DenseLes: slice-wise dense network for multiple sclerosis lesion segmentation and classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Accurate segmentation of multiple sclerosis (MS) lesions in magnetic resonance imaging (MRI) is critical for diagnosis and disease monitoring.
• Automated methods offer efficient solutions for rapid patient data analysis.

Purpose of the Study:

• To propose a convolutional neural network (CNN)-based method, DenseLe, for automated MS lesion segmentation from FLAIR MRI images.
• To evaluate DenseLe's performance against existing methods and human raters.

Main Methods:

• The DenseLe system involves two stages: pre-processing (brain extraction, standardization) and end-to-end slice-wise dense network segmentation.
• Lesion localization was performed in specific anatomical regions: periventricular, (juxta)cortical, infratentorial, and spinal.
• The model was evaluated on a custom dataset and the public MSSEG 2016 challenge dataset.

Main Results:

• DenseLe achieved a significant improvement in segmentation quality, with an average Dice score of 0.80% on the Szeged MS dataset.
• On the MSSEG 2016 dataset, DenseLe yielded Dice scores between 0.32% and 0.73%.
• The performance was comparable to that of human raters on the MSSEG 2016 dataset.

Conclusions:

• The proposed DenseLe method demonstrates robust and efficient automated segmentation of MS lesions from MRI.
• This AI-driven approach shows potential for improving the accuracy and speed of MS diagnosis and monitoring.