Automated Medical Report Generation using Large Language Models

Primary supervisor

Trang Vu

Co-supervisors

  • Thanh Nguyen (NMHS)

Manual medical report writing is time-consuming and subject to variability. Recent advances in large language models (LLMs) create new opportunities for automating this process. This project explores using LLMs to generate medical reports from a very large dataset, aiming to streamline workflows and support clinical decision-making. Students will work on data preprocessing, model fine-tuning, and performance evaluation, contributing to advances in medical AI.

 

Student cohort

Double Semester

Aim/outline

  • Develop data construction and preprocessing pipeline
  • Fine-tune and evaluate an LLM for medical report generation
  • Compare generated reports with expert annotations
  • Investigate challenges and propose improvements
  • An open source code  and possibly a publication in medical AI/NLP venues

URLs/references

  1. Varol Arısoy, M., Arısoy, A. & Uysal, İ. A vision attention driven Language framework for medical report generation. Sci Rep 15, 10704 (2025)  https://www.nature.com/articles/s41598-025-95666-8
  2. Chen, Qi, et al. "A survey of medical vision-and-language applications and their techniques." arXiv preprint arXiv:2411.12195 (2024). https://arxiv.org/abs/2411.12195

 

Required knowledge

  • Good Python programming
  • Interest in AI/ML with prior deep learning experience
  • Familiarity with medical imaging is a plus but not required