Enhancing ECG Classification through Individual 12-lead Analysis and fQRS pattern based on Deep Learning technique

Diabetes patients can experience congestive heart failure due to diastolic dysfunction from its early phases. This study addresses a crucial challenge by proposing a novel methodology for predicting this complication. We introduce an innovative approach by examining the Fragmented QRS (fQRS) pattern within the individual leads of the 12-lead Electrocardiogram (ECG) data. Our approach employs a specialized deep learning model, ResNet1D, trained on each lead separately. The outcomes from these leads are then combined using a Random Forest classifier for more accurate predictions. The experiment results show that our proposed model outperforms other models utilizing combined 12-lead ECG signals. Each lead encapsulates specific and pivotal diagnostic insights, enhancing the model’s performance in predicting ECG signal anomalies.