Development of an agentic AI for processing electronic health records for stroke rehabilitation

"Development of an agentic AI for processing electronic health records for stroke rehabilitation" Stroke rehabilitation represents a critical phase in patient care, where timely access to reliable and personalized information can significantly influence clinical outcomes. In most modern healthcare systems, patient information is stored in electronic health records (EHRs), which clinicians rely on as the primary source for medical history, monitoring progress, and planning treatments. However, these records are often heterogeneous, unstructured, and difficult to process automatically, creating barriers to their effective use in supporting individualized rehabilitation plans. This thesis addresses this challenge by proposing the development of an agentic artificial intelligence (AI) system specifically designed for the extraction, structuring, and summarization of clinical information from EHRs. The main objective of this work is to design and implement a framework capable of transforming raw medical records, structured or unstructured, into usable outputs for healthcare professionals. The system provides two key results: on one hand, a personalized care plan tailored to the patient’s clinical history and rehabilitation needs; on the other, an overview of the patient’s data that facilitates decision-making and progress monitoring. Through the use of designed AI agents, the model is able to extract the most relevant information, generate meaningful summaries, and preserve data quality. The results show that the developed system produces unique and contextually relevant outputs from different inputs. For each EHR processed, the agents successfully extract core medical information and generate structured summaries that retain clinical meaning. Furthermore, the solution demonstrates a favourable balance between the quality of the results and the operational costs. In conclusion, this work demonstrates that agentic AI represents a promising paradigm for the medical domain, and particularly for stroke rehabilitation. The proposed system illustrates how an autonomous agent can bridge the gap between complex, unstructured clinical data and the practical needs of healthcare professionals, while more broadly highlighting the potential of AI to transform healthcare processes through scalable, accurate, and personalized data-driven support.