Automated peripheral nerve segmentation for MR-neurography.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Magnetic resonance neurography (MRN) is a key tool for diagnosing peripheral neuropathies.
• Quantitative MRN analysis requires nerve segmentation, which is currently manual, time-consuming, and prone to errors.
• Automated segmentation methods are needed to integrate quantitative measures into routine clinical practice.

Purpose of the Study:

• To develop and evaluate a deep learning-based neural network for automated segmentation of peripheral nerves in MRN.
• To assess the performance of the automated segmentation model using quantitative metrics.

Main Methods:

• A neural network was trained to segment the sciatic nerve and its branches on MRN scans from 35 healthy individuals (70 training examples) using 5-fold cross-validation.
• Model performance was evaluated on an independent test set of MRN scans from 60 healthy individuals.

Main Results:

• The model achieved a mean Dice similarity coefficient (DSC) of 0.892 and a Jaccard index (JI) of 0.806 during cross-validation.
• On the independent test set, the model yielded a mean DSC of 0.789 and a mean JI of 0.672.
• Mean Hausdorff distance (HD) was 2.146 in cross-validation and 2.118 on the test set.

Conclusions:

• The deep learning model demonstrates promising performance for automated peripheral nerve segmentation in MRN.
• These findings provide a baseline for developing an automated quantitative MRN analysis framework for clinical use.
• Future work will expand training data and include patients with neuropathies to enhance disease characterization.