Holistic AI in medicine; improved performance and explainability.

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Generative AI Applications

Background:

• Existing AI frameworks like Holistic AI in Medicine (HAIM) fuse multimodal data for clinical tasks but lack explainability.
• Current AI models often function as black boxes, limiting clinical trust and utility.

Purpose of the Study:

• To introduce xHAIM (Explainable HAIM), a novel framework enhancing AI prediction and explainability in medicine.
• To address the limitations of task-agnostic data usage and lack of transparency in previous AI models.

Main Methods:

• xHAIM employs a four-step process: identifying task-relevant data, generating patient summaries, improving predictive modeling, and providing linked clinical explanations.
• The framework leverages Generative AI to create comprehensive patient summaries and link predictions to medical knowledge.

Main Results:

• xHAIM significantly improved average AUC from 79.9% to 91.3% on the HAIM-MIMIC-MM dataset for chest pathology and operative tasks.
• The framework demonstrated enhanced predictive performance and provided traceable, patient-specific explanations.

Conclusions:

• xHAIM transforms AI into an explainable decision support system, bridging AI advancements with clinical utility.
• The framework enables clinicians to understand and trust AI-driven predictions by tracing them to relevant patient data.