FLAMeS: A Robust Deep Learning Model for Automated Multiple Sclerosis Lesion Segmentation

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Accurate assessment of brain lesions in multiple sclerosis (MS) is critical for research.
• Manual segmentation of MS lesions on MRI is time-consuming and lacks consistency.

Purpose of the Study:

• To develop an automated algorithm for segmenting MS lesions on T2-weighted fluid-attenuated inversion recovery (FLAIR) MRI.
• To evaluate the performance of the developed algorithm against existing methods.

Main Methods:

• Developed FLAIR Lesion Analysis in Multiple Sclerosis (FLAMeS), a deep learning algorithm using the nnU-Net architecture.
• Trained FLAMeS on 668 MS FLAIR MRI scans (1.5 and 3 Tesla).
• Evaluated FLAMeS on three external datasets and compared it with SAMSEG, LST-LPA, and LST-AI using qualitative and quantitative metrics.

Main Results:

• Qualitative review by blinded experts favored FLAMeS in 17 out of 20 scans.
• FLAMeS achieved a mean Dice score of 0.74, true positive rate of 0.84, and F1 score of 0.78 across testing datasets.
• FLAMeS outperformed benchmark methods in lesion segmentation accuracy, particularly for smaller lesions.

Conclusions:

• FLAMeS is an accurate and robust method for automated MS lesion segmentation.
• The developed algorithm demonstrates superior performance compared to other publicly available methods for MS lesion segmentation.