Using Convolutional Encoder Networks to Determine the Optimal Magnetic Resonance Image for the Automatic Segmentation of Multiple Sclerosis

Abstract

Multiple Sclerosis (MS) is a neuroinflammatory demyelinating disease that affects over 2,000,000 individuals worldwide. It is characterized by white matter lesions that are identified through the segmentation of magnetic resonance images (MRIs). Manual segmentation is very time-intensive because radiologists spend a great amount of time labeling T1-weighted, T2-weighted, and FLAIR MRIs. In response, deep learning models have been created to reduce segmentation time by automatically detecting lesions. These models often use individual MRI sequences as well as combinations, such as FLAIR2, which is the multiplication of FLAIR and T2 sequences. Unlike many other studies, this seeks to determine an optimal MRI sequence, thus reducing even more time by not having to obtain other MRI sequences. With this consideration in mind, four Convolutional Encoder Networks (CENs) with different network architectures (U-Net, U-Net++, Linknet, and Feature Pyramid Network) were used to ensure that the optimal MRI applies to a wide array of deep learning models. Each model had used a pretrained ResNeXt-50 encoder in order to conserve memory and to train faster. Training and testing had been performed using two public datasets with 30 and 15 patients. Fisher’s exact test was used to evaluate statistical significance, and the automatic segmentation times were compiled for the top two models. This work determined that FLAIR is the optimal sequence based on Dice Similarity Coefficient (DSC) and Intersection over Union (IoU). By using FLAIR, the U-Net++ with the ResNeXt-50 achieved a high DSC of 0.7159.

Multiple Sclerosis Diagnosis Using Machine Learning and Deep Learning: Challenges and Opportunities

“…Numerous studies suffer from limited data sizes [ 40 , 66 , 69 , 70 , 77 , 87 , 90 ]. In addition, many studies had access to data from only one center [ 40 , 62 , 70 ], which may introduce bias.…”

“…Ghosh et al [ 90 ] proposed a method of diagnosing MS using four convolutional encoder networks (CENs) with various network architectures including U-Net, U-Net++, Linknet, and feature pyramid network, where all architectures had the ResNeXt-50 encoder. The dataset used contains MRI scans for 45 MS patients and was collected from two public datasets, which are the University Medical Center of Ljubljana (UMCL) and the MSSEG 2016 challenge training dataset.…”

Framework to Segment and Evaluate Multiple Sclerosis Lesion in MRI Slices Using VGG-UNet

“…In order to validate the merit of the proposed scheme with the existing work, the Dice score is considered and is then compared with the dice value available in other earlier works in the literature [ 10 – 14 ]. Figure 8 confirms that the Dice score achieved with the proposed scheme is better compared to other works found in the literature, and this comparison confirms the significance of the proposed VGG-UNet in detecting the MS lesion from MRI slices.…”

“…Ghosh et al presented a CNN segmentation scheme to extract the MS lesion in a brain MRI slice [ 10 ]. This work implemented CNN schemes such as UNet and UNet++ to extract the MS lesion in 2D MRI slices with modalities T 1, T 2, and Flair and achieved a Dice score >75%.…”

“…After recording the image, a personal check by the doctor or a chosen computer algorithm is employed to inspect the MS lesion in the brain. In most of the earlier works, the computer algorithm-based examination is implemented to segment and evaluate the MS lesion from the chosen brain MRI slice [ 10 – 14 ].…”