Contributor name and NetID: Vishal Vyas vyas9 (vyas9@illinois.edu)
Type of contribution: Standalone Task
Original Paper: A Data-Centric Approach To Generate Faithful and High Quality
Patient Summaries with Large Language Models
Link to Original Paper: [https://arxiv.org/abs/2402.15422]
High-level description of implementation: Implements a new standalone task for extracting the patient note samples using the MIMIC - IV Note dataset. This dataset is then cleaned further and used to train Large Language Models to generate patient summaries and detect hallucinations.

File guide listing

New Files
pyhealth/tasks/discharge_note_summarization.py A new Task for extracting patient discharge summary and text samples from MIMIC4-Note dataset. Task also performs some additional processing on the patient discharge events to extract brief hospital course, text , subject id and hadm id.

tests/core/test_discharge_note_summarization.py: Test cases for the discharge_note_summarization.py file

docs/api/tasks/pyhealth.tasks.DischargeNoteSummarization.rst
examples/discharge__summary_samples.ipynb: Note book showing examples of how the Task is used to extract patient summaries and additional cleaning is performed to generate the data set for training LLMs - GPT 4 and Llama 70B

Modified Files
docs/api/tasks.rst: Updated the index
pyhealth/tasks/init.py Registered the Task