3D Compressed Convolutional Neural Network Differentiates Neuromyelitis Optical Spectrum Disorders From Multiple Sclerosis Using Automated White Matter Hyperintensities Segmentations

Abstract

BackgroundMagnetic resonance imaging (MRI) has a wide range of applications in medical imaging. Recently, studies based on deep learning algorithms have demonstrated powerful processing capabilities for medical imaging data. Previous studies have mostly focused on common diseases that usually have large scales of datasets and centralized the lesions in the brain. In this paper, we used deep learning models to process MRI images to differentiate the rare neuromyelitis optical spectrum disorder (NMOSD) from multiple sclerosis (MS) automatically, which are characterized by scattered and overlapping lesions.MethodsWe proposed a novel model structure to capture 3D MRI images’ essential information and converted them into lower dimensions. To empirically prove the efficiency of our model, firstly, we used a conventional 3-dimensional (3D) model to classify the T2-weighted fluid-attenuated inversion recovery (T2-FLAIR) images and proved that the traditional 3D convolutional neural network (CNN) models lack the learning capacity to distinguish between NMOSD and MS. Then, we compressed the 3D T2-FLAIR images by a two-view compression block to apply two different depths (18 and 34 layers) of 2D models for disease diagnosis and also applied transfer learning by pre-training our model on ImageNet dataset.ResultsWe found that our models possess superior performance when our models were pre-trained on ImageNet dataset, in which the models’ average accuracies of 34 layers model and 18 layers model were 0.75 and 0.725, sensitivities were 0.707 and 0.708, and specificities were 0.759 and 0.719, respectively. Meanwhile, the traditional 3D CNN models lacked the learning capacity to distinguish between NMOSD and MS.ConclusionThe novel CNN model we proposed could automatically differentiate the rare NMOSD from MS, especially, our model showed better performance than traditional3D CNN models. It indicated that our 3D compressed CNN models are applicable in handling diseases with small-scale datasets and possess overlapping and scattered lesions.

The Correlation Between White Matter Hyperintensity Burden and Regional Brain Volumetry in Patients With Alzheimer's Disease

“…In the previous study, we found that AccuBrain ® could accurately segment and measure hippocampus volume like FreeSurfer, a well-used tool for brain structure measurement worldwide (Abrigo et al, 2019 ). We also proved that AccuBrain ® had good accuracy and reproducibility in segmenting and measuring WMHs, especially in 3D T2-weighted fluid-attenuated inversion recovery (T2-FLAIR) MRI (Guo et al, 2019 ; Wang et al, 2020b ). Different from previous studies, this cross-sectional study not only explored the relationship between WMHs and quantitative medial temporal lobe atrophy (QMTA) but also analyzed the correlation between WMHs and atrophy of occipital, temporal, frontal, parietal, and insular lobe.…”

Machine Learning Approaches in Study of Multiple Sclerosis Disease Through Magnetic Resonance Images

“…Two of them used different biomarkers and ML learning classifiers to distinguish MS from Neuromyelitis Optical Spectrum Disorders (NMOSD) ( 53 , 54 ). The first article employed a CNN for lesion segmentation of MRI images on 213 patients with MS and 125 patients with NMOSD.…”

“…54 publications were grouped into 4 categories based on the application of ML Learning in MS disease.…”

A joint model for lesion segmentation and classification of MS and NMOSD

“…Due to limited available data, they primarily utilized SqueezeNet to prevent overfitting, achieving an accuracy of 0.81, sensitivity of 0.80, and specificity of 0.83 using common features to classify MS and NMOSD. Wang et al (2020) compressed 3D MS and NMOSD MRI images into multi-channel 2D images and used a 2D ResNet model for classification. By leveraging transfer learning subsequent to pre-training the model on ImageNet, they attained an accuracy of 0.75.…”