Title : Agentic AI framework for End-to-End Medical Data Inference
Author : G.Prabhakar, N.Savitha, B.Yugandhara Chary
Abstract :
Due to fragmented preprocessing procedures, model compatibility challenges, and tight data privacy limitations, building and implementing machine learning solutions in healthcare remains costly and labor-intensive. Through a system of task-specific agents, we automate the full clinical data pipeline in this study, from intake to inference. This architecture is introduced as an Agentic AI system. No human interaction is required for feature selection, model selection, or preprocessing recommendation since these agents can handle both structured and unstructured data. Using geriatric, palliative care, and colonoscopy imaging datasets that are publicly accessible, we assess the system's performance. As an example, the pipeline starts with file-type detection by the "Ingestion Identifier Agent" to ensure privacy compliance. Then, the "Data Anonymizer Agent" identifies the type of data and anonymizes it. This process is repeated for both structured and unstructured data, such as data from c
