Key result
Explainable AI ECG model identifies LQTS mutation carriers with ~0.93 AUC.
Why the study?
Incomplete penetrance of typical LQTS-related mutations makes diagnosis difficult, and neither models capturing morphological ECG alterations nor black-box AI models have been successfully translated into clinical practice. This study investigated the feasibility of an explainable AI approach based on the Frequency Modulated Möbius method for LQTS diagnosis.
Does an explainable artificial intelligence model based on the Frequency Modulated Möbius (FMM) method accurately diagnose genetically confirmed Long QT Syndrome from ECG beats in mutation carriers compared to healthy controls?
Case-Control (n=300)
Does an explainable artificial intelligence model based on the Frequency Modulated Möbius (FMM) method accurately diagnose genetically confirmed Long QT Syndrome from ECG beats in mutation carriers compared to healthy controls?
An explainable AI model utilizing 3DFMMecg morphological features can reliably distinguish LQTS mutation carriers from healthy controls, even when the baseline QT interval is normal (<460 ms).
May aid LQTS diagnosis beyond QTc in incomplete penetrance; leaves open prospective validation before clinical use.
Background Long QT Syndrome (LQTS) is an inherited arrhythmia syndrome characterized by increased susceptibility to potentially lethal arrhythmias1. While prolonged QT interval is diagnostic, typical LQTS-related mutations exhibit incomplete penetrance2 rendering diagnosis difficult. However, the duration of ventricular repolarization represents just one aspect of the electrocardiographic (ECG) manifestations of LQTS, which extend to morphological alterations such as T wave asymmetry, shape and phase. Unfortunately, neither models aiming to objectively capture these morphological alterations nor black-box artificial intelligence (AI) models have not been successfully translated to the clinical practice. In this context, the Frequency Modulated Möbius (FMM) model3 offers a physiologically meaningful framework with strong statistical and computational properties, as unlike black-box models, FMM balances interpretability with predictive accuracy. Purpose To investigate the feasibility of an explainable artificial intelligence approach based on FMM method for the diagnosis of Long QT Syndrome. Methods We enrolled consecutive individuals who tested positive for a single mutation in one of the three genes associated with LQTS. Family members of a proband with a known LQTS-related gene mutation who had been referred for genetic testing and who tested negative mutation identified in the family served as healthy controls. ECGs were digitized using ECGMiner. Following preprocessing and signal splitting into individual beats per patient, 3DFMMecg was fitted to each beat to extract beat-level markers. To test the predictive ability of this approach, a logistic regression model was used. Results We enrolled n=300 individuals: n=200 carriers of a typical LQTS mutation (n=100 in KCNQ1, n=71 in KCNH2, and n=29 in SCN5A), and n=100 individuals were healthy controls. First, we developed a logistic regression model to identify several electrocardiographic markers that contribute meaningfully to the identification of patients affected by genetically confirmed Long QT Syndrome. Importantly, in addition to the duration of the QTc interval (p<0.0001), delayed onset and peak of the T wave (p=0.011 and p<0.001, respectively), less symmetrical T waves (p<0.0001) as well as longer duration of the ST-T segment (p=0.0001). The model showed a strong discrimination ability (cross-validated mean AUC of 0.925±0.023). We then tested the performance of the model in the subset of individuals with normal QT interval (i.e., <460 ms). Crucially, the same predictors were retained with comparable effect sizes and the model’s ability to identify genotype-positive individuals remained robust (cross-validated mean AUC of 0.886). Conclusions Novel explainable artificial intelligence model reliably distinguished carriers of LQTS-causing variants from healthy controls based on coherent morphological features.
No takes yet. Share an insight, caveat, or question.
Kukavica et al. (2026) conducted a case-control in Long QT Syndrome (n=300). Explainable AI model based on 3D Frequency Modulated Möbius (3DFMMecg) method vs. Genotype-negative healthy family member controls was evaluated on Identification of genetically confirmed Long QT Syndrome (AUC 0.925±0.023). An explainable AI model based on FMM ECG features strongly discriminated LQTS mutation carriers from healthy controls (AUC 0.925±0.023), including those with normal QT intervals (AUC 0.886).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: