CardioXNet: A Novel Lightweight Deep Learning Framework for Cardiovascular Disease Classification Using Heart Sound Recordings

The alarmingly high mortality rate and increasing global prevalence of cardiovascular diseases signify the crucial need for early detection schemes. Phonocardiogram (PCG) signals have been historically applied in this domain owing to its simplicity cost-effectiveness. In paper, we propose CardioXNet, a novel lightweight end-to-end CRNN architecture automatic five classes cardiac auscultation namely normal, aortic stenosis, mitral regurgitation valve prolapse using raw PCG signal. process has automated by involvement two learning phases. Three parallel CNN pathways implemented representation phase learn coarse fine-grained features from explore salient variable receptive fields involving 2D-CNN based squeeze-expansion. Thus, phase, network extracts efficient time-invariant converges with great rapidity. sequential residual bidirectional-LSTMs skip connection, can proficiently extract temporal without performing any feature extraction on obtained results demonstrate that proposed yields outstanding performance all evaluation metrics compared previous state-of-the-art methods up 99.60% accuracy, 99.56% precision, 99.52% recall 99.68% F1- score an average while being computationally comparable. This model outperforms works same dataset considerable margin. accuracy both primary secondary combined significantly low number parameters prediction approach makes suitable point care CVD screening resource setups memory constraint mobile devices.