Multimodal Large Language Models for Cystoscopic Image Interpretation and Bladder Lesion Classification: Comparative Study

Key Points

• Cystoscopy image interpretation with multimodal large language models supports biopsy classification effectively, yet shows limited performance in challenging cases.
• Key performance metrics indicate that while the assistive potential is meaningful, optimization is necessary before clinical use.
• Observational analysis employing multimodal large language models for generating text rationales and facilitating diagnosis in cystoscopy shows promise.
• Further optimization is needed for these models to safely integrate into clinical workflows, particularly for complex differential diagnoses.