Detection of Titin-Associated Electrocardiography Features in Dilated Cardiomyopathy Using Conventional and Deep...

Area of Science:

• Cardiology
• Genetics
• Artificial Intelligence in Medicine

Background:

• Titin truncating variants (TTNtvs) are the primary genetic cause of dilated cardiomyopathy (DCM).
• Routine genetic testing for TTNtvs is limited by resource constraints, hindering diagnosis and treatment.
• Identifying predictive markers for TTNtv is crucial for efficient patient stratification.

Purpose of the Study:

• To identify electrocardiography (ECG) parameters that predict TTNtv in DCM patients.
• To compare the predictive performance of conventional ECG analysis with an ECG-based deep neural network (DNN).
• To determine which patients would benefit most from targeted genetic testing for TTNtv.

Main Methods:

• A retrospective multinational study involving 99 DCM patients with TTNtv and 318 gene-elusive DCM patients.
• Extraction of conventional ECG parameters, such as QRS duration.
• Development and training of a DNN to analyze ECGs and extract 21 explainable factors.
• Comparison of predictive model performance using C-statistics, with LASSO regularization for variable selection.

Main Results:

• TTNtv patients were younger, more frequently male, and had lower ejection fraction compared to gene-elusive DCM patients.
• Conventional ECG showed shorter QRS duration and prolonged PR interval associated with TTNtv.
• The DNN identified specific ECG patterns, including T-wave inversions, linked to TTNtv.
• Both conventional ECG (C-statistic=0.83) and DNN (C-statistic=0.86) models demonstrated strong predictive performance for TTNtv.

Conclusions:

• Conventional ECG and DNN analysis show comparable, high predictive accuracy for identifying TTNtv in DCM.
• These ECG-based approaches can serve as valuable clinical tools to guide targeted genetic testing.
• Efficient identification of TTNtv patients can improve access to genetic diagnostics and personalized care.