Do General-Purpose Multimodal Large Language Models Really See Radiologic Images or Rely on Text?
Abstract
understanding of radiologic images.While these models may perform well on standardized examination-style diagnostic questions, their performance appears to depend primarily on textual processing rather than genuine visual interpretation.This editorial aims to comprehensively understand current evidence on multimodal LLMs' performance in radiology, examine whether they truly interpret radiologic images, highlight their limitations in real-world clinical applications, and outline expectations for future development.
Fields of Study & Subjects
Radiology practices and education
Artificial Intelligence in Healthcare and Education
Topic Modeling
Publication details
- Journal / source
- Korean Journal of Radiology
- Publisher
- XMLink
- Volume
- 27
- Issue
- 4
- Pages
- 297–297
- Publication date
- 1 January 2026
- Language
- EN