pone.0347840.t004: Photoplethysmography Signals for Multi-Class Cardiovascular Diagnostics

Amjed Al Fahoum published a dataset on May 15, 2026, containing photoplethysmography (PPG) signals for cardiovascular disease classification. The dataset comprises 2,448 annotated PPG segments from 612 patients across six diagnostic classes: atrial fibrillation, heart failure, acute coronary syndrome, cerebral vascular accident, deep vein thrombosis, and normal sinus rhythm. It was used to train a hierarchical convolutional neural network achieving 93.48% overall accuracy.

Use Cases

• Training multi-class cardiovascular disease classifiers based on raw PPG signal morphology and rhythm.
• Benchmarking deep learning models for photoplethysmography signal analysis based on the six diagnostic classes.
• Developing real-time health monitoring algorithms for wearables based on the sub-5ms inference time demonstrated.
• Studying the confusion between physiologically related conditions like ACS and CVA based on the model's error patterns.

Strengths

• Dataset contains 2,448 annotated PPG segments from 612 patients.
• Model trained on the data achieved an overall accuracy of 93.48% and a Cohen’s Kappa of 0.8968.
• Specific performance metrics are reported, including flawless precision and recall for AF and HF detection.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description notes a limitation: the need for external, multi-center validation.

Provenance

Source

Amjed Al Fahoum

Collection Method

Multi-source dataset of PPG signals annotated across six diagnostic classes.

Freshness

Last updated 2026-05-15 17:38:41; freshness should be verified.