Introducing AIDx, The Physician's AI Copilot

Abstract The dynamic environment of medicine, like the Emergency Department, challenges physicians with an influx of patient data and the need for swift diagnosis and treatment under high-pressure situations. Furthermore, there are numerous AI models developed for medical applications that, despite their potential, rarely see hospital implementation because they fail to integrate into pre-existing clinical workflows. Addressing these challenges, I developed AIDx, a state-of-the-art AI-powered system designed to streamline clinical decision-making, provide accurate disease prediction and treatment planning, offer novel approaches for personalized patient healthcare, and serve as a hub for AI-assisted medical analysis. Unlike standalone systems, AIDx integrates seamlessly with a hospital's pre-existing electronic health records and automatically leverages supplemental tools (like an up-to-date medical knowledge database and capability-enhancing AI models), offering physicians a unified platform for comprehensive AI assistance. I trained AIDx on 400,000 de-identified patient charts and evaluated it on 7,000 medical exam questions, achieving a remarkable accuracy of 83.61% that rivals the performance of industry-leading models like Google's Med-PaLM 2 and OpenAI's GPT-4 with significantly fewer parameters (a model size metric), using only 46.7 billion compared to their 340 billion and 1.7 trillion. This parameter efficiency translates into feasible on-premises deployment, high-speed processing, and adaptability across clinical settings. AIDx also significantly enhances patient-physician communication, simplifying medical terminology and facilitating multilingual translation. Ultimately, AIDx represents a transformative integration of AI in medicine, establishing an unprecedented standard for personalized, reliable, and streamlined patient care, elevating the quality of healthcare, and enhancing the lives of patients and physicians alike.