Auto-Rad: End-to-End Report Generation from Lumber Spine MRI Using Vision-Language Model

Area of Science:

• Artificial Intelligence in Radiology
• Medical Imaging Analysis
• Natural Language Processing

Background:

• Lumbar spinal stenosis (LSS) is a common cause of chronic lower back and leg pain.
• Traditional LSS diagnosis relies on time-consuming radiologist analysis of MRI scans.
• There is a need for efficient diagnostic tools to manage LSS patient care.

Purpose of the Study:

• To develop an automated radiology report generation (ARRG) system for LSS diagnosis.
• To leverage vision-language (VL) models for streamlining the interpretation of lumbar spine MRI scans.
• To reduce the diagnostic workload associated with LSS reporting.

Main Methods:

• A Generative Image-to-Text (GIT) model, adapted from visual question answering (VQA), was fine-tuned for MRI report generation.
• The GIT model was trained on a curated dataset of annotated lumbar spine MRI scans.
• GPT-4 was employed to enhance text coherence for the GIT model's comprehension.

Main Results:

• The ARRG system generated reports that were semantically accurate and grammatically coherent.
• Performance metrics included METEOR (0.37), BERTScore (0.886), and ROUGE-L (0.3).
• The results demonstrate the model's capability to produce clinically relevant diagnostic content.

Conclusions:

• Vision-language models show significant potential for automating medical imaging report generation.
• This technology can effectively reduce the diagnostic burden on radiologists.
• Automated reporting systems offer a promising avenue for improving efficiency in LSS diagnosis.