Convolutional neural network based system for fully automatic FLAIR MRI segmentation in multiple sclerosis diagnosis

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Neurology

Background:

• Multiple Sclerosis (MS) diagnosis relies on accurate lesion segmentation in FLAIR MRI.
• Manual segmentation is time-consuming and prone to inter-rater variability.
• Automated methods are needed to improve efficiency and consistency in MS lesion detection.

Purpose of the Study:

• To develop and evaluate an automated Convolutional Neural Network (CNN) system for segmenting FLAIR MRI images in Multiple Sclerosis (MS).
• To assess the system's performance on both internal and external datasets for clinical applicability.
• To leverage the nnU-Net architecture for robust MS lesion segmentation.

Main Methods:

• Utilized a dataset of FLAIR MRI scans from 103 internal and 10 external patients.
• Applied preprocessing techniques including skull stripping, normalization, resizing, and data augmentation.
• Employed the nnU-Net architecture, trained with fivefold cross-validation for slice-level classification and voxel-level segmentation.

Main Results:

• Slice-level classification achieved 83% accuracy (internal) and 76% (external), with high sensitivity (100%) for both.
• Voxel-level segmentation yielded a Dice Similarity Coefficient (DSC) of 70% (internal) and 75% (external).
• The system demonstrated high negative predictive value (NPV) on both datasets, indicating reliability.

Conclusions:

• The developed CNN-based system with nnU-Net architecture is accurate and reliable for segmenting MS lesions in FLAIR MRI.
• This automated approach has the potential to significantly enhance clinical decision-making in Multiple Sclerosis diagnosis.
• The system's performance on external data suggests good generalizability for clinical use.