GenAI for automated medical reports: A retrieval-augmented framework

This research uses the opportunity of generative artificial intelligence (GenAI) and will solve the problem of effective and precise medical records. The healthcare systems are characterized by numerous growing administrative and clinician demands, and automating medical report generation can be an acceptable solution. This paper presents the use of instruction-tuned, Transformer based large language models (LLM) fine-tuned using supervised learning and expertise feedback-based reinforcement learning (RLEF) to produce realistic and context-driven medical reports. In the patient data and clinical guidelines, a retrieval-augmented generation (RAG) module is inserted into the text generated; it makes sure that the text is factually correct and useful in clinical practice. The proposed methodology is staged in nature, including data curation, model development, rigorous experimental testing, and pilot implementation in a hospital. The key contribution of this study is the presentation of a new model in the creation of medical reports. The overall evaluation of the LLM performance and the comprehension of the ethical aspects of AI-driven health care documentation have been conducted with the use of the automated metrics and the review of the clinicians. The proposed and developed model (LLM + RAG + RLEF) obtained 94% accuracy in contrast to the conventional method. The first limitation of proposed method is patient safety, secondly, the utilization of de-identified electronic health records (EHRs); and thirdly, the tracking of the possible biases. The research directions in future include investigation of the extensibility of the suggested approach to numerous medical specialties and the possibility of combining various sources of multimodal data to increase accuracy of medical reports.