000			02557nam a22001577a 4500
082			_a610
100			_aMansoor, Mariam _9133095
245			_aAn Intelligent Framework for Real-Time PCG Classification and Cardiac Monitoring Using Machine Learning / _cMariam Mansoor
264			_aIslamabad : _bSMME- NUST; _c2026.
300			_a103p. _bSoft Copy _c30cm
500			_aThe accurate and timely diagnosis of cardiovascular diseases, particularly valvular heart disorders, is essential for reducing morbidity and mortality. However, conventional auscultation remains limited by subjectivity, ambient noise, and an inability to record or analyse sounds. This thesis addresses these limitations through a two-fold contribution: the development of a wearable, high-fidelity digital auscultation system and the integration of a real-time, AI-assisted classification framework for automated diagnosis. The first part of the research presents a custom-designed wearable digital stethoscope using a piezoelectric contact sensor embedded in an aluminium chestpiece. This configuration offers superior acoustic coupling and environmental noise rejection. A multi-stage low-noise analog signal chain was designed, incorporating band-specific filtering to isolate heart and lung signals. A miniaturized PCB was fabricated to house the circuit, and a Python-based GUI was developed to visualize, record, and archive phonocardiogram (PCG) signals while generating patient-linked reports. In the second phase, a lightweight 1D convolutional neural network (CardioSynx) was introduced for real-time classification of five PCG classes, including normal and pathological valve conditions. The model trained offline using Butterworth filtering, Hilbert envelope segmentation, and wavelet denoising, achieved 97.5% ± 0.22 accuracy on validation data. For deployment, the system employed a parallel processing architecture to enable simultaneous audio acquisition, denoising, feature extraction, and inference, achieving a frame-level latency of 58 ms, well below clinical standards and an SNR of 16.8 dB. Real-time evaluation on patients yielded over 90% accuracy across key pathologies. Together, the proposed system delivers a scalable, real-time, and clinically robust platform for digital auscultation and decision support in modern healthcare.
650			_aMS Biomedical Engineering (BME) _9119509
700			_aSupervisor: Dr Muhammad Asim Waris _9119524
856			_uhttp://10.250.8.41:8080/xmlui/handle/123456789/57337
942			_2ddc _cTHE
999			_c616088 _d616088