Abnormal Brachial Plexus Differentiation from Routine Magnetic Resonance Imaging: An AI-based Approach

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Brachial plexopathy diagnosis relies on identifying brachial plexus (BP) injuries.
• Accurate localization and identification of neurological injury are crucial for clinical practice.
• Automatic abnormality identification in BP MRI is an emerging area.

Purpose of the Study:

• To develop and evaluate an AI-driven approach for differentiating abnormal brachial plexus (BP) from normal MRI.
• To assess the efficacy of AI models across different MRI sequences (T1, fluid-sensitive, post-gadolinium).

Main Methods:

• Collected a BP dataset curated by radiological experts.
• Employed a semi-supervised AI method (nnU-net) for BP segmentation.
• Utilized radiomics to extract 107 shape and texture features.
• Trained and optimized six machine learning classifiers with dynamic feature selection.

Main Results:

• Shape features were more sensitive than texture features for detecting abnormal BP.
• Logistic and Bagging classifiers demonstrated superior performance.
• Models trained on fluid-sensitive sequences significantly outperformed those on T1 and post-gadolinium sequences.
• Classification accuracy and AUC score exceeded 90% for fluid-sensitive sequences.

Conclusions:

• AI shows substantial potential and feasibility for clinical integration in brachial plexopathy diagnosis.
• Fluid-sensitive MRI sequences are highly effective for AI-based BP abnormality detection.
• The developed AI approach offers a promising tool for improving diagnostic accuracy.