Date of Submission
2026-06-22
Supervisor
Dr. Tariq Mahmood, Professor, School of Mathematics and Computer Science (SMCS)
Committee Member 1
Dr. Sajjad Haider, Examiner –I, Institute of Business Administration (IBA), Karachi
Committee Member 2
Dr. Rizwan Ahmed Khan, Examiner – II, Institute of Business Administration (IBA), Karachi
Degree
Master of Science in Computer Science
Department
Department of Computer Science
School
School of Mathematics and Computer Science (SMCS)
Keywords
Rheumatic Heart Disease, Deep Learning, Autoencoder, Electrocardiogram, Phonocardiogram, Cross-modal Synchronisation, Paediatric Screening
Abstract
Rheumatic Heart Disease (RHD) is a chronic and progressive condition caused by repeated episodes of acute rheumatic fever following untreated streptococcal infections, and remains one of the leading causes of preventable cardiovascular mortality in low-andmiddle-income countries. Thediseaseprogressessilentlyinschool-agedchildrenlongbeforeclinicalsymp toms emerge, yet echocardiography the current gold standard for early detection, remains prohibitively expensive and largely unavailable in the settings where the burden of RHD is greatest. This thesis proposes a deep learning-based framework for the detection of subclinical RHD in children using synchronised Electrocardiogram (ECG) and Phonocardiogram (PCG) signals recorded through a portable digital stethoscope. ECG and PCG capture complementary electrical and mechanical aspects of cardiac activity, which are inherently linked within each heartbeat. However, most existing approaches treat them independently, motivating the need for cross-modal synchronisation The proposed framework employs two independent vanilla autoencoders to learn compact latent representations from ECG and PCG signals separately. These bottleneck features are subsequently synchronised through a cross-modal mapping stage with cycle consistency constraints, enabling the framework to capture the physiological relationship between the electrical and acoustic signals of the heart within the same cardiac event. The synchronised representations are combined with demographic variables and passed to a machine learning classifier for binary RHD classification. The dataset was collected in collaboration with the Sindh Institute of Urology and Transplantation (SIUT), Karachi comprising paired ECG, PCG, and demographic data from 612 school-aged children between 5 and 15 years of age, with echocardiographic confirmation as the reference standard. Cross-modal synchro nisation achieved a mean cosine similarity of 0.731 and a Pearson correlation of 0.924 in the ECG-to-PCG direction. Under Leave-One-Positive-Out (LOPO) cross-validation with ECG and PCG fusion, and using SMOTE oversampling to address the 101:1 class imbalance, Support Vector Machine (SVM) achieved the best overall performance with an AUROC of 0.851 and a recall of 0.667 for the RHD positive class. External validation on an independent dataset from Tabba Heart Institute, Karachi comprising 21 participants including 8 confirmed RHDcases, yielded correct identification of 5 out of 8 positive cases, recall 62.5%. This research introduces the first cross-modal autoencoder synchronisation framework applied specifically to subclinical RHD detection in children, provides a real-world paired ECG and PCG dataset from 612 school-aged children confirmed by echocardiography.
Document Type
Restricted Access
Submission Type
Thesis
Recommended Citation
Kamil, M. (2026). Deep Learning-Based ECG-PCG Synchronisation for Subclinical Rheumatic Heart Disease Detection in School-Aged Children (Unpublished Unpublished graduate thesis). Retrieved from https://ir.iba.edu.pk/etd-ms-cs/4
