Full write-up coming soon — detailed content for this project is being finalized.
Overview
Heart murmurs are often the earliest detectable sign of structural heart disease, but reliably hearing one takes training most primary care settings don't have on hand. This project classifies heart sounds recorded from a low-cost digital stethoscope, flagging likely murmurs for follow-up.
Approach
Recordings are converted to a time-frequency representation and passed through a small convolutional model trained to distinguish common murmur patterns from normal heart sounds and background noise.
Results
The model reliably separated clear murmurs from normal sounds and performed well even on noisy recordings collected outside a controlled clinical setting.