Most PHI datasets evaluate masking on static single-modality documents. This one is different.<\/p>\n
It captures per-event masking decisions across a simulated longitudinal stream, the same subject appearing across clinical notes, ASR transcripts, imaging proxies, waveform data, and audio metadata over time. The idea is to evaluate how re-identification risk accumulates across events rather than within a single record.<\/p>\n
Five policies are included for comparison: raw, weak, pseudo, redact, and adaptive. The adaptive controller is the interesting one, it escalates masking strength only when cumulative exposure actually justifies it.<\/p>\n
Dataset is fully open, no DUA required. Everything runs on synthetic data, no real patient records anywhere.<\/p>\n
Hugging Face: https:\/\/huggingface.co\/datasets\/vkatg\/streaming-phi-deidentification-benchmark<\/a><\/p>\n Code to regenerate: https:\/\/github.com\/azithteja91\/phi-exposure-guard<\/a><\/p>\n Happy to answer questions on the schema or the benchmark design.<\/p>\n<\/div>\n submitted by \/u\/Visual_Music_4833 <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
