Segmentation of spinal rootlets across MRI contrasts with RootletSeg

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• Spinal nerve rootlet segmentation is crucial for accurate spinal level estimation, lesion classification, and therapeutic interventions.
• Existing methods for spinal nerve rootlet segmentation are often manual and time-consuming.
• Automated segmentation can significantly improve the efficiency and consistency of spinal imaging analyses.

Purpose of the Study:

• To develop and validate a deep learning-based method for automatic segmentation of C2-T1 dorsal and ventral spinal nerve rootlets.
• To assess the performance of the developed method across different MRI contrasts and datasets.
• To provide an open-source tool for researchers and clinicians.

Main Methods:

• A deep learning model, termed RootletSeg, was developed using 93 MRI scans from 50 healthy adults.
• The model was trained and evaluated on various MRI sequences including 3T T2-weighted (T2w) and 7T MP2RAGE (T1w INV1, INV2, UNIT1).
• Performance was quantified using the Dice score across different MRI contrasts.

Main Results:

• RootletSeg achieved a mean Dice score ranging from 0.62 ± 0.10 (T1w-INV1) to 0.67 ± 0.09 (T1w-INV2).
• The model demonstrated accurate segmentation of C2-T1 spinal rootlets across T1w and T2w MRI contrasts.
• The segmentation enabled direct determination of spinal levels from MRI scans.

Conclusions:

• RootletSeg provides an effective and automated solution for spinal nerve rootlet segmentation across various MRI modalities.
• The open-source nature of RootletSeg facilitates its integration into diverse downstream neuroimaging analyses.
• This tool has the potential to advance research in spinal cord imaging and related clinical applications.