Interpretable Brain MRI Report Generation Anchored by Lesion Topography.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Increasing radiologist workloads necessitate efficient and accurate brain MRI report generation.
• Current methods lack fine-grained, explainable reporting capabilities.

Purpose of the Study:

• To develop and evaluate a novel system for grounded automatic brain MRI report generation.
• To introduce a benchmark dataset and a framework to support explainable AI in radiology.

Main Methods:

• Release of RadGenome-Brain MRI dataset with multi-modal scans and expert annotations.
• Proposal of AutoRG-Brain framework combining anomaly segmentation and visual prompting language models.
• Extensive quantitative and expert evaluations in clinical settings.

Main Results:

• The system significantly enhances junior radiologists' ability to detect subtle abnormalities.
• Improved quality and structure of radiology reports generated by the system.
• Demonstrated clinical utility in narrowing the performance gap between junior and senior radiologists.

Conclusions:

• The novel system offers a significant advancement in automated brain MRI report generation.
• The RadGenome-Brain MRI dataset and AutoRG-Brain framework will foster further research in explainable AI for medical imaging.
• Public release of resources aims to accelerate development and adoption in clinical practice.