PaReGTA: A Temporally Aware LLM-Based Patient Representation Framework for EHR Analytics

PaReGTA is an LLM-based system that converts long medical records into structured text and learns patient representations. It uses domain adaptation to refine these texts and a hybrid pooling method to combine recent and important visits, resulting in fixed-dimensional, time-aware patient profiles for EHR analysis.

Open the full episode

More from AI Radio