Machine learning for classification of advanced rheumatic heart disease using electrocardiogram in cardiology ward

Background Rheumatic heart disease (RHD) arises from untreated streptococcal throat infections caused by beta-hemolytic group A streptococci, leading to progressive damage to cardiac valves. While echocardiography is the gold standard for RHD diagnosis, its use in low-income countries is limited due to scarce resources and a lack of trained professionals. Automated RHD detection via echocardiography and phonocardiography data has shown promising, but the effectiveness of electrocardiogram (ECG) for detecting RHD in endemic regions at cardiac wards with limited resource remains uncertain. This study explores the viability of ECG as a cost-effective tool for RHD detection in cardiac wards. Methods The study utilizes a dataset comprising single-lead ECG recordings from 124 confirmed RHD patients and 46 healthy controls collected at a major referral hospital in Ethiopia. Additionally, an extended-RHDECG dataset, which consists of age-matched ECGs from the Physikalisch-Technische Bundesanstalt (PTB-XL) dataset and RHD ECGs, was utilized. The single lead ECG segments of 10-second duration were resampled at 250Hz. Temporal and relative wavelet energy (RWE) features combined with Convolutional Neural Network (CNN) model features were employed for classification of prevalent cardiovascular diseases in the context of the Global South. Results A 5-folds cross-validation on RHDECG dataset using CNN model showed an accuracy of 88.6  ±  0.2% in detecting RHD from healthy controls. This result was improved with combined features of CNN, temporal and RWE to an accuracy of 94.6  ±  0.1%. Similar evaluation on extended-RHDECG yielded an accuracy of 51.2  ±  0.3%, which reached to 69.1  ±  0.5% considering the top-2 predictions of the six classes. The RWE and mainly the PR interval were the most relevant features for classification. Moreover, the RWE concentrations at higher frequency bands were significantly greater in the RHD group than in the healthy control group. Conclusion The findings suggest that single-lead ECG in combination with machine learning could serve as a potential tool for automatically classifying patients with advanced RHD. ECG time‒frequency features effectively capture abnormalities associated with RHD, enhancing the performance and generalizability of deep CNN models. This approach offers a cost-effective tool that might be useful in addressing the RHD burden at resource-constrained cardiac clinics in alignment with global health development goals.