Deep Learning-Based Knee MRI Classification for Common Peroneal Nerve Palsy with Foot Drop

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Foot drop is a challenging condition with various causes, including common peroneal nerve (CPN) injuries.
• Accurate diagnosis of CPN injury is crucial for effective treatment and management of foot drop.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for classifying foot drop caused by CPN injury using knee MRI axial images.
• To compare the performance of different convolutional neural network (CNN) models in identifying CPN injuries.

Main Methods:

• A retrospective study utilizing knee MRI data from patients with confirmed CPN injury and those with non-traumatic knee pain.
• Training and validation of CNN models including EfficientNet-B5, ResNet152, and VGG19 on a dataset of 2286 MRI scans.
• Performance evaluation using metrics such as Area Under the Receiver Operating Characteristic Curve (AUC), precision, recall, accuracy, and F1 score.

Main Results:

• The EfficientNet-B5 model achieved the highest performance in classifying CPN injury versus non-CPN cases, with an AUC of 0.946.
• EfficientNet-B5 demonstrated superior precision, recall, accuracy, and F1 score compared to ResNet152 and VGG19.
• Saliency map analysis indicated that EfficientNet-B5 focused on the nerve area for accurate CPN injury detection.

Conclusions:

• Deep learning analysis of knee MRI images is a successful method for differentiating CPN injuries in patients with foot drop.
• The EfficientNet-B5 algorithm shows significant potential for clinical application in diagnosing CPN-related foot drop.
• AI-driven analysis of medical imaging offers a promising avenue for improving the diagnostic accuracy of neurological conditions.