Segmentation of Peripheral Nerves From Magnetic Resonance Neurography: A Fully-Automatic, Deep Learning-Based...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Peripheral neuropathies require accurate diagnosis using methods like magnetic resonance neurography (MRN).
• Manual segmentation of peripheral nerves in MRN images is time-consuming and subject to inter-rater variability.

Purpose of the Study:

• To develop and evaluate a fully-automatic segmentation method for peripheral nerves in the thigh using MRN.
• To assess the accuracy and efficiency of the automated method compared to manual segmentation.

Main Methods:

• A fully-convolutional neural network was trained to segment peripheral nerves from T2-weighted MRN sequences.
• The method was evaluated on MRN data from 10 healthy volunteers and 42 patients with sciatic neuropathy.
• Performance was quantified using Dice coefficients, Hausdorff distances, and volumetric similarity, compared against manual segmentation.

Main Results:

• The automated method achieved high accuracy, with Dice coefficients of 0.859 and 0.719 for healthy and patient cohorts, respectively.
• Segmentation completion time was under one second, drastically reducing the average manual segmentation time of 19 minutes.
• The method demonstrated good volumetric similarity and Hausdorff distance metrics in both cohorts.

Conclusions:

• Fully-automatic segmentation of healthy and neuropathic sciatic nerves is feasible from standard MRN images with good accuracy.
• The rapid, automated approach offers a clinically viable tool for diagnosing peripheral neuropathies and quantifying lesion burden.
• Potential applications include biomarker development for lesion burden and defining regions of interest for advanced MRN techniques.