MediAI - AI Powered Multilingual Clinical Decision Support System (CDSS)

Abstract

Patient data almost never shows up in one neat package. A single visit can leave behind structured vitals and symptom checklists, a scanned or handwritten lab slip, a medical image, and a handful of spoken complaints, and most AI health tools out there are only really built to deal with one of these at a time. This paper presents MediAI, a modular clinical decision support system (CDSS) that pulls together machine-learning-based disease-risk prediction, OCR-driven laboratory report interpretation, CNN-based medical image classification, an explainability layer built on feature attribution, a constrained large-language-model (LLM) explanation component, and multilingual text and voice interaction in English, Hindi, and Kannada, all inside one clinician-supervised workflow. Instead of letting a general-purpose LLM reason its way toward a diagnosis, we keep the language model on a short leash here: its only job is to translate the outputs of specialized, independently validated models into plain language, and every AI-generated result has to pass a human review step before anyone treats it as a confirmed clinical record. The paper lays out the proposed architecture, situates it against related work in disease-risk modelling, medical OCR, CNN-based imaging, explainable AI, and multilingual clinical speech recognition, and sets out the experimental protocol needed to validate each module.