レポートからのキースライス自動選択

This research explores the development of an intelligent system that identifies diagnostically relevant 2D slice(s) from 3D CT volumes based on radiology reports. By fine-tuning multi-modal large language models (MLLMs) to jointly interpret report impressions and corresponding images, we aim to bridge textual and visual clinical information.

This approach reduces the need for manual slice selection and accelerates the process of locating key diagnostic views, ultimately supporting radiologists in efficient and accurate clinical decision-making.