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

Accurate and reliable segmentation of multiple sclerosis (MS) lesions from magnetic resonance imaging (MRI) is essential for diagnosis and monitoring disease progression. Therefore, a robust and efficient automated approach can rapidly provide information about the patient. Here, a convolutional neural network-based method is proposed to segment lesions from FLAIR images. The DenseLessystem includes two stages: pre-processing of image data (brain extraction, standardization), then segmentation of MS lesions using an end-to-end slice-wise dense network. We also identified the segmented lesions in specific locations [periventricular, (juxta)cortical, infratentorial, and spinal]. DenseLesis evaluated and compared to other methods on our assembled data and the public MSSEG 2016 MS challenge dataset. Our model demonstrates a significant improvement in segmentation quality over previous approaches, achieving an average Dice score of 0.80% on the Szeged MS dataset. On the MSSEG 2016 dataset, our method achieved Dice scores ranging from 0.32% to 0.73%, comparable to those of human raters.