A Deep Learning Approach for Nerve Injury Classification in Brachial Plexopathies Using Magnetic Resonance...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Brachial plexopathies (BPs) present diagnostic challenges due to complex anatomy and overlapping symptoms.
• Magnetic Resonance Neurography (MRN) offers advanced imaging but requires specialized interpretation.
• Accurate classification of nerve injury severity (neuropraxia, axonotmesis, neurotmesis) is crucial for effective treatment.

Purpose of the Study:

• To develop and validate an AI-based framework for improved classification of brachial plexopathies using MRN data.
• To integrate deep learning (DL) with an optimized feature selection algorithm for enhanced diagnostic accuracy.
• To differentiate between normal and abnormal nerve conditions and classify injury severity.

Main Methods:

• A framework combining MobileNetV4 for feature extraction and a modified Hiking Optimization Algorithm (MHOA) with Comprehensive Learning (CL) for feature selection.
• Utilized MRN data from 39 patients with brachial plexopathies across STIR, T2, T1, and DWI sequences.
• Classified injuries based on Seddon's criteria, distinguishing normal/abnormal states and injury severity.

Main Results:

• Achieved 1.0000 accuracy in distinguishing normal from abnormal conditions using STIR and T2 sequences.
• Demonstrated high accuracy (0.9820) in classifying injury severity using STIR, outperforming other metaheuristic algorithms.
• Reported high classification accuracy (0.9667) on DWI sequences, with overall high sensitivity and specificity.

Conclusions:

• The AI framework significantly enhances brachial plexopathy diagnosis by accurately classifying nerve injury types.
• Integration of DL and optimization techniques reduces diagnostic variability, offering a valuable tool for clinical settings.
• This framework has the potential to improve clinical decision-making and patient outcomes through precise diagnoses.