated Gaussian noise into the clipped gradients before they are aggregated, obscuring the contribution of any one record. The strength of this protection is formally quantified by a privacy budget, expressed as epsilon and delta. In this study, the model was trained under a privacy budget of approximately epsilon equal to 16.5 with delta set to 10 to the power of minus 5, a setting that provides a mathematically provable bound on how much any single participant’s information can leak into the model or its outputs.

Crucially, this formal privacy came at a surprisingly modest cost in accuracy. The DP-SGD trained logistic regression model, combined with a feature importance selection step that retained the most informative variables, achieved a cross-validated accuracy of 0.833 plus or minus 0.035 and an area under the receiver operating characteristic curve of 0.888 plus or minus 0.019. The evaluation used a stratified five-fold cross-validation framework in which all pre-processing steps were performed strictly within each training fold, a design choice that prevents subtle data leakage from inflating performance estimates. For comparison, the researchers also trained baseline models including a standard logistic regression, random forest classifiers and artificial neural networks, providing a benchmark for how much predictive power was preserved once the privacy machinery was switched on.

The second, more ambitious layer of the pipeline attempted to hide patient data even at the moment of prediction, using fully homomorphic encryption. This cryptographic technique allows computation directly on encrypted values, meaning a hospital could send an encrypted feature vector to an untrusted server and receive an encrypted diagnosis without the server ever seeing the underlying physiological measurements. The team implemented the Cheon-Kim-Kim-Song scheme, often abbreviated CKKS, using the TenSEAL library, which is designed for arithmetic on approximate real numbers and is well suited to encrypting continuous biomedical features. Because homomorphic encryption supports only addition and multiplication, non-linear activation functions must be approximated by polynomials, and the researchers used a degree-3 polynomial approximation of the sigmoid function to perform encrypted classification.

Here the study delivered an honest and instructive negative result. While the differentially private model performed well, inference carried out entirely under encryption degraded sharply, with the area under the curve falling to roughly 0.53, barely better than a coin flip. The culprit was approximation error: the polynomial surrogate for the sigmoid, combined with the noise inherent in CKKS arithmetic, distorted the decision boundary enough to destroy much of the model’s discriminative power. The authors are careful to frame this as a feasibility assessment rather than a failure, noting that the encrypted pipeline was applied only at the prediction stage and that optimised homomorphic architectures are a clear direction for future work. The finding matters for the field because it quantifies, in a real clinical setting, the gap between privacy-preserving training and fully secure inference, showing that the former is nearly free in utility terms while the latter remains a genuine engineering challenge.

Interpretability, the third pillar of the framework, was assessed with two widely used explanation techniques, LIME and SHAP, which attribute a model’s predictions back to individual input features. Both methods converged on the same picture of what the model had learned. Tilt-phase haemodynamic measurements, specifically systolic and diastolic blood pressure recorded during the tilt (SBP_T and DBP_T), together with autonomic balance indices derived from heart rate variability and blood pressure variability in the low-frequency to high-frequency bands (LFHF_RRI_T and LFHF_SBP_T), emerged as the key predictors of a positive vasovagal syncope diagnosis. This alignment is clinically reassuring, because these features correspond to the physiological mechanisms physicians already associate with tilt-induced fainting: an exaggerated drop in vascular tone and a shift in autonomic control when the body is tilted upright.

The study’s authors are explicit about the limits of what their results demonstrate. All reported performance figures are internal, cross-validated estimates derived from a single-centre dataset of 137 participants, and the team cautions that they should not be read as evidence of generalizable clinical performance. The relatively modest sample size also meant that subgroup analyses were omitted, a decision consistent with the data minimisation principle that underpins the privacy framework itself. External validation on multi-centre cohorts, and independent confirmation that the identified haemodynamic and autonomic features hold across different populations and tilt-test protocols, remain essential next steps before any deployment in a clinical environment.

Even so, the work arrives at a moment when healthcare systems worldwide are wrestling with how to exploit the predictive power of machine learning without breaching patient trust or privacy law. Differential privacy, homomorphic encryption and explainable AI are usually studied in isolation; this study is notable for assembling all three around a concrete diagnostic problem and reporting exactly where each succeeds and where each breaks down. The demonstration that a formally private model can match conventional baselines with minimal utility loss offers a template for other biomedical prediction tasks, from arrhythmia detection to sepsis early warning, where training data are scarce, sensitive and ethically charged.

The research was funded by the UTAR Research Fund from Universiti Tunku Abdul Rahman and received ethics approval from both the UTAR Scientific and Ethical Review Committee and the UMMC Medical Research Ethics Committee, with written informed consent obtained from all participants. As hospitals move toward federated learning and cloud-based clinical AI, studies of this kind provide the empirical grounding needed to decide which privacy technologies are ready for the clinic today and which still require the next generation of cryptographic engineering. For patients whose fainting spells have defied easy diagnosis, the promise is an AI assistant that reads their tilt test accurately, explains its reasoning, and never exposes their data in the process.

Subject of Research: Privacy-preserving and explainable machine learning for vasovagal syncope detection from head-up tilt test signals

Article Title: Differentially private and explainable machine learning for vasovagal syncope detection: a feasibility study of homomorphic encryption for secure inference

Article References: Ferdowsi, M., Kwan, B.-H., Tan, M. P., & Goh, C.-H. (2026). Differentially private and explainable machine learning for vasovagal syncope detection: a feasibility study of homomorphic encryption for secure inference. BioMedical Engineering OnLine. https://doi.org/10.1186/s12938-026-01626-2

Image Credits: AI Generated

DOI: 10.1186/s12938-026-01626-2

Keywords: vasovagal syncope, head-up tilt test, differential privacy, DP-SGD, homomorphic encryption, CKKS scheme, explainable AI, LIME, SHAP, secure inference, biomedical engineering, electrocardiogram

The appeal of the technique lies in its simplicity. A cold plasma microjet is essentially a pencil-thin plume of ionized helium gas that remains close to room temperature even while it carries a menagerie of reactive species: electrons, ions, radicals, and energetic ultraviolet photons. When the researchers aimed this plume at magnesium wire submerged in liquid, the plasma’s chemistry attacked the metal surface, driving oxidation and dissolution reactions that ultimately precipitated magnesium hydroxide directly in the solution. Because no reducing chemicals are added at any stage, the process sidesteps many of the environmental and purification headaches associated with conventional wet-chemical nanoparticle synthesis, in which reagents such as sodium borohydride or organic solvents must later be removed from the final product.

To confirm that they had actually built what they intended to build, the team subjected their nanoparticles to a battery of characterization techniques. X-ray diffraction, which probes the arrangement of atoms by measuring how X-rays scatter from crystal planes, produced patterns that matched crystalline magnesium hydroxide exactly, and analysis of the peak widths revealed an average crystal size of approximately 10.5 nanometers. That is astonishingly small. For scale, a single human hair is roughly 7,000 times wider than one of these particles, and at that dimension the surface-to-volume ratio becomes enormous, meaning a large fraction of every particle’s atoms sit at the surface where they can interact directly with biological targets.

Scanning electron microscopy coupled with energy-dispersive X-ray spectroscopy provided the second pillar of evidence. The electron micrographs showed particles with a near-spherical morphology and a tight, uniform distribution, while the accompanying elemental analysis confirmed high chemical purity with no detectable contaminant phases introduced by the plasma process. Perhaps most importantly for any eventual biomedical use, zeta potential measurements returned values between +30 and +40 millivolts. Zeta potential is a measure of the electrical charge a particle presents to its surroundings, and values of this magnitude indicate strong electrostatic repulsion between neighboring particles. In practical terms, the nanoparticles resist clumping together and settling out of suspension, which means a stable, well-dispersed formulation can be prepared and stored without specialized stabilizers.

With the material thoroughly characterized, the researchers turned to the question that drives most of nanomedicine: what does it do to cancer cells? They exposed HeLa cells, a famous and hardy line of cervical cancer cells first isolated in 1951, to nanoparticle concentrations ranging from 0.025 to 1.0 parts per million, then measured cell viability using the MTT assay at 24, 48, and 72 hours. The MTT assay relies on the fact that metabolically active cells reduce a yellow tetrazolium compound into purple formazan crystals; the amount of purple dye produced is directly proportional to the number of living, functioning cells. It is one of the most widely trusted readouts of cytotoxicity in cell biology, and the results here followed a clear and reproducible pattern.

Cell viability dropped steadily as both concentration and exposure time increased, a dose- and time-dependent relationship that is exactly what one expects from a genuine cytotoxic agent rather than a measurement artifact. At the highest concentration after 72 hours, cytotoxicity exceeded 80 percent, meaning that more than four out of every five cancer cells in the treated cultures had lost viability. Under the microscope, the researchers documented the physical consequences of that toxicity: treated cells visibly shrank and their membranes showed signs of damage, morphological hallmarks consistent with progressive cellular collapse rather than a transient growth slowdown.

The mechanism behind magnesium hydroxide’s anticancer effect is not fully mapped, and the study’s authors are careful on this point, emphasizing that the molecular pathways involved still need to be elucidated in future work. The existing literature, however, offers several plausible leads. Magnesium hydroxide is mildly alkaline and slowly releases hydroxide ions, which can perturb the delicate pH homeostasis that tumor cells work hard to maintain. Prior studies have shown that magnesium hydroxide nanoparticles can physically damage bacterial cell walls, raising the possibility that similar mechanical stress on the more fragile membranes of cancer cells contributes to the observed membrane damage. Related magnesium oxide nanoparticles have been linked to elevated reactive oxygen species, oxidative stress being one of the classic triggers of programmed cell death. Disentangling these candidate mechanisms, and determining which dominates at the low parts-per-million concentrations used in this study, is the obvious next experimental frontier.

The researchers are equally candid about the most important caveat: selectivity. A compound that kills cancer cells but harms healthy cells equally well is simply a poison, not a therapy. This study evaluated cytotoxicity exclusively against HeLa cells, so the question of how the nanoparticles behave toward normal cervical epithelial cells, fibroblasts, or other healthy tissue types remains open. Earlier work on magnesium-based nanomaterials has often reported favorable biocompatibility profiles, and magnesium itself is a biologically essential element that the human body regulates and tolerates well, which gives the field reason for optimism. But optimism is not evidence, and the authors explicitly call for comparative studies against normal cell lines before any clinical relevance can be claimed.

If those selectivity studies succeed, the potential applications extend beyond anticancer therapy. Magnesium hydroxide nanoparticles have already attracted interest as antibacterial agents, wound-dressing components, and drug-delivery platforms, with prior reports describing magnesium hydroxide nanocomposite hydrogels for infected wound care and fluorescent magnesium hydroxide nanosheets for antimicrobial bandages that also monitor wound pH. A synthesis method that is cheap, green, and reproducible, as the cold plasma microjet route demonstrably is, lowers the barrier to producing the quantities of well-defined nanomaterial that such applications demand. The plasma approach also produces the particles in a single step from a metallic precursor, avoiding the surfactants, high temperatures, and prolonged hydrothermal processing that other routes require.

What makes the study genuinely notable, then, is the combination of three results that rarely appear together: an unusually clean and sustainable synthesis, a rigorously characterized and colloidally stable product, and a striking biological effect at remarkably low concentrations. Parts per million is an extraordinary range in which to see strong cytotoxicity, and if future work confirms that the effect is selective for transformed cells, cold-plasma-synthesized magnesium hydroxide nanoparticles could move from the physics lab toward preclinical evaluation. For now, the study stands as a persuasive demonstration that one of the most extreme tools in modern physics can craft one of chemistry’s humblest compounds into a serious candidate for cancer research, and it offers a template for how plasma science and biomedicine can be married to greener effect.

The choice of helium as the plasma-forming gas is itself worth noting. Helium’s high ionization energy and low breakdown voltage make it the workhorse gas for atmospheric-pressure plasma jets, producing a discharge that is dense in reactive oxygen and nitrogen species yet gentle enough to operate in open air near liquid surfaces. Researchers in the same laboratory group have previously applied this technique to other materials, including zinc oxide, copper oxide, iron oxide, and selenium nanoparticles, suggesting that the microjet platform functions as a general-purpose synthesis tool rather than a one-off method tailored to a single compound.

The positive surface charge reported for the particles also has biological implications beyond shelf stability. Positively charged nanoparticles generally interact more readily with the negatively charged membranes of mammalian cells, which can promote cellular uptake through endocytosis. Prior work on related magnesium-based nanomaterials has traced toxicity to caveolin-1-mediated endocytosis in endothelial cells, illustrating how surface charge and internalization pathways can shape a nanoparticle’s biological behavior.

The concentration range examined here deserves emphasis. At 0.025 to 1.0 parts per million, the effective doses are far below those typically reported for many metal oxide nanoparticles in similar assays, where tens or hundreds of parts per million are often required to achieve comparable killing. Whether this heightened potency reflects the small crystal size, the high purity, the colloidal stability, or some combination of these properties remains an open question that comparative studies against conventionally synthesized magnesium hydroxide could resolve.

It is also useful to place the safety picture in context. Animal studies of magnesium oxide nanoparticles have generally found limited acute toxicity, and investigations of magnesium hydroxide nanoparticles in normal biological systems have reported favorable biosafety profiles at antibacterial doses. That record, combined with magnesium’s status as an essential physiological element, provides a plausible foundation for the selectivity studies the authors now call for, though in vitro potency at parts-per-million levels will need careful re-examination in more complex biological systems before therapeutic claims can be entertained.

Subject of Research: Cold plasma synthesis of magnesium hydroxide nanoparticles and their in vitro cytotoxicity against HeLa cervical cancer cells.

Article Title: Cold Plasma Synthesis Mg(OH)₂ Nanoparticles: In Vitro Cytotoxic Evaluation Against HeLa Cervical Cancer Cells

Article References: saad akram, R., Majeed, N. F., Abdalameer, N. K., & Zaydan, E. A. (2026). Cold Plasma Synthesis Mg(OH)₂ Nanoparticles: In Vitro Cytotoxic Evaluation Against HeLa Cervical Cancer Cells. Journal of Medical and Biological Engineering. https://doi.org/10.1007/s40846-026-01055-5

Image Credits: AI Generated

DOI: 10.1007/s40846-026-01055-5

Keywords: cold plasma, magnesium hydroxide, nanoparticles, HeLa cells, cervical cancer, cytotoxicity, MTT assay, green synthesis, nanomedicine, biomedical engineering, X-ray diffraction, zeta potential

The Segment Anything Model, or SAM, was introduced by Meta AI researchers in 2023 as a foundation model for image segmentation, trained on more than a billion masks across eleven million images. Its promise was audacious: the ability to segment, or precisely outline, virtually any object in any image without task-specific training. In medical imaging, remote sensing, pathology, and even planetary geology, researchers have rushed to harness SAM’s general-purpose vision capabilities. But SAM has a well-documented Achilles heel. When confronted with images dominated by dense, repetitive, fine-grained textures—the branching vessels of a retina, the tangled architecture of a surgical field, or in this case the grooves, channels, and tread patterns of a shoe sole—the model’s performance drops sharply. Footwear outsoles are exactly this kind of challenge. Their tread designs combine geometric regularity with manufacturing variation, wear patterns, and material contrasts that confound a model trained mostly on natural scenes and everyday objects. Shaghayegh Chavoshian, Ali Barzegar Khanghah, and Atena Roshan Fekr, based at the KITE Research Institute of the Toronto Rehabilitation Institute and the Institute of Biomedical Engineering at the University of Toronto, set out to close that gap.

Their strategy was transfer learning, a technique in which a model pretrained on a vast general dataset is retrained, or fine-tuned, on a smaller but highly specific dataset so that its general visual knowledge is redirected toward a narrow task. Fine-tuning foundation models has already proven fruitful in medical image segmentation, where adapted SAM variants have been used for anatomical structures, tumor delineation, and surgical video analysis. The Toronto team reasoned that the same principle should apply to footwear science, where manual annotation of outsole images is notoriously slow and expensive. Segmenting an outsole by hand requires a trained expert to trace, pixel by pixel or polygon by polygon, the boundaries between tread features and the spaces between them—a process that consumes substantial time per image and becomes prohibitive when applied across the many shoes a serious slip-resistance study requires. Automating this bottleneck would open the door to analyzing footwear at a scale previously impractical, potentially transforming how safety standards are written and how shoes are evaluated for consumers, workers, and older adults. The motivation is reinforced by a long line of prior work showing that outsole features such as tread groove geometry, sole hardness, and material wear all influence the friction available at the shoe-floor interface, which makes accurate, scalable measurement of outsole geometry a genuinely valuable scientific target rather than a mere convenience.

To fine-tune the model, the researchers assembled a dataset of forty footwear outsoles, each manually annotated using a graphical annotation tool that allows precise polygonal outlining of image features. Forty shoes may sound modest next to SAM’s eleven-million-image pretraining corpus, but that is precisely the point of transfer learning: the general model already understands edges, shapes, and textures; it needs only a comparatively small volume of domain-specific examples to learn what matters in a shoe sole. The ground truth against which the model was judged came not from mechanical friction devices alone but from human-centered data, in which footwear had been classified as having either low or high slip resistance based on real human testing. Some of that human-centered footwear data was drawn from open access material available through the Rate My Treads website, a resource that aggregates winter footwear performance information. This choice of labels matters. Mechanical slip testers measure friction under controlled conditions, but human slip resistance emerges from an interplay of gait biomechanics, loading rates, and perception that benchtop devices only approximate. By anchoring the labels to how shoes actually perform on people’s feet, the researchers kept the machine learning pipeline aligned with the outcome that ultimately counts—whether a person slips.

The results of the fine-tuning were substantial. Compared with the original, out-of-the-box SAM, the adapted model reduced segmentation loss by 8.11 percent and lifted the intersection over union, the standard overlap metric between predicted and true segmentation masks, to 70.45 percent. Perhaps more strikingly, pixel accuracy climbed from 56.90 to 78.10 percent, and the F1 score, which balances precision and recall, rose from 53.70 to 66.30 percent. Those numbers tell a clear technical story: without task-specific adaptation, SAM could barely delineate outsole features better than chance on some measures, but after fine-tuning it captured roughly two-thirds to three-quarters of the relevant structure. The researchers also examined which image quality factors influenced performance, finding that resolution, contrast, and intensity all significantly affected segmentation quality. This is consistent with a growing literature showing that deep segmentation networks are sensitive to the spectral and spatial characteristics of their inputs; a model fine-tuned on images of one resolution or contrast profile may degrade when fed imagery that differs. For anyone hoping to deploy such systems in the field—photographing shoes in a store, a clinic, or a workplace—the finding underscores that image capture protocols must be standardized for reliable results.

The segmentation outputs were then put to work. From each predicted outsole mask, the pipeline estimated the outsole-ground contact areas—the regions of the sole that would actually press against a floor during walking. These contact features were fed into a downstream machine learning classifier tasked with predicting the slip resistance category of the shoe, low or high, using the human-derived ground truth labels. With an 80/20 train-test split, the classification model achieved 70 percent accuracy and a 64 percent F1 score. While these figures are a distance from clinical certainty, they represent a meaningful proof of concept: the geometry of a shoe’s contact patch, extracted automatically by an adapted foundation model, carries enough information to forecast, with better-than-chance reliability, how the shoe will behave under a human foot on a slippery surface. Earlier work by the same group and collaborators had shown that convolutional neural networks and machine learning models could predict slip resistance from engineered tread features, including studies of winter footwear on glycerol-contaminated surfaces; the new study shortens that pipeline by replacing manual feature extraction with learned segmentation.

The broader context makes the advance more than an academic exercise. Winter footwear rated for slip resistance is still evaluated largely through mechanical tests and, increasingly, through human-centered trials such as those conducted with the maximum achievable incline method, in which participants walk up progressively steeper icy slopes until they slip. Portable slip simulators, cart-type friction measurement devices, and computational models of shoe-floor friction have each expanded the toolkit, but these approaches remain expensive, time-consuming, and difficult to scale to the flood of new footwear models reaching the market each year. Rating programs that rely on such testing simply cannot keep pace. If a vision model can screen outsole designs computationally, flagging promising candidates for human testing and discouraging poor performers before they reach consumers, the entire evaluation ecosystem could accelerate. The authors point to implications for two populations in particular: older adults, for whom a fall can trigger a cascade of fractures, hospitalization, loss of independence, and mortality, and workers in occupations—from construction and healthcare to marine and winter industries—who depend on protective footwear to stay safe in hazardous environments.

The study also contributes to a rapidly evolving conversation about when and how to adapt foundation models. Across computer vision research, SAM has been shown to struggle in concealed scenes, camouflaged object detection, and numerous specialized domains, prompting a wave of adaptation techniques ranging from lightweight adapter layers and parameter-space reconstruction to knowledge distillation, in which a compact student model learns to mimic a large teacher. The Toronto study adds footwear science to the list of domains where modest fine-tuning unlocks outsized gains, and its finding that image quality variables materially shape performance echoes parallel observations in brain tissue segmentation and satellite imagery classification. The researchers acknowledge the inherent constraints of their dataset size—forty annotated outsoles leaves the classifier with limited statistical power—and the 70 percent classification accuracy should be read as an early benchmark rather than a deployable performance ceiling. Scaling the annotated corpus, standardizing image acquisition, and refining the contact-area features that bridge segmentation and prediction are the obvious next steps.

Funded by the Digital Research Alliance of Canada and reviewed by the University Health Network Research Ethics Board, the work forms part of a broader research program at KITE that spans mechanical and human-centered slip testing, gait analysis with multimodal transformers, and slip detection during real human walking trials, alongside earlier investigations of how the edge of a footwear sole influences measured slip resistance. Together these strands sketch a future in which the slip resistance of a shoe could be assessed rapidly, cheaply, and at scale—by photographing its sole, letting a fine-tuned foundation model trace its contact geometry, and letting a classifier render a verdict before the shoe ever meets an icy sidewalk. For the millions of people who navigate winter sidewalks, wet kitchens, oily factory floors, and slick hospital corridors every day, that future could translate into fewer falls, fewer fractures, and a far more transparent footwear market in which slip resistance is not a marketing slogan but a measurable, machine-verified property.

Subject of Research: Fine-tuning the Segment Anything Model for automated footwear outsole segmentation to predict slip resistance.

Article Title: Transfer Learning on Segment Anything Model for Footwear Outsole Segmentation to Predict Footwear Slip Resistance

Article References: Chavoshian, S., Khanghah, A. B., & Fekr, A. R. (2026). Transfer Learning on Segment Anything Model for Footwear Outsole Segmentation to Predict Footwear Slip Resistance. Annals of Biomedical Engineering. https://doi.org/10.1007/s10439-026-04331-2

Image Credits: AI Generated

DOI: 10.1007/s10439-026-04331-2

Keywords: Segment Anything Model, transfer learning, footwear outsole segmentation, slip resistance, machine learning, computer vision, fall prevention, biomedical engineering, slip and fall injuries, image segmentation, footwear safety, deep learning

The backdrop of this research is set against the crucial need for effective imaging modalities that can provide real-time insights into dynamic biological systems. Traditional imaging techniques, while valuable, often lack the capability to provide the detailed, high-quality data required in various clinical scenarios. By addressing these limitations, the authors of this study advocate for a shift towards more advanced computational techniques, particularly those harnessing the capabilities of neural networks.

At the core of their investigation is the utilization of “neural implicit models,” a concept that leverages deep learning to represent complex shapes and intensity patterns. By employing these models, researchers can create highly detailed representations of anatomical structures, allowing for a more nuanced understanding of their spatial relationships and changes over time. This advancement is particularly important in dynamic scenarios where the target anatomy is not static and may undergo significant transformations during the imaging process.

Dynamic stereo-radiography, the focus of this study, is a relatively novel technique that combines stereo imaging with radiographic methods to capture moving biological processes. While this technique provides substantial benefits, it also introduces challenges related to the accurate registration of 2D and 3D data. The authors propose that by integrating neural implicit models, these challenges can be effectively mitigated. This promising integration could facilitate more precise alignments between two-dimensional images and their corresponding three-dimensional representations, ultimately leading to improved outcomes in various medical applications.

One of the primary advantages highlighted in the study is the ability of neural implicit models to learn and adapt from vast amounts of imaging data. Unlike conventional models, which may rely heavily on predefined geometric parameters, these neural networks can extract complex features directly from data, allowing them to adapt dynamically to varying shapes and intensities encountered in different scenarios. This adaptability is particularly crucial in medical imaging, where variability among patients and pathological conditions can be significant.

Moreover, the study emphasizes the potential for scan-free applications of these neural models. Traditional imaging methods often require extensive scans that can be time-consuming and expose patients to unnecessary radiation. By developing techniques that can infer shape and intensity information without the need for extensive scanning, the researchers open up the possibility of safer, more efficient imaging protocols. This approach not only prioritizes patient safety but also addresses the practical limitations often faced in clinical settings.

Implementing these neural implicit models in dynamic stereo-radiography could lead to breakthroughs in the diagnosis and monitoring of various conditions. For instance, in the realm of orthopedic surgery, accurate 2D-3D registration can significantly enhance pre-operative planning, allowing surgeons to visualize complex anatomical structures with an unprecedented level of detail. This visual clarity can diminish the likelihood of intraoperative complications and improve patient outcomes.

The implications extend beyond surgical practice; they also resonate within the fields of cardiology, nephrology, and oncology, where dynamic imaging plays a vital role in assessing disease progression and treatment efficacy. By enabling a robust connection between 2D and 3D representations, the research stands to transform how medical professionals interpret imaging data and make clinical decisions.

Moreover, the collaborative effort of the research team underscored the interdisciplinary nature of advancing biomedical technologies. By merging expertise from neural network design and medical imaging techniques, the authors provide a comprehensive understanding of how computational advancements can directly impact clinical practices. This orchestration of knowledge highlights the need for collaborative frameworks in research initiatives, combining insights from engineering, medicine, and data science.

In conclusion, the research findings put forth by Burton, Myers, and Rullkoetter signify a transformative approach to imaging in healthcare. The integration of neural implicit shape and intensity models with dynamic stereo-radiography not only addresses existing limitations in traditional imaging but also paves the way for innovative, scan-free methodologies that prioritize patient safety and operational efficiency. This development is poised to usher in a new era of precision medicine, where the interplay between deep learning and medical imaging profoundly enhances the quality of care delivered to patients around the globe.

In an era where medical technology continues to evolve at a rapid pace, studies like these reaffirm the importance of leveraging advanced computational techniques to tackle real-world challenges in healthcare. The journey towards improved imaging modalities is just beginning, and the implications of this research reach far beyond theoretical applications, laying the groundwork for practical solutions that could define the future of medical diagnostics.

As the healthcare landscape shifts towards more integrated and technology-driven approaches, the collaboration between researchers and clinical practitioners will be vital in realizing the full potential of these advancements. The insights gained from such studies not only contribute to scientific literature but translate into actionable benefits for patients, ultimately driving improvements in health outcomes across diverse medical domains.

As we look to the future, the developments in neural implicit modeling and dynamic imaging technologies underscore the importance of interdisciplinary dialogue and collaboration within the scientific community. By fostering partnerships that bridge clinical and technical expertise, we can continue to push the boundaries of what is possible in the realm of medical imaging and beyond.

In this context, the role of ongoing research and innovation remains critical, as it fuels the progress necessary to navigate the complexities of modern healthcare. The findings of this study may be just the starting point for a broader exploration of how artificial intelligence can revolutionize the healthcare sector, and as we move forward, it will be exciting to witness the transformative potential these technologies hold.

By embracing change and remaining committed to the pursuit of knowledge, the intersection of technology and medicine can cultivate an environment ripe for groundbreaking discoveries that ultimately improve patient care and outcomes.

Subject of Research: Neural Implicit Shape and Intensity Models for 2D-3D Registration in Dynamic Stereo-Radiography

Article Title: Neural Implicit Shape and Intensity Models for Scan-Free 2D-3D Registration in Dynamic Stereo-Radiography

Article References:
Burton, W., Myers, C. & Rullkoetter, P. Neural Implicit Shape and Intensity Models for Scan-Free 2D-3D Registration in Dynamic Stereo-Radiography. Ann Biomed Eng (2025). https://doi.org/10.1007/s10439-025-03911-y

Image Credits: AI Generated

DOI: https://doi.org/10.1007/s10439-025-03911-y

Keywords: Neural Implicit Models, Dynamic Stereo-Radiography, 2D-3D Registration, Medical Imaging, Artificial Intelligence, Biomedical Engineering, Precision Medicine, Clinical Applications.

Hillman’s appointment is rooted in her remarkable history as a pioneer in the field of imaging. She holds an impressive track record in developing high-speed microscopes and advanced in-vivo imaging systems for studying living tissues. Under her leadership, a range of talented faculty members is expected to join the department, collectively driving the advancement of imaging techniques that span from microscopic imaging at the sub-cellular scale to comprehensive medical imaging processes. This diverse expertise will ideally enhance scientific studies while simultaneously improving patient care outcomes.

The establishment of this department is a clear indication of St. Jude’s commitment to innovation in biomedical research, especially focused on children. “Elizabeth is a renowned physicist, gifted biomedical engineer, and prolific inventor of new technologies,” remarked James R. Downing, MD, the president and CEO of St. Jude Children’s Research Hospital. The ambitious vision involves not merely building a functional department but rather creating a hub of excellence that integrates cutting-edge imaging technology into multidisciplinary research and clinical applications for children experiencing severe health challenges.

One of the primary objectives of this new department will be to develop and refine imaging and measurement methodologies that can facilitate transformative scientific studies. By leveraging advanced imaging technologies, researchers will likely better grasp disease processes and treatment outcomes, creating pathways for groundbreaking innovations in patient care. Hillman’s deep-rooted beliefs regarding the synergy between environment and innovation underscore the importance of St. Jude’s unique collaborative landscape. She acknowledges that local collaborations and shared scientific inquiries have significantly influenced her creative endeavors throughout her career.

Prior to joining St. Jude, Hillman made remarkable contributions during her tenure at Columbia University, serving as both a Herbert and Florence Irving Professor and a tenured professor in biomedical engineering and radiology. Her extensive 20-year career is marked by the successful development and application of a wide array of novel imaging and data analysis methods. These methods have not only advanced scientific inquiry but have also paved the way for potential commercial applications, evidenced by technologies she developed that have been licensed to major industry players like PerkinElmer and Leica Microsystems.

The broader implications of Hillman’s appointment extend beyond mere technological advancements. J. Paul Taylor, MD, PhD, St. Jude’s executive vice president and scientific director, articulated the revolutionary potential of recent advances in visualization and quantification methodologies. This revolutionary potential is expected to catalyze significant improvements in biomedical research specifically tailored to combating childhood diseases. Therefore, St. Jude’s commitment to propelling the institution forward in biomedical imaging innovation could manifest profound benefits for children diagnosed with life-threatening illnesses.

As Hillman transitions into her new role, she emphasizes the unique combination of talent and passion present at St. Jude, which she considers to be critical in addressing some of the most challenging questions in child health. The hospital’s environment presents a stimulating atmosphere where facilitators of scientific discovery can collaborate toward shared goals, maximizing the impact of their findings in real-time patient care. Hillman asserts that working in an inspiring environment like St. Jude will foster creativity and significantly heighten the immediate impacts of innovative discoveries.

Hillman’s academic pedigree includes a PhD in medical physics and bioengineering from University College London, one of the leading institutions known for driving scientific advancements. Furthermore, her post-doctoral work at the Martinos Center for Biomedical Engineering, affiliated with Massachusetts General Hospital and Harvard Medical School, provided her with foundational expertise in biomedical engineering, focusing on imaging sciences. She has authored over 100 research papers featured in esteemed journals such as Science, Nature Methods, Nature Photonics, and Nature Biomedical Engineering, showcasing her prominent role in advancing the field.

Moreover, Hillman has made substantial contributions to augmenting the scientific community’s understanding of critical biological processes and disease mechanisms. Her work reflects a synthesis of theory and applied sciences, which illustrates the value of interdisciplinary collaboration in driving biomedical progress. As a testament to her innovative contributions, she holds over 20 issued patents and was elected to the National Academy of Inventors in 2022. This remarkable recognition underscores her dedication to fostering an environment rich in innovation and invention, ensuring young researchers also have the opportunities to thrive within this dynamic landscape.

St. Jude Children’s Research Hospital has solidified its position as a preeminent institution in transforming how childhood diseases are understood, treated, and cured. With a unique focus as the only National Cancer Institute-designated Comprehensive Cancer Center exclusively dedicated to children, the hospital has played a critical role in improving pediatric treatment outcomes over its 60-plus-year history. Specifically, the treatment advancements achieved at St. Jude have propelled the childhood cancer survival rate from a mere 20% to 80%, representing a drastic shift and a beacon of hope for countless families across the globe.

Notably, the breakthroughs generated at St. Jude do not remain confined within its walls. The institution is deeply committed to sharing its discoveries, allowing healthcare providers worldwide to enhance treatment quality and care for children suffering from life-threatening conditions. Whether through its digital platforms or social media presence, St. Jude actively engages in disseminating vital knowledge that can have a lasting influence on partners in the healthcare community.

As the new Department of Imaging Sciences embarks on its groundbreaking journey under Hillman’s leadership, it signifies not just a commitment to scientific advancement but also a profound dedication to the lives of the children it serves. By uniting cutting-edge technology with a comprehensive understanding of pediatric diseases, the collaborative efforts within this department could redefine the contours of research excellence at the intersection of imaging and healthcare, ultimately transforming the future landscape of pediatric medicine for generations to come.

Subject of Research: Imaging and Measurement Approaches in Pediatric Medicine
Article Title: Elizabeth Hillman Appointed Founding Chair of St. Jude’s Imaging Sciences Department
News Publication Date: October 2023
Web References: St. Jude Children’s Research Hospital
References: N/A
Image Credits: Credit: St. Jude Children’s Research Hospital

Keywords

Imaging, Biomedical Engineering, Pediatric Medicine, High-Speed Microscopy, In-Vivo Imaging, Technology Innovation, Scientific Research, Child Health Care, Imaging Sciences, Medical Imaging.