Automatic Radiology Report Generation Based on State-Space Model.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology Report Generation
• Deep Learning for Healthcare

Background:

• Physician workload in radiology report generation impacts efficiency and patient care.
• Accurate detection of subtle pathologies in X-rays is challenging due to minimal inter-image differences.
• Existing methods struggle with the complexity of medical image interpretation and report generation.

Purpose of the Study:

• To develop an automated radiology report generation system that enhances efficiency and accuracy.
• To address the challenge of identifying subtle pathologies in X-ray images.
• To improve the consistency between medical images and generated radiology reports.

Main Methods:

• Proposed a novel method with three modules: Self-Attention Mamba (Self-Mamba), Cross-Attention Mamba (Cross-Mamba), and Sparse Mask Loss Function (Sparse-Loss).
• Self-Mamba module models global information for extracting abnormal area features in X-rays.
• Cross-Mamba module optimizes cross-modal interaction between images and reports; Sparse-Loss addresses sample imbalance.

Main Results:

• The proposed approach demonstrated superior performance compared to existing models on key metrics.
• Achieved excellent results on the IU-Xray and COV-CTR publicly available datasets.
• The method effectively extracts features of abnormal areas and enhances image-report consistency.

Conclusions:

• The novel AI method significantly improves radiology report generation efficiency and accuracy.
• The Self-Attention Mamba and Cross-Attention Mamba modules offer a promising direction for medical image analysis.
• This approach has the potential to reduce patient waiting times and alleviate physician burden.