RWT-Net: A Hybrid ResNet-wavelet-transformer for Early Detection of Left Ventricular Hypertrophy

Abstract

Early detection of left ventricular hypertrophy (LVH), a key predictor of heart failure and stroke, is critical. However, standard 12-lead electrocardiogram (ECG) criteria suffer from low sensitivity. While deep learning shows promise, a research gap exists for models that robustly integrate diverse signal fea tures to improve detection, especially sensitivity. We propose ResNet-wavelet-transformer net (RWT-Net), a hybrid architecture that fuses deep morphological features from a ResNet1D with statistical time-frequency features from awavelet packet transform (WPT) using a transformer encoder. The model was evaluated on the PTB-XL dataset (11,201 recordings) using a stringent, patient level 5-fold cross-validation. RWT-Net achieved a mean area under the curve (AUC)of0.9868andF1-scoreof0.8725. Critically, its wavelet-enhanced stream yielded significantly higher sensitivity compared to a ResNet-transformer base line (0.8964 vs. 0.8716, p=0.0039), better addressing the clinical need to mini mize false negatives. A key limitation is the reliance on ECG-based labels, not an echocardiography gold standard. RWT-Net demonstrates potential as a reliable, automated screening tool to prioritize at-risk patients for further clinical assessment.