biosqi

Explainable signal-quality assessment for biomedical recordings.

biosqi divides ECG, PPG, or EEG recordings into windows, computes transparent signal-quality indicators (SQIs), and returns:

• a calibrated 0–100 usability score for each window;
• a boolean mask suitable for downstream processing;
• evidence for likely artifacts such as clipping, flatline, baseline wander, mains interference, and poor periodic consistency;
• machine-readable JSON/CSV and a local interactive dashboard.

Scope and safety

biosqi measures properties of a recording. It does not diagnose a patient, determine whether a physiological pattern is normal, or replace review by a qualified clinician. The bundled model is validated on the datasets stated in Validation; performance may not transfer to a different sensor, population, acquisition protocol, or artifact distribution.

Design principles

1. Evidence before labels. Every quality decision exposes the indicator values that contributed to it.
2. Calibration before confidence. A score is presented as a probability only when its calibration procedure and validation population are stated.
3. No hidden preprocessing. Sampling rate, profile, window length, overlap, threshold, and model version are included in every report.
4. Reproducible claims. Published metrics are generated by benchmarks/run_all.py from public datasets.

Continue with the Quickstart.