Multimodal AI for Early Disease Detection

Early-stage medical screening can be supported effectively through intelligent digital systems that combine multiple AI techniques.This paper presents a multimodal healthcare support framework that applies machine learning, computer vision, and optical character recognition to assist users during preliminary health assessment.The system performs symptom-based disease prediction, detects facial acne using a deep learning object detection approach, and automatically extracts clinically relevant information from uploaded medical reports.To improve accessibility and usability, the platform includes multilingual voice-based interaction, structured doctor-patient workflow management, and secure connectivity with external services through OAuth-based authorization.From an architectural standpoint, the backend follows a dual-service design in which the core web application operates independently from a GPU-enabled vision sidecar, improving scalability, reliability, and deployment flexibility.Experimental evaluation demonstrates strong system performance, with ensemblebased disease classification achieving an accuracy of approximately 97.6% and the acne detection module reaching an mAP50 score of 0.645 on a custom dataset.Although the system is not intended to replace professional medical diagnosis, it serves as a supportive tool for early screening and workflow automation.