Deep Learning Framework for Multi-Class Classification of Brachial Plexopathy MRI Using Transfer Learning

Abstract

Diagnosing brachial plexus injuries is challenging due to the network’s complex anatomy, overlapping clinical symptoms, and reliance on radiologist expertise, often leading to inconsistent interpretations. To address this limitation, this study proposes a deep learning framework for automatic classification of brachial plexopathy MRI images using transfer learning. Six convolutional neural network architectures, DenseNet121, EfficientNetB0, InceptionV3, MobileNetV2, NAS-NetMobile, and ResNet50, were compared, and a modified MobileNetV2-SlimARC model was introduced. Experimental results showed that MobileNetV2-SlimARC achieved the best performance, with 99.32% and 97.26% accuracy for binary and multi-class classification, respectively. These findings demonstrate the model’s efficiency and clinical potential as a lightweight, reliable AI-assisted diagnostic tool for evaluating brachial plexopathy nerve injuries.