The research, published in Holistic Integrative Oncology, was led by a team at Sun Yat-sen University Cancer Center in Guangzhou. The investigators built what they call a combined radiomics model, one that fuses quantitative features extracted from two of the most common imaging tests in breast care: ultrasound and dynamic contrast-enhanced magnetic resonance imaging, or DCE MRI. Their goal was ambitious but precise—to distinguish luminal from non-luminal breast cancers preoperatively, using only images that patients already receive in standard clinical workups.

The clinical logic behind the effort is straightforward. Luminal cancers, which express hormone receptors, typically respond well to endocrine therapy and carry the highest five-year survival rates. Non-luminal tumors—HER2-positive and triple-negative cancers—behave differently. Because they tend to be more sensitive to chemotherapy, guidelines often recommend neoadjuvant treatment given before surgery for these patients. Knowing the subtype in advance therefore shapes the entire treatment sequence. Today, that knowledge comes from a core needle biopsy analyzed with immunohistochemistry, an invasive procedure whose results depend on where in the tumor the sample was taken and how a pathologist interprets the staining.

That sampling problem is not trivial. Studies cited by the team indicate that roughly 5.9 to 18.5 percent of biopsy specimens fail to match the molecular subtype of the final surgical specimen, a discrepancy rooted in tumor heterogeneity—different regions of the same cancer can carry different receptor profiles. A non-invasive imaging-based prediction, the researchers argue, could complement biopsy by assessing the whole lesion rather than a sliver of it, and could serve as a decision-support tool when biopsy is delayed, contraindicated, or declined.

To build the model, the team assembled 655 patients with primary invasive breast cancer from two hospitals. A cohort of 551 patients from Sun Yat-sen University Cancer Center was split in a 3:1 ratio into a training set of 413 cases and an internal test set of 138 cases, while 104 patients from the First Affiliated Hospital of Gannan Medical University formed an independent external validation set—an essential step, since many previous AI models in this field were trained and tested on data from a single center, inflating their apparent accuracy. Importantly, images were acquired on many different scanners from multiple manufacturers, with varied acquisition parameters, testing whether the model could generalize beyond any single device.

The technical pipeline was exhaustive. For each patient, radiologists delineated the tumor on a representative ultrasound image and, slice by slice, on the DCE MRI phase showing the strongest enhancement. Software then extracted thousands of quantitative features—1,561 from each ultrasound image and 1,316 from each MRI scan—capturing first-order statistics, lesion shape, and intricate texture patterns, including features derived from wavelet and other image transformations. Through a multi-stage filtering process involving inter-observer agreement checks, statistical testing, correlation pruning, and LASSO regression with tenfold cross-validation, the combined signature was distilled to just 25 key features: nine first-order, one shape, and fifteen texture descriptors.

Those features were fed into eight different machine learning algorithms, from logistic regression and random forests to gradient boosting methods and support vector machines. Because non-luminal cancers made up only about a quarter or less of each dataset, the team applied SMOTE, a synthetic over-sampling technique that generates new examples of the minority class to prevent the algorithms from simply defaulting to the more common luminal category. The effect was striking in some cases: the support vector machine’s performance on the internal test set jumped from an AUC of 0.731 without balancing to 0.872 with it, although the authors note that this sensitivity to preprocessing should be kept in mind when reproducing the model.

The support vector machine ultimately emerged as the winner. Its combined model achieved an AUC of 0.973 in the training set and 0.872 in the internal test set, significantly outperforming the ultrasound-only model (AUCs of 0.701 and 0.751) and the MRI-only model (0.919 and 0.742). On the external validation cohort from the second hospital, the combined model posted an AUC of 0.835, again numerically ahead of the single-modality models at 0.684 for ultrasound and 0.740 for MRI, though the differences there did not reach statistical significance. Perhaps most clinically meaningful were the model’s safety-oriented metrics: a sensitivity of 0.833 and a negative predictive value of 0.927 in external validation, and even higher values of 0.926 and 0.976 in the internal test set. In plain terms, the model rarely missed a non-luminal cancer—the very patients who need neoadjuvant chemotherapy—while patients the model classified as negative were very likely to have luminal disease suitable for endocrine therapy.

Decision curve analysis reinforced the statistical results, showing that the combined model delivered the highest net benefit across a range of clinical threshold probabilities compared with either single-modality approach. The authors position the tool not as a replacement for pathology but as a complementary layer in the diagnostic workflow: a preoperative risk-stratification aid that could help flag patients likely to benefit from chemotherapy before surgery, or provide guidance in settings where tissue sampling is impractical. Because the model draws on two-dimensional ultrasound and three-dimensional MRI views simultaneously, it captures complementary information—ultrasound excels at lesion margins and echotexture, while contrast-enhanced MRI reveals how a tumor perfuses and enhances, a window into its proliferative activity.

The study is not without caveats, which the authors acknowledge candidly. It was retrospective, meaning image quality and completeness could not be fully controlled, and the numbers of HER2-positive and triple-negative cases were relatively small, reflecting their real-world prevalence. Manual segmentation of MRI lesions, performed layer by layer, introduces subjectivity and labor costs that future deep learning approaches could automate. The team calls for prospective, multi-center collection of more cases to sharpen sensitivity and positive predictive value further. Still, the demonstration that a radiomics model trained on heterogeneous, multi-vendor images from two hospitals can hold its performance on completely unseen external data marks a meaningful step toward bringing AI-assisted molecular subtype prediction out of the research lab and into the breast clinic—where a few minutes of computation on images patients already have could help ensure the right therapy reaches the right tumor at the right time.

Subject of Research: Radiomics-based prediction of breast cancer molecular subtypes using ultrasound and DCE MRI

Article Title: The radiomics models based on ultrasound and DCE MRI for predicting molecular subtypes of breast cancer

Article References: Mao, R., Zheng, H., Tang, X., Yang, L., Zhang, Y., & Zhou, J. (2026). The radiomics models based on ultrasound and DCE MRI for predicting molecular subtypes of breast cancer. Holistic Integrative Oncology, 5(1), Article 28. https://doi.org/10.1007/s44178-026-00243-2

Image Credits: AI Generated

DOI: 10.1007/s44178-026-00243-2

Keywords: breast cancer, radiomics, molecular subtypes, ultrasound, DCE MRI, machine learning, support vector machine, neoadjuvant therapy, luminal cancer, triple-negative breast cancer, non-invasive diagnosis, external validation

In a study published in Holistic Integrative Oncology, investigators at Shanghai Pulmonary Hospital affiliated with Tongji University, working with collaborators at Jiangxi University of Chinese Medicine and Jilin University, enrolled 410 participants — 308 lung cancer patients and 102 healthy individuals — and analyzed the gases they exhaled using a technique called extractive electrospray ionization mass spectrometry, or EESI-MS. Their machine learning model, built on the molecular fingerprints of breath, distinguished cancer patients from healthy controls with accuracy exceeding 90 percent in both positive and negative ion modes. The work was registered as a clinical trial (NCT06086587) before data collection, lending additional rigor to the design.

The underlying biology is elegantly simple. Tumors are metabolic engines, and their altered chemistry releases volatile organic compounds, or VOCs, that diffuse from cancer cell membranes and surrounding tissue into the bloodstream. Once in the blood, these molecules travel to the lungs, cross the alveolar membrane, and are exhaled. Each breath therefore carries a chemical record of what is happening at the tissue level — a molecular signature that, in principle, can be read without a single needle, scalpel, or X-ray. Previous studies have shown that compounds such as pentane, hexane, heptane, and nonane are elevated in the breath of lung cancer patients, but small sample sizes and the absence of independent validation cohorts have kept breath analysis out of the clinic.

The Shanghai-led team addressed both problems. Participants were recruited between January and August 2024 and randomly allocated into three independent cohorts using computer-generated random numbers. Cohorts 1 and 2, which together contained 216 lung cancer patients and 74 healthy individuals, were used for metabolite screening, model training, and internal cross-validation. Cohort 3 — 92 cancer patients and 28 healthy controls — was held back as a completely independent testing set to evaluate how well the final model generalized to unseen data. Strict exclusion criteria filtered out anyone with a history of other malignancies, active lung infections, or prior anti-cancer treatment, while healthy volunteers had to show no respiratory symptoms and normal chest imaging.

Sample collection itself was re-engineered from the ground up. Conventional breath studies rely on collection bags such as Tedlar, which fail to concentrate trace compounds and can lose VOCs to condensation on their inner walls, or on thermal desorption tubes, which risk contamination and cross-infection because exhaled breath carries aerosol particles, proteins, nucleic acids, and even viruses. The team fabricated an in-house collector fitted with unidirectional valves and a 0.45-micrometer silver nanoparticle filter on both the blowing and exhausting ports, with an extraction liquid inside the bottle to enrich volatile compounds and an automatic virus inactivation function. Participants fasted for at least six hours, avoided spicy food, alcohol, and coffee the night before, and breathed normally through a latex mouthpiece for five minutes under tightly controlled temperature and humidity. Ambient air was sampled before and after each session and subtracted as background.

The analytical engine, EESI-MS, is where the technique departs sharply from established methods. Gas chromatography-mass spectrometry, the traditional workhorse of VOC analysis, demands complex pretreatment and often takes more than 30 minutes per sample. Commercial sensor arrays detect only single compounds and are notoriously destabilized by water vapor. In EESI, two sprayers converge in front of the mass spectrometer inlet: one delivers neutral sample, the other delivers charged reagent droplets. Through collision and liquid-liquid extraction, analyte molecules are extracted into the reagent and ionized within seconds, with no enrichment or pretreatment required. Crucially, the method captures both volatile and non-volatile compounds — a capability most breath technologies lack entirely. The instrument operated at a resolution of 70,000, scanning masses from 50 to 750 m/z in both positive and negative ion modes.

To distill meaning from the resulting spectral fingerprints, the researchers applied partial least squares discriminant analysis, cross-validated with permutation testing to guard against overfitting. The models separated patients from controls with exceptional fit — R-squared values of 0.998 or higher and Q-squared values above 0.987 in both ion modes across both screening cohorts. Intersecting the significant hits from Cohorts 1 and 2 yielded 17 characteristic compounds in positive ion mode and 11 in negative ion mode, 28 significantly differential metabolites in total, each identified against the Human Metabolome Database at Level 2 confidence. A support vector machine algorithm then built the final diagnostic model, which was tested on the untouched Cohort 3.

The results were remarkable. In positive ion mode, the 17-compound panel achieved an area under the ROC curve of 0.98, with 92.50 percent accuracy, 92.39 percent sensitivity, and 92.86 percent specificity. In negative ion mode, the 11-compound panel reached an AUC of 0.92, with 90.83 percent accuracy and an even higher sensitivity of 94.57 percent, though specificity dropped to 78.57 percent. Cross-validation in the training cohorts was even stronger: AUC values surpassed 0.97, accuracy exceeded 97 percent in positive ion mode, and specificity hit 100 percent in several configurations. Because smoking is the dominant risk factor for lung cancer and a potential confounder of breath chemistry, the team stratified participants by smoking status and found no significant difference in model performance between smokers and non-smokers — evidence that the biomarkers reflect the metabolic reprogramming of cancer cells themselves rather than residues of tobacco exposure.

Among the detected molecules were non-volatile compounds rarely explored in breath research, including adenylosuccinic acid, a key intermediate in purine metabolism whose dysregulation has been linked to tumor growth and drug-tolerant persister states in lung cancer, and lipids such as lysophosphatidylcholine, monoglyceride, and phosphatidic acid. Lysophosphatidylcholine can be converted by phospholipase D into lysophosphatidic acid, a potent proliferation-signaling molecule implicated in tumorigenesis, and plasma levels of LPC are known to be diminished in lung adenocarcinoma patients. Their appearance in exhaled breath opens a new window onto cancer metabolism that conventional VOC-only approaches cannot provide.

The authors are candid about the limitations. This was a single-center, exploratory study, and the control group consisted only of healthy individuals — the far harder clinical challenge of distinguishing lung cancer from benign pulmonary diseases such as COPD or pneumonia remains untested. Breath composition is also sensitive to diet, medications, and environmental exposures. Still, the team argues that the diagnostic threshold can be tuned to clinical purpose: a lower cutoff to maximize sensitivity for mass population screening, and a higher cutoff to maximize specificity in outpatient nodule clinics, reducing unnecessary invasive procedures. With high initial instrument costs offset by reagent-free, seconds-per-sample throughput in centralized hub laboratories, the researchers now plan a prospective, multi-center trial to validate the approach — and to determine whether a simple breath test can truly become the stethoscope of cancer’s earliest whispers.

Subject of Research: Non-invasive lung cancer detection through mass spectrometry-based analysis of exhaled breath biomarkers

Article Title: Mass spectrometry-based analysis of exhaled gases: a promising diagnostic tool for lung cancer detection

Article References: Zhao, W., Li, Y., Chen, X., Zhang, N., Liu, X., Zhao, L., Wang, H., Ye, L., Xu, K., Zhang, W., Chen, Z., Liu, Y., Zhang, Q., Liu, B., Chen, H., Su, R., & He, Y. (2026). Mass spectrometry-based analysis of exhaled gases: a promising diagnostic tool for lung cancer detection. Holistic Integrative Oncology, 5(1), Article 34. https://doi.org/10.1007/s44178-026-00247-y

Image Credits: AI Generated

DOI: 10.1007/s44178-026-00247-y

Keywords: lung cancer, breath analysis, mass spectrometry, EESI-MS, volatile organic compounds, early detection, machine learning, support vector machine, biomarkers, non-invasive diagnosis, metabolomics, clinical study

The new system, described in the journal Discover Artificial Intelligence, is called CGO-PC2BM, short for Cultural Guidance Optimized Patch-Mix Contrastive Learning enabled Convolutional Neural Network Light Gradient Boosting Machine. Behind the unwieldy acronym lies a carefully coordinated pipeline that takes raw respiratory audio, strips away noise, extracts a rich tapestry of acoustic features, and then classifies the recording into categories including healthy, asthma, bronchial, COPD, pneumonia, COVID-19, and symptomatic. On the Asthma Detection Dataset Version 2, the framework achieved a specificity of 96.08 percent, an accuracy of 96.61 percent, an F1-score of 96.39 percent, a precision of 95.76 percent, and a sensitivity of 97.03 percent, numbers that place it ahead of a range of established baselines.

The problem the researchers set out to solve is one that has dogged respiratory-audio analysis for years. Asthma narrows the airways, driven by genetic, immune, and environmental factors such as smoke, exercise, or cold air, and these pathological changes leave fingerprints in the voice and breath. Coughs and wheezes carry diagnostic information, but the acoustic signatures of asthma overlap heavily with those of other respiratory conditions, and simple machine learning models trained on individual feature types tend to be fragile. Earlier approaches, from decision trees and random forests to deep architectures combining convolutional neural networks with LSTM layers and Mel-frequency cepstral coefficients, showed promise but struggled with noise sensitivity, limited generalization to unseen recordings, and high computational demands.

The first pillar of the new framework is a feature-extraction mechanism the authors call Spectrogram Statistical Audio Features, or S2AF. Rather than relying on a single acoustic descriptor, S2AF fuses three complementary views of each recording. VGGish, a pretrained audio network originally developed for large-scale sound classification, contributes high-level semantic embeddings that capture the most representative characteristics of the respiratory signal. A hybrid spectrogram representation combines the Constant-Q Transform, which offers flexible time-frequency resolution with a constant Q-factor and excels at harmonic tracking, with the Short-Time Fourier Transform, which provides robust spectral information over time. Finally, a battery of statistical descriptors, including energy, zero-crossing rate, spectral centroid, spectral flux, spectral rolloff, spectral entropy, chroma features, and Mel-frequency cepstral coefficients, captures the rapidly changing dynamics of wheezes and abnormal breathing patterns across short-term windows, with delta features and their means and standard deviations adding temporal context.

Once these heterogeneous feature streams are concatenated into a unified representation, the framework reshapes them into patches and applies its second innovation: Patch-Mix Contrastive Learning. During training, feature patches from two recordings belonging to the same disease class are blended together, with a mixing coefficient randomly sampled between 0.3 and 0.7. The resulting mixed representations are semantically consistent, because both parents come from the same class, yet they introduce intra-class diversity that forces the model to learn robust latent features rather than memorizing individual recordings. A contrastive loss then pulls the mixed and original representations of a sample together in the embedding space while pushing them away from samples of other classes, sharpening the boundaries between clinically similar conditions. Crucially, no patch-mix augmentation is applied to test samples, which remain untouched during evaluation.

The third pillar is the classification architecture itself. A convolutional neural network, built from Conv2D layers with ReLU activation, max-pooling, flattening, dropout, and dense layers, performs deep feature extraction on the patch embeddings, producing a 32-dimensional latent vector for each recording. That vector is then handed to LightGBM, a gradient-boosting framework that grows decision trees leaf-wise, splitting nodes at the point of maximum loss reduction. The division of labor is deliberate: the CNN excels at learning discriminative representations from complex audio, while LightGBM is fast, efficient with high-dimensional data, and adept at carving out precise decision boundaries. Together they form a deep ensemble that outperforms either component alone.

Tuning this multi-component system is itself a formidable challenge, and that is where the fourth innovation comes in. The Cultural Guidance Optimization Algorithm, or CGOA, is a hybrid of the Walrus Optimization Algorithm and the Coyote Optimization Algorithm. The walrus-inspired component contributes broad exploration of the hyperparameter space, while the coyote-inspired component adds a population-level cultural tendency, a vector derived from the ranked social condition of many candidate solutions rather than from the single best one. This collective knowledge guides the search toward promising regions without collapsing diversity, and a guidance-driven exploitation phase led by the fittest solution helps the optimizer escape local optima. CGOA searched the learning rate between 0.0001 and 0.01, batch sizes of 16 to 128, dropout rates between 0.1 and 0.5, convolutional filter counts of 32 to 128, and training epochs between 20 and 100, maximizing the average of accuracy, sensitivity, and specificity. In convergence tests, CGOA drove the loss down to roughly 1.42 times ten to the minus sixtieth power by epoch 97, dramatically lower than its parent algorithms.

The experimental protocol was designed with care to avoid the pitfalls that often inflate reported results. Recordings from the Asthma Detection Dataset Version 2, which contains 288 asthma, 104 bronchial, 401 COPD, 133 healthy, and 255 pneumonia samples, and from the large-scale COUGHVID crowdsourcing dataset were split 80:20 with stratification to preserve class proportions. Region-of-interest extraction removed silent and low-energy segments by thresholding frame energy at 20 percent of each recording’s maximum amplitude, and spectral gating suppressed background noise before feature extraction. All augmentation, contrastive learning, and hyperparameter optimization were confined to the training subset, preventing information leakage into the test set. Against baselines including BiGRU, CALMNet, Google’s Health Acoustic Representations model, standalone CNN, LightGBM, and the unoptimized PC2BM, the full framework delivered consistent gains, for example improving accuracy over CALMNet by 4.18 percent and over HeAR by 7.26 percent on the asthma dataset, and by 10.23 percent and 9.14 percent respectively on COUGHVID.

The authors are candid about the limits of their work. Statistical significance was not uniform across all metrics and datasets: sensitivity reached significance on the asthma dataset with a p-value of 0.04, while accuracy and sensitivity on COUGHVID did not clear the p-less-than-0.05 threshold. Subject-independent partitioning was not enforced, so recordings from the same participant could appear in both training and test sets, and strict cross-dataset validation, in which a model trained on one dataset is tested on another without retraining, was not performed. A dedicated noise-stress experiment across controlled signal-to-noise ratios was also absent, and detailed class-wise confusion matrices were not retained. The researchers position CGO-PC2BM as a computer-assisted screening tool rather than a replacement for clinical diagnosis, noting that a false-negative asthma prediction could delay care while a false positive could trigger unnecessary follow-up testing.

Even with those caveats, the study marks a meaningful step toward non-invasive, scalable respiratory screening. Because the framework requires only audio, it could be deployed in telemedicine platforms, remote patient monitoring systems, and mobile health applications, bringing diagnostic support to settings where spirometry equipment and specialist physicians are scarce. The authors outline a roadmap for the future that includes subject-independent and cross-dataset validation, multi-center clinical trials, controlled noise-robustness testing, the integration of explainable AI techniques to make the model’s decisions transparent to clinicians, and comparisons with a broader range of hyperparameter optimizers. If those validations succeed, the sound of a cough, analyzed by an algorithm that borrows its search strategy from the social behavior of coyotes, may one day become a routine first line of defense against one of the world’s most common chronic diseases.

Subject of Research: A deep learning framework for detecting and classifying asthma from respiratory audio recordings

Article Title: Cultural guidance optimized patch mix contrastive learning enabled ensemble model for asthma detection and classification from respiratory audio

Article References: Shivapur, S., & Chavan, P. (2026). Cultural guidance optimized patch mix contrastive learning enabled ensemble model for asthma detection and classification from respiratory audio. Discover Artificial Intelligence, 6(1), Article 1325. https://doi.org/10.1007/s44163-026-02308-7

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02308-7

Keywords: asthma detection, respiratory audio, machine learning, contrastive learning, CNN, LightGBM, hyperparameter optimization, spectrogram features, non-invasive diagnosis, deep learning, COUGHVID dataset, telemedicine

MASLD has become one of the most common chronic liver conditions in the pediatric population, tracking closely with the global rise in childhood obesity. The disease involves the abnormal accumulation of fat within liver cells and, in a subset of patients, can progress to inflammation, fibrosis, and eventually cirrhosis. The challenge for clinicians is that the liver is remarkably silent in its early stages: children with accumulating hepatic fat typically feel well, and standard liver enzymes such as alanine aminotransferase can remain within normal limits even when fat infiltration is well established. The current diagnostic gold standard, liver biopsy, is rarely justified in children because of its invasiveness, cost, and sampling variability. Abdominal ultrasound is the usual screening tool, but it is operator-dependent and loses sensitivity when hepatic fat content is modest. This diagnostic gap is precisely what the Chinese team set out to address.

The researchers, led by Huanyu Wang and Zongming Yang of Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, with corresponding authors Shimin Wu and Xiaoping Luo, conducted a cross-sectional study of children with obesity who presented to the hospital’s Department of Pediatrics between January 2023 and September 2024. The final analysis included 84 children, 53 boys and 31 girls, of whom 43—just over half—were diagnosed with MASLD based on abdominal ultrasound findings. Each child underwent anthropometric assessment, including body mass index z-score and waist-to-height ratio, alongside fasting metabolic panels. The team then quantified serum concentrations of five candidate biomarkers using enzyme-linked immunosorbent assays: cathepsin D, leptin, adiponectin, insulin-like growth factor 1, and growth differentiation factor 15.

These five molecules were not chosen at random. Each represents a distinct biological axis implicated in obesity-related liver disease. Cathepsin D is a lysosomal protease involved in intracellular protein degradation, and emerging evidence links it to lipid metabolism, insulin signaling, and hepatic inflammation. Leptin, the prototypical adipocyte-derived hormone, signals energy stores to the brain and participates in immune and fibrotic pathways in the liver. Adiponectin, another adipokine, is generally insulin-sensitizing and anti-inflammatory, and its levels tend to fall as metabolic health deteriorates. Insulin-like growth factor 1 mediates growth hormone action and is often altered in states of insulin resistance. Growth differentiation factor 15 is a stress-responsive cytokine of the transforming growth factor beta superfamily that rises in mitochondrial stress and systemic inflammation, and it has attracted attention as a broad marker of cardiometabolic risk.

The statistical approach was deliberately conservative. Rather than simply asking whether biomarker levels differed between children with and without MASLD, the investigators calculated odds ratios across tertiles of each biomarker after adjusting for established risk factors: age, sex, body mass index z-score, and waist-to-height ratio. This adjustment matters because any marker that merely reflects the degree of obesity adds little clinical value; the question was whether these proteins carry independent information about liver fat beyond what a clinician already knows from measuring a child’s size. The team also tested the shape of each association, allowing for nonlinear relationships rather than assuming a straight line.

The results were striking for cathepsin D. Children in the highest tertile of logarithmically transformed serum CTSD had roughly twelve-fold higher odds of MASLD compared with those in the lowest tertile, with an odds ratio of 12.63 and a 95 percent confidence interval spanning 2.57 to 84.62, and the association was linear across the distribution. Growth differentiation factor 15 also showed a robust positive association: children in the top tertile had nearly five times the odds of MASLD relative to the bottom tertile, with an odds ratio of 4.86 and a confidence interval of 1.14 to 23.57. Leptin told a more nuanced story. After full adjustment, its relationship with MASLD risk was inverse U-shaped, meaning risk peaked at intermediate leptin concentrations rather than rising monotonically. Children in the middle tertile had an odds ratio of 6.70 compared with the lowest tertile, and the test for nonlinearity yielded a p-value of 0.004. Adiponectin and insulin-like growth factor 1, by contrast, lost any significant independent association once conventional risk factors were accounted for.

Association alone, however, does not make a biomarker clinically useful. The researchers therefore asked whether adding each protein to a base prediction model containing age, sex, BMI z-score, and waist-to-height ratio actually improved diagnostic performance. They evaluated this with three complementary metrics: the area under the receiver operating characteristic curve, which measures overall discrimination; the continuous net reclassification improvement, which quantifies how many patients are moved into more accurate risk categories; and the integrated discrimination improvement, which assesses gains in predicted probabilities. Individually, each biomarker produced only marginal and statistically non-significant increases in the optimism-corrected AUC—for example, adding cathepsin D yielded a change in AUC of 0.047 with a p-value of 0.096.

Yet the reclassification analyses told a more encouraging story. Incorporating cathepsin D produced the most prominent improvement in risk stratification, with a continuous net reclassification improvement of 1.147 (95 percent confidence interval 0.152 to 1.657, p = 0.032) and an integrated discrimination improvement of 0.142 (95 percent confidence interval 0.014 to 0.339, p = 0.008). Leptin and insulin-like growth factor 1 also generated significant reclassification gains, with all p-values below 0.05. In practical terms, this means that even when a biomarker does not dramatically shift the average discrimination statistic, it can still meaningfully re-sort individual children into more appropriate risk categories—correctly elevating the estimated risk of those who truly have hepatic steatosis and lowering it for those who do not. For a condition in which the decision to pursue further imaging or specialist referral hinges on risk thresholds, such reclassification can translate directly into better-targeted care.

The biological plausibility of the three signal-bearing markers strengthens the case. Cathepsin D’s strong association fits with its emerging role as a mediator of lysosomal lipid handling and inflammatory signaling in hepatocytes. The leptin finding is particularly intriguing: the inverse U-shaped curve suggests that both relatively low and very high leptin concentrations carry different risk implications than intermediate levels, a pattern that could reflect receptor desensitization, leptin’s dual roles in appetite regulation and hepatic fibrogenesis, or confounding by adiposity distribution that the adjustment model only partially captures. Growth differentiation factor 15’s positive association aligns with its established behavior as a marker of cellular metabolic stress, which would be expected to rise as hepatic fat accumulation imposes mitochondrial and inflammatory burdens on the liver.

The authors are careful to frame these results as preliminary. With only 84 participants, the confidence intervals are wide, and the odds ratios—especially the double-digit estimate for cathepsin D—carry substantial statistical uncertainty. The cross-sectional design also means the study captures a single snapshot; it demonstrates association, not causation, and cannot establish whether elevated cathepsin D or GDF15 precede the development of steatosis or accompany it. The single-center recruitment from one pediatric obesity clinic further limits generalizability, and ultrasound-based diagnosis, while standard in clinical practice, is an imperfect reference. The team explicitly calls for validation in larger, multicenter cohorts before these markers enter clinical decision-making. Still, the study was approved by the Tongji Hospital Ethics Committee and supported by China’s National Key Research and Development Program and the Natural Science Foundation of Hubei Province, and it offers a concrete, testable hypothesis: that a small panel of serum proteins, measured alongside routine anthropometry, could sharpen the non-invasive detection of MASLD in children. If larger studies confirm the reclassification gains observed here, pediatricians may one day identify which of their young patients with obesity carry silent liver disease with a routine blood test—catching, at an early and potentially reversible stage, a condition that otherwise hides in plain sight for decades.

Subject of Research: Serum cardiometabolic biomarkers for diagnosing metabolic dysfunction-associated steatotic liver disease in children with obesity

Article Title: Association and diagnostic value of serum cardiometabolic biomarkers for metabolic-associated steatotic liver disease in children with obesity

Article References: Wang, H., Yang, Z., Wu, S., Weng, Y., Hou, L., Wu, W., & Luo, X. (2026). Association and diagnostic value of serum cardiometabolic biomarkers for metabolic-associated steatotic liver disease in children with obesity. BMC Pediatrics. https://doi.org/10.1186/s12887-026-07697-3

Image Credits: AI Generated

DOI: 10.1186/s12887-026-07697-3

Keywords: MASLD, pediatric obesity, cathepsin D, leptin, GDF15, biomarkers, fatty liver disease, non-invasive diagnosis, risk stratification, ELISA, children, hepatic steatosis

The study, published in JAMA Dermatology, centers on a technology known as line-field confocal optical coherence tomography, or LC-OCT. This imaging platform merges two established techniques into a single device. The first is optical coherence tomography, commonly abbreviated OCT, which uses low-coherence light to capture cross-sectional images of tissue and provides reliable information about the thickness and lateral extent of a tumor. The second is line-field confocal microscopy, which delivers sharply resolved images at the cellular level, allowing clinicians to examine individual cell structures within the upper layers of the skin without taking a biopsy.

The combination is what gives LC-OCT its unusual diagnostic power. Whereas OCT alone offers depth but limited cellular detail, and confocal microscopy alone offers cellular detail but limited depth, the hybrid technique produces vertical, horizontal and fully three-dimensional images of the skin in real time, with a resolution on the order of micrometers, that is, thousandths of a millimeter. For perspective, a single human hair is roughly 50 micrometers thick. The result is a live, non-invasive window into the architecture of the skin that approaches what a pathologist would see under a microscope, but without cutting, staining or waiting for laboratory processing.

LC-OCT is already in clinical use as a standard tool for detecting basal cell carcinoma in its early stages and for monitoring how well non-invasive therapies are working. What the FAU team has added is a layer of artificial intelligence trained to interpret the images. The AI analyzes the optical data as it is acquired and overlays a color-coded probability value onto the image, indicating the likelihood that basal cell carcinoma is present in each region examined. This allows the physician to recognize suspicious areas at a glance, dramatically reducing the cognitive load of scanning large or subtle lesions frame by frame.

Importantly, the researchers emphasize that the algorithm does not replace clinical judgment. It is still the physician who makes the final diagnosis, weighing the AI-assisted image alongside the patient’s history, the appearance of the lesion and, where necessary, a conventional biopsy. The technology is best understood as a decision-support system: it directs attention to areas that merit closer scrutiny and can help distinguish malignant tissue from benign look-alikes, but accountability for the diagnostic decision remains with the treating clinician.

The clinical stakes of earlier detection are considerable. Basal cell carcinoma is a locally aggressive tumor: it spreads within the skin in its vicinity and destroys the tissue it invades, which can damage delicate structures such as the nose or the eyes when it arises on the face. Although it rarely metastasizes, its capacity for local destruction makes timely intervention essential. As Dr. Moritz Ronicke of FAU notes, detecting basal cell carcinoma early allows for less invasive treatment options, and in some cases all that is required is a medicated cream rather than surgical excision. Earlier and more confident diagnosis therefore translates directly into gentler treatment, better cosmetic outcomes and reduced burden on both patients and health systems.

The truly forward-looking element of the FAU work is a concept the researchers call SUBSCAN. Rather than imaging a lesion that has already attracted clinical suspicion, SUBSCAN would extend LC-OCT screening to people who carry a high risk of developing skin cancer but who do not yet have a visible tumor. In principle, the AI-assisted imaging could identify the microscopic signatures of basal cell carcinoma forming beneath an apparently normal skin surface, enabling a diagnosis even earlier in the disease process than current practice allows. If such risk-stratified optical screening were routinely adopted, Ronicke suggests, it could lead to more widespread use of topical creams for treating basal cell carcinoma, in some cases avoiding the need for surgery altogether.

That vision, however, is not yet ready for the clinic. The researchers are careful to point out that SUBSCAN cannot currently be offered routinely, for two principal reasons. First, there is still a lack of data on the sensitivity of the method, meaning it is not yet established how reliably the technique detects all early tumors and how often it might produce false alarms when applied to unselected high-risk skin. Second, the examination remains too time-consuming for routine deployment in a busy dermatology practice. Large prospective studies will be needed to quantify diagnostic performance, streamline the imaging workflow and define precisely which patient groups would benefit most from preventive scanning.

Even so, the trajectory of the technology is striking. Dermatology has been moving for years toward non-invasive diagnostics, from dermoscopy to reflectance confocal microscopy to OCT, each generation trading some simplicity for a deeper and sharper view of living skin. LC-OCT with AI-assisted image recognition represents a convergence of that trend: micrometer-scale, three-dimensional, real-time imaging paired with machine learning that distills complex optical data into an immediately interpretable probability map. For patients at elevated risk of skin cancer, the prospect of a diagnosis delivered before a tumor ever becomes visible, and of treatment with a simple cream instead of a scalpel, marks a compelling glimpse of where early cancer detection may be heading.

Subject of Research: AI-assisted line-field confocal optical coherence tomography for early detection of basal cell carcinoma

Article Title: Detecting skin cancer even before it becomes visible

Article References: Detecting skin cancer even before it becomes visible. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: basal cell carcinoma, skin cancer, LC-OCT, optical coherence tomography, confocal microscopy, artificial intelligence, image recognition, early detection, non-invasive diagnosis, dermatology, JAMA Dermatology, SUBSCAN