phonocardiogram

This project builds a deep-learning-based heartbeat sound classification system using MFCC features and multiple models including CNN, BiLSTM, and a Hybrid CNN–BiLSTM architecture. The system detects and classifies heart sounds into normal, murmur, and artifact categories, supporting early cardiac abnormality detection.

Non-Invasive Valvular Disorder Screening from Phonocardiograms. Upload any digital stethoscope PCG recording (WAV, MP3, M4A, FLAC, OGG) or select a clinical reference sample to analyze cardiac cycles for murmurs across 5 pathology classes.

Research prototype for screening abnormal heart-sound recordings with audio features and classical machine learning. Includes a local WAV demo, evaluation code, and research results. Not for clinical diagnosis.

A Streamlit app for CWT & FFT signal analysis with PCG feature detection, built entirely from scratch.

Optimization of CNN, GRU, and CNN-GRU models for compressed PCG classification with deployment on STMicroelectronics boards, including the Neural-ART accelerator.

The main objective of this project is to design and implement an automatic classification method, CNN, to classify cardiovascular diseases (CVDs). Diseases covered include Mitral Stenosis (MS), Aortic Stenosis (AS), Mitral Regurgitation (MR), Mitral Valve Prolapse (MVP), along with normal heart sound.

Phonocardiogram (PCG) signal processing and analysis for Internet of Medical Things (IoMT) applications, including filtering, heart sound visualization, and feature extraction.

Open-source AI heart murmur detection for primary care: multi-location masked-attention fusion, official PhysioNet-2022 challenge metric (s_murmur 0.70, ~median rank 20/40)

FPGA design for extracting PSD, Hilbert, Wavelet, and Homomorphic envelograms from phonocardiogram (PCG) recordings.

Fonogramma: auscultazione digitale (PWA) per fonendoscopio via jack - fonocardiogramma, waterfall, analisi S1-S4, soffi, crepitii, visite e referto

Myocardial infarction detection from phonocardiogram (PCG) heart sounds using a hybrid CNN-LSTM model, with a Flask web interface. 560 recordings from 140 subjects, collected at Hasan Sadikin Hospital, Indonesia.

Exploratory computational framework for modeling cardiac acoustic signals as latent physiological topology using manifold learning, trajectory dynamics, anomaly detection, and cyclic state-space analysis.

Smart Digital Stethoscope - EEE495/EEE496 senior design project for PCG acquisition, embedded streaming, signal processing and phantom-based characterisation. Repository codename: AuscultaForge.

Heart Sound Segmenter implemented using a Convolutional Neural Network (CNN) on heterogeneus platforms (RISC-V included, through AIRISC architecture)

Non-proprietary code and dummy workflow for AI-enhanced auscultation to identify left ventricular dysfunction.

A toolkit for BLE digital stethoscopes — capture, DSP, log-mel spectrograms, and murmur classification on the Apple Neural Engine

Final year capstone project for heart disease detection using a multi-modal approach. We trained separate ML models on ECG, PCG, and PPG signals and fused their confidence scores using a mathematical method to provide a more accurate and holistic risk assessment. Project Members: Anandakrishnan A and Aryan Matte

Listen2YourHeart is a publically available, extendable framework for training Neural Networks via Contrastive SSL learning, for Phonocardiogram classification.

Public repository associated with: "honocardiogram classification using 1-dimensional inception time convolutional neural network"