Applications of Single-Modal and Multi-Modal Models in Medical Diagnosis and Rare Disease Diagnosis

Abstract

Artificial Intelligence (AI), as a multifaceted tool, has gradually developed into a tool that can assist the medical field. This article discusses AI from two aspects: unimodal and multimodal data diagnosis. Unimodal diagnosis is the starting point and foundation of AI in medical diagnosis, using supervised deep learning models for end-to-end anomaly detection or classification based on specific, structured data types. Multimodal diagnosis constructs multimodal fusion networks to integrate heterogeneous data, achieving comprehensive disease assessment, differential diagnosis, and prognostic prediction. This article demonstrates the current advantages and disadvantages of AI models as diagnostic tools, such as the advantages of efficient diagnosis and comprehensive assessment of rare diseases, as well as the many hidden dangers such as insufficient datasets, limitations of dark knowledge, and privacy risks. This study underscores the necessity of improving data quality, interpretability, and ethical regulation to facilitate the reliable clinical adoption of AI-based diagnostic systems.