The competition, hosted by the American Heart Association’s Center for Technology & Innovation, drew a record number of submissions from health technology companies around the world. Volunteer experts spanning healthcare, science and industry carried out a rigorous multi-stage review before narrowing the field to five finalists. What unites them is a focus on two of the largest burdens in modern medicine: diseases of the heart and of the brain and nervous system. The finalists will present their solutions in the Health Innovation Pavilion during Scientific Sessions 2026, the association’s flagship global event for the latest research and clinical advancements in cardiovascular science, taking place November 6 through 9 at McCormick Place in Chicago. Competition pitches are scheduled for November 7 and 8, with winners announced on November 8.

The first finalist, HEARTio of Pittsburgh, Pennsylvania, is developing investigational artificial intelligence software that analyzes a standard electrocardiogram to estimate obstructive coronary artery disease, the leading cause of heart attacks. The significance of the approach lies in what it does not require: no dyes, no radiation, no incisions and no catheters. Coronary artery disease is typically evaluated with invasive angiography or imaging techniques that carry cost, access and safety considerations, so an algorithm capable of extracting diagnostic signal from the routine ECG, one of the cheapest and most widely available cardiac tests in existence, could dramatically widen the funnel of early detection if its performance holds up in clinical validation.

MEGI, based in Chicago, Illinois, takes aim at a different gap in cardiovascular care with an AI-powered digital health platform focused on women’s heart health, maternal care and hypertension management. High blood pressure remains one of the most consequential and undercontrolled risk factors worldwide, and cardiovascular conditions in women, particularly around pregnancy, have historically been under-recognized. A platform that combines artificial intelligence with continuous digital engagement could help patients and clinicians manage hypertension more proactively, especially in populations where traditional touchpoints with the health system are infrequent.

MedRhythms of Portland, Maine, brings a distinctly different modality to the finalist pool. The company operates at the intersection of digital health and neurotherapeutics, combining sensors, software and music to help people improve their walking and mobility after a neurological injury or disease. The idea builds on a long-observed phenomenon in rehabilitation science: rhythmic auditory stimulation, in which music provides a timing cue for movement, can help patients retrain gait patterns disrupted by stroke or other neurological conditions. By pairing that therapeutic approach with sensor-based measurement, MedRhythms aims to deliver and quantify rehabilitation in ways that were previously confined to the clinic.

Seerlinq, headquartered in Bratislava, Slovakia, addresses one of the most persistent problems in heart failure management: detecting deterioration early enough to intervene before a crisis. The company’s technology estimates cardiac filling pressure non-invasively from a short fingertip photoplethysmography measurement taken at home. Rising filling pressures are a hallmark of worsening heart failure, and today they are often assessed invasively or during hospital visits, by which point a patient may already be decompensating. By catching deterioration weeks before hospitalization, Seerlinq aims to give clinicians time to adjust therapy, potentially averting the costly and dangerous admissions that heart failure so frequently triggers.

The fifth finalist, Vascudyne of St. Paul, Minnesota, works at the frontier of bioengineering. The company has developed bioengineered blood vessels that regenerate into the patient’s own tissue. For coronary bypass patients, the technology could eliminate the need to harvest a vein from the leg, an invasive step that adds morbidity to an already major surgery. For children with congenital defects, the implications may be even more profound: an implant that grows with the patient could spare growing children the repeated surgeries that conventional grafts, which do not grow, currently necessitate.

The competition is deliberately structured for emerging companies whose technologies are clinically informed and validated, who can show early evidence of impact and who have a viable path to scale. Finalists were selected for solutions that may engage patients, support clinicians and demonstrate the potential to improve cardiovascular and brain health outcomes across diverse care settings. During the final presentations, a panel of expert volunteer judges will evaluate each company against three core criteria: the strength of the supporting clinical or pilot data, the innovativeness of the underlying science or technology, and how well the solution addresses a meaningful market need. The judging panels themselves draw from across the clinical and commercial spectrum, including cardiologists, digital health executives, academic leaders and investors from institutions such as Stanford University, Yale School of Medicine, Weill Cornell Medicine, Kaiser Permanente and the University of Texas Southwestern Medical Center.

Behind the competition lies a frustration that the American Heart Association’s leadership has articulated plainly. “Too many promising health technologies never reach the people who could benefit from them most,” said Lee A. Shapiro, J.D., volunteer chair of the American Heart Association and managing partner and co-founder at 7wireVentures, an investment firm headquartered in Chicago. “The American Heart Association’s Health Tech Competition highlights solutions that are grounded in science, informed by real-world needs and designed to improve care for people at risk for or living with cardiovascular and brain health conditions.” The statement captures the program’s central thesis: that scientific grounding and real-world relevance, not novelty alone, should determine which technologies earn attention and resources.

The association has also been careful to set expectations about what finalist status means. Selection as a finalist does not constitute an endorsement, certification or validation of a company, product, clinical claim or commercial offering by the American Heart Association. That distinction matters in a field where digital health claims often outpace evidence, and it signals that the competition is a platform for scrutiny rather than a seal of approval. Even so, the visibility that comes with presenting at Scientific Sessions, before an audience of clinicians and researchers from around the world, represents a meaningful accelerant for young companies navigating the long road from promising data to adopted practice.

Taken together, the five finalists sketch a portrait of where cardiovascular and neurological care may be heading. Artificial intelligence is being applied to the humble ECG and to the management of hypertension and maternal health. Sensors and music are being repurposed as rehabilitation tools. A fingertip measurement at home may soon flag heart failure weeks before the hospital does. And bioengineered vessels may grow alongside the children who receive them. None of these outcomes is guaranteed; each depends on the clinical evidence, scientific innovation and market fit that the judges will weigh in Chicago. But the competition’s premise is that the bottleneck in health technology is often not the science itself. It is the pathway between a validated idea and the clinic, and it is precisely that pathway the American Heart Association is trying to shorten.

Subject of Research: The American Heart Association's 2026 Health Tech Competition finalists developing cardiovascular and brain health technologies

Article Title: Health tech competition finalists advance solutions for heart and brain health

Article References: Health tech competition finalists advance solutions for heart and brain health. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: American Heart Association, Health Tech Competition, cardiovascular disease, artificial intelligence, electrocardiogram, coronary artery disease, hypertension, heart failure, digital health, neurorehabilitation, bioengineered blood vessels, Scientific Sessions 2026

The review, led by Thanuja M and Soumyashree M. Panchal of the Manipal Institute of Technology Bengaluru together with Prasanna Lakshmi G S of Sai Vidya Institute of Technology, followed the PRISMA 2020 guidelines that govern rigorous evidence synthesis. The team searched PubMed, IEEE Xplore, Scopus, Web of Science, and other major databases, screening 1,247 records before settling on the 83 studies that met their inclusion criteria. Each study was then scored with a customized quality-assessment tool covering data sources, preprocessing clarity, model development rigor, evaluation practices, external validation, reproducibility, and interpretability. The verdict was sobering: only 32 studies, or 38.6 percent, showed a low risk of bias. Another 49.4 percent carried moderate risk, and 12 percent were judged high risk. In other words, nearly two-thirds of the published evidence base for AI-based ECG diagnosis rests on studies with methodological weaknesses that could inflate or distort their reported performance.

The electrocardiogram itself is a deceptively simple signal. Electrodes on the skin capture the heart’s electrical activity, producing the familiar P wave, QRS complex, and T wave that correspond to atrial depolarization, ventricular depolarization, and ventricular repolarization. Clinicians read these waveforms for subtle clues: a slightly elevated ST segment may signal an evolving myocardial infarction, an irregularly irregular rhythm points to atrial fibrillation, and a prolonged QT interval suggests vulnerability to dangerous arrhythmias. These are pattern-recognition tasks that demand years of training and remain subject to variability between observers. Automating them has been a dream almost as old as the ECG itself, but early rule-based systems, which measured predefined features and compared them against fixed criteria, plateaued well below expert human performance because they could not capture the holistic, nuanced pattern recognition that experienced interpreters develop.

Machine learning changed the equation by learning features directly from data rather than relying on hand-crafted measurements. The review found that traditional machine learning methods, used in 49.4 percent of the studies surveyed, still hold real advantages in specific settings. Ensemble classifiers such as Random Forest and Gradient Boosting consistently outperform single classifiers by 4 to 7 percent in accuracy, and expert-derived features grounded in clinical knowledge of ECG morphology add another 5 to 10 percent compared with generic statistical features. Traditional approaches also demand far less training data, making them the sensible choice when fewer than 1,000 recordings are available, and they offer inherent interpretability through feature importance scores and coefficient magnitudes. But on raw or minimally preprocessed signals, they trail deep learning by 8 to 15 percent, a gap that explains why neural networks have come to dominate the field, appearing in 78 percent of studies from 2023 to 2025.

Among deep architectures, convolutional neural networks reign supreme, employed in 67 percent of the reviewed studies. CNNs apply learnable filters across the one-dimensional ECG signal, automatically detecting local morphological patterns such as QRS complexes and ST-segment changes, with pooling layers providing translation invariance so that clinically relevant features are recognized regardless of their exact timing. Their weakness is temporal: they struggle to capture dependencies spanning multiple heartbeats. That role falls to Long Short-Term Memory networks, a form of recurrent architecture whose internal memory gates preserve information across hundreds of time steps, making them well suited to rhythm analysis and to catching paroxysmal atrial fibrillation hidden in long recordings. Hybrid CNN-LSTM models, which pair convolutional feature extraction with temporal modeling, appeared in 34 percent of recent studies and consistently outperform standalone architectures by 3 to 8 percent on benchmarks, particularly for conditions involving both morphological abnormalities and rhythm disturbances. Emerging tools, including transformer self-attention models in 15 percent of 2024 and 2025 studies and graph neural networks that model relationships between ECG leads, are pushing the frontier further, while transfer learning from large repositories such as PTB-XL, which contains 21,799 recordings, cuts training data requirements by 60 to 80 percent.

Before any model sees data, however, the signal must be cleaned, and the review devotes detailed attention to this unglamorous but critical stage. Raw ECG recordings are contaminated by baseline wander, powerline interference at 50 or 60 hertz, muscle activity, respiratory artifacts, and motion noise from patient movement or electrode displacement. Traditional frequency-selective filters, including bandpass filters with cutoffs between 0.5 and 150 hertz and Butterworth designs prized for their maximally flat response, remain workhorses, but newer techniques are gaining ground. Wavelet transforms decompose the signal in time and frequency simultaneously, isolating noise while preserving transient diagnostic features. Empirical mode decomposition adaptively breaks the signal into intrinsic components for selective noise removal. Adaptive filters such as Least Mean Squares algorithms adjust their parameters dynamically, and Kalman filters support real-time estimation. Most strikingly, neural networks themselves, including denoising autoencoders and generative adversarial networks, now learn noise patterns directly from data. Segmentation, normalization, and R-peak alignment complete the pipeline, and the review notes that even deep learning models, which supposedly bypass manual feature engineering, remain heavily dependent on high-quality preprocessing.

The most alarming findings concern the data itself. Ninety-five percent of the reviewed studies relied on publicly available ECG repositories, with only 3 percent using private institutional data and 2 percent collecting their own recordings. That reliance on open datasets aids reproducibility but concentrates risk: 78 percent of public ECG data originates from North America or Europe, ethnicity is reported in fewer than 10 percent of datasets, pediatric populations account for less than 3 percent of recordings, and average female representation stands at 41.5 percent. Annotation quality is equally uneven, with only 35 percent of datasets specifying annotator qualifications and just 22 percent reporting inter-annotator agreement. The consequences are measurable. Studies performing cross-dataset validation report average performance degradation of 8 to 15 percent when models are tested on data from different sources, and only 28 percent of studies include any form of external validation. A model trained primarily on middle-aged European patients may simply fail when deployed in a demographically different hospital, a phenomenon known as dataset shift that raises serious fairness and equity concerns.

Transparency poses another barrier. Deep networks function as black boxes, producing accurate predictions with little insight into their reasoning, and clinicians justifiably hesitate to act on alerts they cannot interrogate. Explainable AI methods aim to open that box: SHAP values, rooted in game theory, quantify each feature’s marginal contribution to a prediction, while LIME approximates a complex model locally with an interpretable surrogate. Visual techniques such as Class Activation Maps and saliency maps highlight which portions of the waveform drove a classification, letting observers see exactly which ECG segments influenced the model’s decision. Yet only 52 percent of recent studies incorporate any explainability method, demographic fairness assessments remain rare, and the review argues that current explanations often satisfy technical curiosity without answering the questions clinicians actually ask, such as whether a model’s reasoning aligns with established pathophysiology and whether its recommendation can be trusted for a specific patient.

The reproducibility picture is equally troubling. Only 29 percent of reviewed studies made their code publicly available, preprocessing details were frequently too sparse to permit replication, and the field lacks standardized benchmarks, making cross-study comparison nearly impossible. Few papers include comprehensive ablation analyses quantifying which preprocessing steps or architectural components actually drive performance. Evaluation practices compound the problem: most studies emphasize overall accuracy while neglecting class-specific performance, calibration, and uncertainty quantification, all of which matter enormously in clinical decision-making where a false negative on a lethal arrhythmia carries very different stakes than a false alarm.

The authors close with a roadmap that reads as both a critique and a call to action. In the near term, they urge standardized evaluation benchmarks, mandatory demographic reporting, and routine external validation; in the medium term, multi-center data consortia and explainability integrated with clinical workflows; and in the long term, prospective validation in diverse practice settings and continuous learning systems that adapt as populations change. They also point to federated learning, which trains models across institutions without centralizing sensitive patient data, as a promising route around privacy barriers. The convergence of abundant ECG data, powerful architectures, and pressing clinical need creates a genuine opportunity, the review concludes, but seizing it will require engineers to work alongside clinicians from project inception, researchers to prize generalizability over benchmark scores, and the whole community to treat transparency and fairness not as optional extras but as the foundation of AI-assisted cardiovascular care.

Subject of Research: Artificial intelligence methods for ECG-based cardiovascular disease prediction

Article Title: A systematic survey of artificial intelligence methods for ECG-based cardiovascular disease prediction

Article References: M, T., Panchal, S. M., & S, P. L. G. (2026). A systematic survey of artificial intelligence methods for ECG-based cardiovascular disease prediction. Discover Artificial Intelligence, 6(1), Article 1259. https://doi.org/10.1007/s44163-026-02170-7

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02170-7

Keywords: artificial intelligence, electrocardiogram, cardiovascular disease, deep learning, machine learning, explainable AI, systematic review, clinical validation, dataset bias, CNN, LSTM, signal preprocessing

The new system, called JEPA-Mamba, is the work of Luyang Dai and Yuwen Li of the School of Instrument Science and Engineering at Southeast University. Its name compresses two of the most consequential ideas in contemporary machine learning. The first is the Joint-Embedding Predictive Architecture, or JEPA, a self-supervised learning strategy that shot to prominence when Meta AI researchers demonstrated it on natural images in 2023. The second is Mamba, a state-space-based sequence model introduced in late 2023 that promises to handle very long data streams with linear computational cost, sidestepping the quadratic bottleneck of conventional attention-based transformers. Dai and Li fused these ingredients with a custom attention module tailored to the peculiar anatomy of a twelve-lead ECG, and the resulting model outperformed established self-supervised baselines across a battery of diagnostic classification tasks.

To understand why this matters, it helps to grasp how JEPA differs from the self-supervised methods that preceded it. Early approaches to learning without labels, such as contrastive learning, worked by showing the network two augmented views of the same recording and teaching it to pull their representations together while pushing apart representations of different samples. This required careful engineering of the augmentations, and for medical signals it is far from obvious which distortions are safe: flip a T-wave or shift a baseline and you may destroy the very clinical information the model needs. JEPA takes a different path. Instead of comparing raw signals, it predicts representations of masked regions in embedding space. In JEPA-Mamba, the authors apply high-ratio contiguous-block masking, hiding large stretches of the waveform, and train an exponential moving average encoder alongside a narrow predictor network that must guess what the hidden segments mean. Because the model operates in abstract feature space rather than pixel-by-pixel reconstruction, it learns high-level diagnostic structure without any hand-crafted augmentation, a property the authors argue is crucial for medical time series.

The second pillar, the bidirectional Mamba encoder, addresses a fundamental tension in ECG analysis. A twelve-lead recording sampled over ten seconds, and often much longer in clinical practice, is a long sequence, and transformers that process it with full self-attention become computationally expensive as length grows. Recurrent networks process sequences cheaply but struggle to retain distant dependencies. Mamba, built on selective structured state spaces, scans a sequence with input-dependent dynamics that let it filter and remember information selectively, achieving linear-time processing that scales gracefully. Dai and Li run Mamba in both directions, so the model can integrate context from earlier and later parts of the cardiac cycle simultaneously. They enhance this backbone with rotary position embeddings, which encode temporal position in a way that respects relative offsets, and a register mechanism that helps the network organize global information, then couple the state-space modeling with convolutional waveform extraction that captures fine-grained morphological features of P waves, QRS complexes, and T waves.

Perhaps the most domain-aware component is the channel-temporal attention module, or CTAM. A standard clinical ECG is not one signal but twelve simultaneous recordings, each lead viewing the heart’s electrical activity from a different spatial angle, and different leads carry different diagnostic weight for different conditions. CTAM recalibrates the contributions of the twelve leads adaptively, borrowing ideas from squeeze-and-excitation and convolutional block attention designs, while also focusing temporal attention on the clinically salient segments of the recording. In other words, the model learns both which leads matter and which moments in time matter, rather than treating the full multi-lead tensor as a homogeneous block of numbers. Attention maps produced in the study suggest the trained model concentrates on the regions cardiologists would also scrutinize.

The evaluation was deliberately stringent. The team pretrained the encoder on MIMIC-IV-ECG, a large public dataset of diagnostic ECGs matched to clinical records, and then transferred the learned representations to PTB-XL, the largest publicly available twelve-lead ECG benchmark, using a strict patient-level split so that no patient appeared in both training and test sets. This cross-dataset setup is a demanding test of generalization, because the recordings come from different devices, patient populations, and labeling conventions. On the fully annotated PTB-XL tasks, JEPA-Mamba achieved macro-averaged area under the ROC curve of 0.9296 plus or minus 0.002 and 0.9306 plus or minus 0.002, scores that place it ahead of prior self-supervised approaches on most downstream tasks. Crucially, it beat the most direct JEPA-based competitor in the ECG domain across every task tested, and the improvements over external baselines on the super-class, subclass, and diagnostic-level tasks were statistically significant.

The authors did not rely on headline numbers alone. They conducted ablation studies to quantify the contribution of each architectural component, confirming with DeLong tests that the differences observed when removing the bidirectional design, the register mechanism, or the attention module were statistically meaningful rather than noise. Per-class performance analyses revealed where the model excelled and where it struggled, while t-SNE visualizations of the learned embeddings showed diagnostic categories separating into coherent clusters, evidence that the representations encode clinically relevant structure even before fine-tuning. Robustness experiments exposed the model to noise and baseline wander, two pervasive artifacts of real-world ECG acquisition, and demonstrated that its performance degrades gracefully, an important property for deployment outside pristine laboratory conditions.

The significance of this work extends beyond one benchmark. Self-supervised pretraining on large unlabeled archives addresses the labeled-data bottleneck that has slowed clinical AI adoption, and the cross-dataset results suggest the learned features are not merely tailored to one recording environment. Because Mamba operates in linear time, the architecture is well suited to the long, continuous streams produced by wearable monitors and intensive-care telemetry, where transformer-based models become impractical. The pipeline also requires no hand-crafted augmentations, removing a layer of human guesswork from the training recipe. Together these properties sketch a realistic path toward foundation-model-style ECG encoders that hospitals could adapt to their own populations with modest local fine-tuning.

Caveats remain, and the authors are careful to frame their findings as preliminary evidence rather than a finished clinical tool. The study used only public, de-identified datasets, so the model has not yet faced the messier distributions of routine hospital practice, and regulatory validation, prospective trials, and integration with clinical workflows all lie ahead. Nonetheless, the combination of JEPA-style predictive learning, bidirectional state-space modeling, and channel-temporal attention represents a coherent template for physiological signal representation, one that could plausibly extend to electroencephalography, arterial pressure waveforms, and other multichannel biosignals where labels are scarce and sequences are long.

What makes the story compelling is its timing. Self-supervised learning transformed computer vision and natural language processing over the past half-decade, and medicine has been waiting for an equivalent breakthrough that respects the constraints of clinical data. JEPA-Mamba shows that the joint-embedding predictive paradigm, when rebuilt around the physics and cardiology of the heartbeat rather than borrowed wholesale from images, can deliver state-of-the-art diagnostic representations without a single expert label during pretraining. If the approach generalizes as its cross-dataset results suggest, the machines that watch over our hearts may soon learn their craft the way residents once did: by observing vast numbers of patients, guided only by the signals themselves.

Subject of Research: Self-supervised deep learning for electrocardiogram representation and classification

Article Title: JEPA-Mamba: a self-supervised joint-embedding architecture with bidirectional Mamba for ECG representation learning

Article References: Dai, L., & Li, Y. (2026). JEPA-Mamba: a self-supervised joint-embedding architecture with bidirectional Mamba for ECG representation learning. Biomedical Engineering Letters. https://doi.org/10.1007/s13534-026-00620-w

Image Credits: AI Generated

DOI: 10.1007/s13534-026-00620-w

Keywords: electrocardiogram, self-supervised learning, JEPA, Mamba, state space models, attention mechanism, PTB-XL, MIMIC-IV-ECG, transfer learning, cardiovascular disease, biomedical engineering, ECG classification

Turner syndrome is a sex chromosome disorder affecting roughly one in 2,000 to 2,500 live female births, caused by a complete or partial absence of one X chromosome. Its classic features include short stature, gonadal dysgenesis that leads to ovarian failure, and a striking degree of clinical heterogeneity, meaning that no two patients present quite the same way. For decades, the medical focus was on the developmental and reproductive consequences of the condition. But as cardiac surgery, growth hormone therapy, and hormone replacement have improved survival, a new problem has moved to the foreground: adults with Turner syndrome are living long enough to accumulate the chronic diseases of metabolism.

That shift is precisely what makes this case report timely. The authors, led by Zhenhan Li of Chongqing College of Traditional Chinese Medicine along with colleagues at Chongqing Traditional Chinese Medicine Hospital and Chongqing Medical University, describe a young patient whose metabolic assessments revealed significant insulin resistance accompanied by metabolic dysfunction-associated steatotic liver disease, or MASLD, and hyperlipidemia. Insulin resistance is the physiological state in which the body’s tissues respond weakly to insulin, forcing the pancreas to secrete ever-larger amounts of the hormone to keep blood glucose in check. Over time, this compensation fails, and type 2 diabetes mellitus emerges, which is exactly what had happened in this patient by the time of her admission.

The technical heart of the paper is what the authors call a metabolic–cardiac interaction model. Rather than treating diabetes, hypertension, fatty liver, and tachycardia as four separate diagnoses, the model frames them as nodes in a network of reinforcing feedback loops. Insulin resistance promotes fat accumulation in the liver; hepatic fat worsens systemic insulin resistance and dyslipidemia; elevated free fatty acids and inflammatory signaling impair vascular function, contributing to hypertension; and hyperglycemia itself increases sympathetic nervous system tone, the body’s fight-or-flight drive, which raises heart rate. In this framing, the cardiovascular system is not merely a bystander damaged by metabolic disease but an active participant whose behavior reflects the state of the whole network.

It is within this framework that sinus tachycardia takes on new significance. Sinus tachycardia means the heart’s natural pacemaker, the sinoatrial node, is firing faster than normal, typically above 100 beats per minute at rest. It is a common finding with many benign explanations, including anxiety, fever, dehydration, anemia, thyroid excess, caffeine, and medications. During the January 2022 admission, persistent sinus tachycardia was documented in the setting of severe hyperglycemia and multiple metabolic abnormalities, yet the underlying cause remained uncertain even after clinical evaluation. The authors propose that, in a patient with Turner syndrome and metabolic deterioration, this otherwise nonspecific finding might represent a clinically accessible signal, something a clinician can detect with nothing more sophisticated than an electrocardiogram or a pulse reading.

The appeal of that idea lies in its practicality. Modern metabolic medicine has powerful tools, from continuous glucose monitors to liver elastography to detailed lipid fractionation, but these are expensive and unevenly distributed. A heart rate measurement costs nothing. If resting heart rate turns out to track with metabolic deterioration in Turner syndrome, it could serve as a triage tool, flagging patients who need deeper investigation before complications such as advanced liver fibrosis or diabetic vascular disease take hold. The authors are careful to frame this as hypothesis-generating rather than proven: their conclusion explicitly states that prospective studies are needed to validate the idea, and that the diagnostic value of electrocardiographic changes such as sinus tachycardia in early intervention remains to be established.

The biological plausibility of a metabolic-heart-rate link rests on several well-characterized mechanisms. Chronic hyperglycemia increases blood osmolality and plasma volume shifts, and it is associated with elevated circulating catecholamines, the hormones adrenaline and noradrenaline that drive cardiac acceleration. Insulin itself has cardiovascular effects, stimulating sympathetic outflow even as it fails to clear glucose effectively, a phenomenon sometimes described as selective insulin resistance in the vasculature. Meanwhile, MASLD is increasingly understood not as a passive fat deposit but as a metabolically active organ state, exporting inflammatory cytokines and triglyceride-rich lipoproteins that stiffen arteries and perturb autonomic regulation of the heart. Hypertension, in turn, increases cardiac afterload, and the heart may compensate with a faster rate. Each loop in the network feeds the others, which is why the authors chose the language of an interaction network rather than a simple chain of cause and effect.

Turner syndrome adds its own cardiovascular layer to this picture. The condition is well known for structural cardiac abnormalities, including bicuspid aortic valve, coarctation of the aorta, and aortic dilatation, which is why guidelines recommend regular cardiovascular imaging for affected individuals. But the metabolic comorbidities documented in this case, including type 2 diabetes, hypertension, MASLD, and hyperlipidemia, are increasingly recognized as major contributors to morbidity in adults with the condition. Estrogen deficiency from gonadal dysgenesis may compound the problem, since estrogen influences insulin sensitivity, lipid metabolism, and vascular health, and the timing and form of hormone replacement therapy can shape metabolic trajectories. The case report lists hormone replacement therapy among its key abbreviations, reflecting its place in the patient’s overall management.

What the authors advocate, on the strength of this single case, is a multidimensional and multidisciplinary management strategy. In practice, that means an adult with Turner syndrome and complex metabolic disease should not be managed by any one specialty alone: endocrinology for glycemic control and hormone replacement, cardiology for rhythm and structural assessment, hepatology for steatotic liver disease, and often nutrition and metabolic medicine for weight and lipid management. The metabolic interaction network model is, in part, an argument for why this fragmentation is dangerous. If the diseases reinforce one another, treating them in isolation risks optimizing one node while the network as a whole continues to deteriorate.

The report, prepared according to the CARE guidelines for case reports and published with the patient’s written informed consent, is open access and carries the DOI 10.1186/s12902-026-02661-9. Its authors are candid about its limits: a single patient cannot establish that sinus tachycardia predicts metabolic decline, and the uncertainty about the cause of this patient’s tachycardia is stated plainly rather than resolved. But case reports have historically played exactly this role in medicine, converting a bedside observation into a testable question. Here, the question is whether an ordinary electrocardiogram, interpreted through the lens of a metabolic interaction network, could give clinicians treating Turner syndrome an early warning that is currently missing. The answer will require the prospective cohort studies the authors call for, following groups of patients with Turner syndrome over time to see whether resting heart rate and other electrocardiographic changes track with insulin resistance, liver fat, and blood pressure before overt disease appears. If they do, one of medicine’s oldest vital signs may gain a new role in one of its most metabolically complex genetic conditions.

Subject of Research: Metabolic comorbidities and sinus tachycardia as a potential early warning sign in Turner syndrome

Article Title: Metabolic interaction network and cardiac abnormalities: a case report of Turner Syndrome complicated by T2DM, hypertension, fatty liver, and sinus tachycardia

Article References: Zhou, L., Chen, Z., Bian, Y., Jing, C., & Li, Z. (2026). Metabolic interaction network and cardiac abnormalities: a case report of Turner Syndrome complicated by T2DM, hypertension, fatty liver, and sinus tachycardia. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02661-9

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02661-9

Keywords: Turner syndrome, sinus tachycardia, type 2 diabetes, insulin resistance, MASLD, hypertension, metabolic syndrome, hormone replacement therapy, cardiovascular risk, case report, multidisciplinary management, electrocardiogram

Researchers led by Salah Al-Zaiti and colleagues tested an AI-enhanced ECG model, known as the ECG-SMART AI score, in symptomatic outpatients scheduled for elective coronary angiography at a tertiary medical center. The question was deceptively simple: could a model originally trained to spot acute coronary occlusion, the complete blockage of a heart artery during a heart attack, also identify patients with angiographically significant coronary artery disease before they ever reached the catheterization laboratory?

The stakes are considerable. Each year, large numbers of patients with chest pain or other cardiac symptoms undergo elective coronary angiography, an invasive procedure in which a catheter is threaded to the heart and contrast dye reveals the anatomy of the coronary arteries. Yet a substantial fraction of these procedures finds no artery narrow enough to warrant intervention, exposing patients to radiation, contrast dye, catheter-related risks, and cost without therapeutic benefit. A noninvasive tool that could refine the pretest probability of significant disease would allow clinicians to select candidates for angiography more judiciously.

The study enrolled 363 patients, with an average age of about 59 years and a slight male majority. Before their procedures, each patient received a standard 12-lead ECG, which was then analyzed offline by the previously validated AI model. The researchers classified patients into two risk categories using predefined thresholds: low risk or intermediate-to-high risk. The benchmark for success was strict and clinically meaningful: angiographically significant coronary artery disease, defined as at least 70 percent narrowing in a major epicardial vessel or at least 50 percent narrowing of the left main coronary artery, the vessel that supplies most of the heart’s blood.

The results were striking. Overall, 36.1 percent of the patients had angiographically significant disease. The AI score flagged 44 percent of the cohort as intermediate-to-high risk and 56 percent as low risk. Crucially, after statistical adjustment for conventional clinical characteristics such as standard risk factors and symptoms, the intermediate-to-high risk classification remained independently associated with significant coronary disease, carrying an odds ratio of 3.12. In plain terms, patients the AI flagged as higher risk were roughly three times more likely to have a critically narrowed artery than comparable patients the model deemed low risk.

Discrimination, the statistical measure of how well a test separates the diseased from the healthy, was good: the model achieved an area under the receiver operating characteristic curve of 0.79, with a 95 percent confidence interval of 0.74 to 0.84. An AUROC of 0.5 would indicate performance no better than a coin flip, while 1.0 represents perfect classification. At a clinically useful operating point, the model delivered 60 percent precision at 80 percent recall, meaning that when it identified patients as high risk, a majority indeed had significant disease, and it captured most of the patients who truly did.

What makes this finding scientifically notable is the concept of generalizability. The model was not built for this task. It was developed to detect acute coronary occlusion, a different and more dramatic physiological state in which the ECG often shows dramatic changes such as ST-segment elevation. Coronary artery disease in stable, symptomatic outpatients is subtler: a 70 percent stenosis may produce only faint perturbations in the electrical signal, perturbations that human readers and conventional criteria routinely miss. That a model trained on one condition can transfer to another suggests the deep learning system has learned features of the ECG that reflect underlying coronary physiology rather than the surface signatures of a single disease state.

The technical implications extend beyond cardiology. Modern AI-ECG systems typically use convolutional or transformer-based neural networks that consume the raw voltage waveforms of all twelve leads simultaneously, detecting spatial and temporal patterns invisible to the naked eye. These networks can pick up on subtle changes in QRS morphology, T-wave vector, and beat-to-beat variability that encode information about myocardial ischemia, scar, and electrical remodeling. The ECG-SMART study adds to a growing body of evidence that such models can infer structural and hemodynamic information, from reduced ejection fraction to atrial fibrillation risk, from signals once thought to carry only rhythm and conduction data.

Cautious interpretation remains essential. The study was conducted at a single tertiary center in a population already selected for elective angiography, so the prevalence of significant disease, 36 percent, is far higher than in the general population of symptomatic outpatients. Precision and recall will shift when the same model is applied to lower-prevalence settings, and prospective validation in independent, more diverse cohorts is the necessary next step. The authors also emphasize that the AI score is not a replacement for clinical judgment or for stress testing, computed tomography angiography, or other established modalities, but rather a complement that adds noninvasive information to refine pretest probability.

Even with those caveats, the practical appeal is hard to overstate. The 12-lead ECG is cheap, fast, painless, and universally available, from rural clinics to emergency departments worldwide. If an AI layer can convert this ubiquitous tracing into a reliable estimate of the likelihood of obstructive coronary disease, clinicians could spare low-risk patients an unnecessary invasive procedure while accelerating high-risk patients toward definitive care. The ECG-SMART AI score, validated here in an elective angiography population, offers a compelling glimpse of that future, one in which a century-old diagnostic tool, augmented by machine learning, becomes a far sharper instrument for seeing inside the heart’s arteries without ever breaking the skin.

Subject of Research: AI-enhanced electrocardiography for detecting obstructive coronary artery disease before elective angiography

Article Title: Diagnostic performance of the ECG-SMART AI score for detecting angiographically significant coronary artery disease in patients undergoing elective coronary angiography

Article References: Bani Hani, D. A., Alshraideh, J. A., Alduraidi, H., Saleh, A., Riek, N. T., Daoud, K., Ji, R. Q., Saba, S., Callaway, C. W., & Al-Zaiti, S. (2026). Diagnostic performance of the ECG-SMART AI score for detecting angiographically significant coronary artery disease in patients undergoing elective coronary angiography. PLOS Digital Health, 5(10), e0001701. https://doi.org/10.1371/journal.pdig.0001701

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001701

Keywords: AI-ECG, coronary artery disease, electrocardiogram, coronary angiography, deep learning, diagnostic accuracy, PLOS Digital Health, acute coronary occlusion, machine learning, cardiology, pretest probability, prospective cohort study

The work, published in Communications Engineering by a collaboration spanning the Toyota Research Institute of North America, the University of Southern California, and Northwestern University, combines two ideas that rarely appear together: a nineteenth-century acoustic device known as a Helmholtz resonator, and a modern neural network trained to translate sound into voltage waveforms. The result is a proof-of-concept sensor system that listens to the heart in the 50 to 120 hertz frequency band and reconstructs the timing of each cardiac cycle with clinically meaningful accuracy.

A Helmholtz resonator is deceptively simple. It consists of a cavity of air connected to the outside world through a narrow neck — the physics behind the tone produced when you blow across the mouth of an empty bottle. The volume of air in the neck acts as a mass and the air in the cavity acts as a spring, and together they oscillate at a characteristic resonant frequency determined by the geometry of the neck and cavity. By carefully designing those dimensions, the researchers engineered a cavity that resonates precisely within the low-frequency band where cardiac acoustic activity concentrates. At resonance, the device amplifies the faint pressure fluctuations produced by the beating heart, boosting a signal that would otherwise drown in ambient noise and in the sensitivity limits of conventional microphones.

This passive amplification is what distinguishes the approach from earlier attempts at acoustic heart sensing. Microphones and accelerometers can pick up cardiac sounds, but the signals of interest sit at extremely low frequencies, where electronic noise, motion artifacts, and environmental rumble dominate. A resonant cavity tuned to the cardiac band acts as a mechanical filter and amplifier in one, emphasizing the physiological content before the signal ever reaches an electronic transducer. Because the amplification is acoustic rather than electronic, it adds no noise of its own and requires no power — a meaningful advantage for any future device that must operate unobtrusively for hours at a time.

Collecting the sound, however, is only half the challenge. The acoustic signal contains rich information about the timing of cardiac events, but it does not look like an electrocardiogram. To bridge that gap, the team developed a neural network-based machine learning algorithm that learns the mapping between the resonator-enhanced acoustic signal and the simultaneously recorded electrical activity of the heart. The network is trained on paired data — acoustic recordings and reference ECG waveforms acquired at the same time — and learns to reconstruct the characteristic features of the cardiac electrical cycle from the acoustic input, including the timing of the R-peak, the sharp spike that marks ventricular depolarization and anchors nearly every heart-rate analysis.

The performance figures reported in the study demonstrate that this translation is more than a curiosity. Compared against a simultaneously acquired reference ECG, the system estimated heart rate with an error of 4.23 plus or minus 4.50 beats per minute, and it located R-peaks with a mean absolute timing error of 36.50 plus or minus 53.17 milliseconds. For context, the two quantities most commonly extracted from ambulatory cardiac monitoring — heart rate, the average beat-to-beat value over time, and heart rate variability, the fluctuation of beat-to-beat intervals — depend directly on how accurately the inter-beat interval can be measured. An error of a few dozen milliseconds in R-peak timing is small enough to support meaningful inter-beat interval determination, which is precisely what the team set out to establish.

Why go to such lengths to recreate an ECG from sound when reliable electrode-based monitors already exist? The answer lies in the friction of everyday health monitoring. Standard ECG measurement requires galvanic contact with the skin, typically through adhesive electrodes or the conductive surfaces of a wrist-worn device. That contact introduces practical constraints: it demands correct placement, adequate skin preparation, sustained pressure, and user compliance. Contact-based sensing dominates today’s wearable landscape, but it is awkward in situations where the user is seated, clothed, or engaged in activities that make electrode contact unreliable. A sensor that works entirely through sound could in principle be embedded into a seat, a headrest, or a cabin structure, monitoring cardiac activity without the wearer ever putting anything on.

That use case is not incidental to the research. The authors explicitly point toward the potential extension of the hardware-plus-software system for non-invasive cardiac biomonitoring of seated users in a dynamic vehicle environment. A car cabin is, in this framing, a plausible sensing platform: the occupant sits in a fixed position for the duration of a trip, and surfaces in contact with or near the body could host an acoustic resonator tuned to the cardiac band. In such a scenario, the vehicle itself would continuously estimate the occupant’s heart rate and heart rate variability, potentially flagging stress, fatigue, or emerging cardiac events during driving. The team’s funding disclosure supports this reading — the experimental work was supported in part by MIRISE Technologies, a supplier of in-cabin sensing technologies for the automotive industry.

The commercial trajectory of the idea is already visible in the intellectual property record. Two of the authors hold a granted United States patent titled Vehicle heart rate detection device, resonators, and methods of detecting a heart rate, and a further patent application covering systems and methods for noncontact monitoring of cardiac activities lists several of the same researchers. The published paper itself, meanwhile, is framed deliberately as a design, experimental testing, and proof-of-concept validation — the authors are careful about what has and has not been demonstrated. The measurements were validated against reference ECGs in controlled conditions; translating that performance to the noise, vibration, and motion of a real vehicle remains future work, as the authors themselves discuss.

Even so, the conceptual advance is striking. It suggests a division of labor between physics and machine learning that could reshape how physiological signals are harvested: let a carefully engineered passive structure do the work of isolating and amplifying the signal of interest, then let a trained network do the work of translating that signal into the clinical vocabulary of medicine. The heart has always been audible if you knew how to listen. What this research demonstrates is that, with the right cavity and the right algorithm, the listening can be precise enough to recover the electrical rhythm itself — beat by beat, without touching the patient.

Subject of Research: Non-invasive cardiac sensing using an acoustic Helmholtz resonator and machine learning-based ECG waveform reconstruction

Article Title: Non-invasive cardiac sensing via an acoustic Helmholtz resonator cavity with electrocardiogram waveform reconstruction

Article References: Schmalenberg, P. D., Hamidi Shishavan, H., Singh, P., Avramidis, K., Severgnini, F. M. Q., Pardo, B., Lee, T., & Dede, E. M. (2026). Non-invasive cardiac sensing via an acoustic Helmholtz resonator cavity with electrocardiogram waveform reconstruction. Communications Engineering. https://doi.org/10.1038/s44172-026-00792-4

Image Credits: AI Generated

DOI: 10.1038/s44172-026-00792-4

Keywords: cardiac sensing, Helmholtz resonator, electrocardiogram, machine learning, neural network, heart rate variability, non-invasive monitoring, acoustics, biomonitoring, in-ear sensing, wearable health, Communications Engineering

The core of the proposal is a fuzzy radial basis inference network, abbreviated by the authors as FRBN, wrapped inside a larger architecture that they call a fuzzy inference network with multimodal feature knowledge embedding. The full pipeline consists of five stages: a multi-channel modal feature input layer, a fuzzy radial basis neuron network layer, a modal category feature aggregation layer, a fuzzy rule layer, and finally a T-S fuzzy classifier that delivers the diagnosis. The design philosophy is deliberately different from mainstream deep learning. Rather than asking a large neural network to discover everything from scratch, the model embeds prior category knowledge directly into its structure, using that knowledge to constrain how the network interprets each modality of data.

The technical challenge the authors set out to solve is well defined. Different modalities of medical data differ in their spatiotemporal distribution, their information granularity, and their semantic representation. An electrocardiogram is a high-frequency time series whose diagnostic content lives in the shape of voltage waveforms over milliseconds. An echocardiogram is a spatiotemporal video in which clinically relevant information is spread across moving tissue structures. Myocardial enzyme measurements are sparse scalar values whose meaning depends on laboratory reference ranges. Clinical symptoms are textual or categorical descriptions with inherent vagueness. Forcing all of these into a single shared feature space, as conventional fusion methods do, risks destroying the very structure that makes each modality informative. The proposed framework instead processes each modality through its own channel and then reconciles them through fuzzy logic.

The embedding of prior knowledge happens through a clever use of clustering. For each classification task, the model first extracts high-order semantic features from each modality separately. It then constructs a similarity measurement function tailored to the features of that modality, which allows the system to judge how close two patient samples are within that particular data type. On top of this similarity structure, the authors apply a fuzzy dynamic C-means clustering algorithm to select representative sample features from subsets of different categories within each modality. These representative samples serve as the kernel centers of the fuzzy radial basis neuron network. In doing so, the model implicitly encodes prior category knowledge: the kernel centers are not arbitrary learned parameters but meaningful prototypes drawn from actual patient subgroups, and they anchor the network’s reasoning in clinically plausible territory.

This choice of architecture matters most when data is limited. Deep learning models typically demand large, complete datasets, and medical datasets rarely qualify. Patient privacy restrictions, the cost of imaging studies, and the rarity of certain conditions all conspire to keep sample sizes small and feature sets incomplete. By embedding prior feature knowledge from each modality into the network’s mechanism, the new framework imposes both structural and data constraints on the model. Those constraints reduce the requirement for sample completeness, meaning the classifier can still function when some modalities are missing or when the dataset does not cover every possible clinical presentation. The authors also emphasize that the design maintains semantic consistency both between modalities and within each modality, so that the meaning of a feature does not drift as information moves through the network.

Once the fuzzy radial basis layer has processed each modality, the aggregation stage takes over. In the modal category feature aggregation layer, the diverse feature information belonging to each category subclass within each modality is aggregated toward the category level. A key mathematical consequence of this aggregation is that it generates a non-convex class interface. In plain terms, the boundary separating one disease class from another is allowed to bend and wrap in feature space rather than being forced into a simple convex shape. This flexibility is clinically important because disease subtypes frequently overlap and interleave in feature space; a rigid linear or convex boundary would misclassify borderline patients that a non-convex boundary can accommodate.

The reasoning engine that sits above the aggregation layer is built on fuzzy computational logic. Instead of the crisp yes-or-no decisions of classical logic, fuzzy logic allows partial membership: a patient can belong to a diagnostic category to a degree. The authors establish fuzzy inference rules for the classification task using fuzzy multiplication operations, and the final decision is produced by a T-S fuzzy classifier, a well-established framework in which the outputs of individual fuzzy rules are combined into a single overall decision. This layered reasoning structure, from per-modality feature extraction through clustering-based knowledge embedding, aggregation, rule formation, and fuzzy classification, is what allows the system to reconcile heterogeneous data without collapsing their individual meanings.

To test the framework, the researchers applied it to one of the most consequential tasks in clinical medicine: the classification of heart disease. Using multimodal medical data including echocardiograms, multi-lead electrocardiograms, myocardial enzyme examinations, and clinical symptoms, the system performed four types of heart disease classification and achieved an accuracy rate of 83.37 percent. That figure, reported in the paper, serves as the authors’ verification of both the effectiveness and the application value of the method. Four-way classification of cardiac conditions from raw multimodal inputs is a demanding benchmark, since each input type carries partial and noisy evidence, and the result suggests that the fuzzy embedding approach can extract clinically meaningful signal from data sources that are rarely combined in a single automated pipeline.

The implications extend beyond cardiology. The architecture is, in principle, modality agnostic: any collection of data streams with distinct spatiotemporal distributions and semantic structures could be routed through the same multi-channel input, fuzzy radial basis embedding, aggregation, and rule-based classification stages. Domains such as neurology, oncology, and intensive care monitoring, where clinicians routinely juggle imaging, electrophysiology, laboratory panels, and narrative notes, face precisely the fusion and small-sample problems this framework addresses. The emphasis on embedding prior knowledge also aligns with a broader movement in machine learning toward structured, interpretable models that do not require internet-scale datasets to perform well, a movement driven in part by the realities of medical data governance.

There are, of course, caveats worth keeping in view. The reported accuracy comes from the authors’ own evaluation, and the paper describes the published version as one that may still be subject to further editorial refinement. Independent validation on external patient cohorts, comparison against clinical baselines, and prospective testing in real hospital workflows would all be needed before such a system could influence actual diagnostic decisions. The authors declare no commercial conflicts of interest, and the work is published open access under a Creative Commons license, which should make it easy for other research groups to scrutinize and build upon. What the study offers now is a concrete, technically detailed demonstration that fuzzy inference and knowledge embedding can tame the messy plurality of real-world medical data, and a measured step toward diagnostic AI that reasons the way clinicians do: by weighing multiple imperfect sources of evidence together, under uncertainty, and arriving at a graded conclusion.

Subject of Research: A fuzzy radial basis inference network with multimodal feature knowledge embedding for multimodal medical data fusion and heart disease diagnosis

Article Title: A fuzzy inference network with multimodal feature knowledge embedding and its application in medical diagnosis

Article References: Xu, C., Liu, K., Wu, J., Zhang, F., & Hou, Y. (2026). A fuzzy inference network with multimodal feature knowledge embedding and its application in medical diagnosis. Complex & Intelligent Systems. https://doi.org/10.1007/s40747-026-02427-9

Image Credits: AI Generated

DOI: 10.1007/s40747-026-02427-9

Keywords: fuzzy inference network, multimodal feature fusion, fuzzy radial basis neuron network, knowledge embedding, small sample classification, T-S fuzzy classifier, heart disease diagnosis, echocardiogram, electrocardiogram, machine learning, medical artificial intelligence, Complex & Intelligent Systems

The research team, led by Javier Sanmartin and Roberto Elosua of the University of Vic-Central University of Catalonia, followed PRISMA guidelines and prospectively registered their protocol in PROSPERO. They searched PubMed, Scopus, and Web of Science from database inception through September 2023, with an updated search extending to June 2025. From 5,963 initially identified articles, the investigators retained 115 original studies after removing duplicates and screening titles, abstracts, and full texts. Risk of bias was assessed with the Joanna Briggs Institute Critical Appraisal Tool for Cross-Sectional Studies, and pooled estimates were generated using random-effects models with the Freeman-Tukey double arcsine transformation for proportions.

The headline number is 4 percent. Using the International-Seattle criteria, which define T-wave inversion as a negative T-wave of at least 1 millimeter in two contiguous leads, the pooled prevalence among athletes older than 16 years was 4 percent, with a 95 percent confidence interval of 3 to 6 percent. That figure, however, masks enormous variability: individual studies reported prevalences ranging from 0 to 42 percent, and the between-study heterogeneity, quantified by the I-squared statistic, exceeded 80 percent in most meta-analyses. The authors caution that the pooled estimate should be interpreted with this heterogeneity firmly in mind.

Location on the electrocardiogram matters enormously. Inversions were most common in the anterior leads V2 through V4, at roughly 3 percent, while inferior leads showed about 1 percent and lateral leads essentially 0 percent. This anatomical gradient carries clinical weight because lateral T-wave inversion is the pattern most frequently linked to confirmed structural heart disease. In the studies that evaluated cardiomyopathies among athletes with T-wave inversion, the most commonly identified conditions were hypertrophic cardiomyopathy, arrhythmogenic ventricular cardiomyopathy, and dilated cardiomyopathy, and lateral-lead inversions were disproportionately represented among those diagnosed.

Ethnicity emerged as a major modifier of prevalence. Black athletes showed an overall T-wave inversion rate of 13 percent, compared with 3 percent in Caucasian athletes, a difference consistent with known genetic, endocrine, and hemodynamic factors, including greater myocardial wall thickness and left ventricular mass. Notably, recent work cited in the review found that geographical origin within Africa also matters, with repolarisation abnormalities significantly more common among West African and Middle African athletes than among those from East or North Africa. Sex differences were smaller but present: women showed a slightly higher overall prevalence than men, 7 percent versus 5 percent, driven by anterior and inferior leads.

Perhaps the most striking comparative finding is that T-wave inversion is roughly 49 percent more common in athletes than in non-athletes. Across 14 studies that included control populations, the pooled prevalence ratio was 1.49, with a confidence interval of 1.13 to 1.97. The effect was strongest in the inferior leads, where the ratio reached 1.70. This supports the idea that intense training itself, through ventricular remodeling and autonomic changes, produces repolarisation patterns that would be considered abnormal in a sedentary person but are often physiological in an athlete.

The mechanisms behind these training-related inversions are diverse. The authors describe ventricular remodeling effects, including a high papillary muscle to left ventricular mass ratio and lateral displacement of the cardiac apex, which can explain anterior inversions. Structural variants such as left ventricular false tendons have also been implicated. Enhanced vagal tone in trained athletes activates acetylcholine-controlled potassium channels, reducing the action potential gradient between the epicardium and endocardium and altering the repolarisation vector. Myocardial ischemia, by contrast, can invert the repolarisation gradient in a pathological direction, which is one reason the finding cannot simply be dismissed.

Age adds another layer of nuance. In athletes aged 16 or younger, a juvenile T-wave inversion pattern in leads V1 through V3 is considered a normal variant, reflecting right ventricular dominance in early adolescence, and it generally regresses after age 16. The meta-analysis found this juvenile pattern in about 6 percent of young athletes. Interestingly, meta-regression identified age as the only variable that partially explained the heterogeneity between studies, with older athletes showing higher prevalence, a relationship the authors attribute to cumulative lifetime training volume.

On the critical prognostic question, the picture is murkier. Among 34 studies that assessed structural cardiomyopathies in athletes with T-wave inversion, most reported disease prevalence below 1 percent, though nine studies found rates between 1 and 10 percent and six exceeded 10 percent, likely reflecting deeper inversions, lateral predominance, referral-center recruitment, and more sensitive imaging such as cardiac magnetic resonance. Follow-up data were similarly heterogeneous: annual incidence rates of newly diagnosed cardiomyopathy ranged from 0 to 54.9 cases per 1,000 athletes, and major cardiovascular events ranged from 0 to 83.33 per 1,000 athlete-years. The authors attribute this wide spread to small samples, short follow-up, and selection biases, leaving the true event risk statistically uncertain.

The clinical message is a careful balance. T-wave inversion in athletes is more common than in the general population, most often involves the anterior leads, and frequently reflects benign exercise-induced remodeling, particularly in Black athletes and in younger age groups. Yet because it can also herald hypertrophic, arrhythmogenic, or dilated cardiomyopathy, especially when it appears in lateral leads or is accompanied by ST-segment depression, a thorough evaluation remains essential. The authors call for international, standardized, prospective studies with adequate sample sizes, uniform diagnostic protocols incorporating echocardiography, exercise testing, Holter monitoring, and cardiac magnetic resonance, and long-term follow-up to finally pin down which athletes with inverted T-waves truly need to worry.

Subject of Research: Prevalence and prognostic significance of T-wave inversions on electrocardiograms in athletes

Article Title: Frequency and Prognostic Significance of T-wave Inversions in Athletes: A Systematic Review and Meta-analysis

Article References: Sanmartin, J., Cartanya-Bonvehi, J., Valenzuela, H., Carballeira, L., Montserrat, S., Baranchuk, A., & Elosua, R. (2026). Frequency and Prognostic Significance of T-wave Inversions in Athletes: A Systematic Review and Meta-analysis. Sports Medicine – Open, 12(1), Article 147. https://doi.org/10.1186/s40798-026-01115-9

Image Credits: AI Generated

DOI: 10.1186/s40798-026-01115-9

Keywords: T-wave inversion, athletes, electrocardiogram, sudden cardiac death, cardiomyopathy, sports cardiology, meta-analysis, athlete's heart, ECG screening, ventricular remodeling, hypertrophic cardiomyopathy, prevalence

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