PCG Heart Sound Dataset: 3,126 Recordings for Cardiovascular Detection

3,126 phonocardiogram recordings developed for the PhysioNet/Computing in Cardiology Challenge 2016. The dataset was created to classify heart sounds as normal or abnormal and serves as a benchmark for automated analysis. It includes recordings from multiple sources in clinical and non-clinical environments, featuring both healthy subjects and patients with cardiac pathology.

Use Cases

• Binary classification of heart sounds as normal or abnormal based on audio recordings
• Benchmarking machine learning models for automated phonocardiogram analysis
• Training audio-based diagnostic tools using data from clinical and non-clinical environments
• Studying acoustic patterns associated with cardiac pathology

Strengths

• 3,126 heart sound recordings provide a substantial corpus for training
• Serves as a widely recognized benchmark dataset for a specific research challenge
• Includes data from both healthy subjects and patients with cardiac pathology

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

PhysioNet/Computing in Cardiology Challenge 2016

Collection Method

Collected from several contributing sources across clinical and non-clinical environments