Etiology-based classification of brain white matter hyperintensity on magnetic resonance imaging

Abstract

Brain white matter lesions found upon magnetic resonance imaging are often observed in psychiatric or neurological patients. Individuals with these lesions present a more significant cognitive impairment when compared with individuals without them. We propose a computerized method to distinguish tissue containing white matter lesions of different etiologies (e.g., demyelinating or ischemic) using texture-based classifiers. Texture attributes were extracted from manually selected regions of interest and used to train and test supervised classifiers. Experiments were conducted to evaluate texture attribute discrimination and classifiers' performances. The most discriminating texture attributes were obtained from the gray-level histogram and from the co-occurrence matrix. The best classifier was the support vector machine, which achieved an accuracy of 87.9% in distinguishing lesions with different etiologies and an accuracy of 99.29% in distinguishing normal white matter from white matter lesions.

Rationale, design and methodology of the image analysis protocol for studies of patients with cerebral small vessel disease and mild stroke

“…; Leite et al. ) or supervised classifiers (Anbeek et al. ) seeking to minimize observer input in the assessment of T2W/FLAIR hyperintensities.…”

Deep Learning vs. Conventional Machine Learning: Pilot Study of WMH Segmentation in Brain MRI with Absence or Mild Vascular Pathology

“…Secondly, we automatically extract 44 first-and second-order statistical measures out of the intensity values from 5 × 5 2D ROIs by using histogram analysis (i.e., mean, variance, skewness, kurtosis and 1%, 10%, 50%, 90% and 99% percentiles), grey-level co-occurrence matrix or GLCM (i.e., using 0 • , 45 • , 90 • and 135 • orientations and distance of 1 and 2 of neighbouring voxels), grey-level run-length matrix or GLRLM (i.e., using 0 • , 45 • , 90 • and 135 • orientations) and statistical analysis of gradient (i.e., mean, variance, skewness, kurtosis and percentage of voxels with non gradient) as in study done by Leite et al [4]. Lastly, 125 MR image's grey scale values and 1875 response values from Gabor filter (i.e., 32 filters from 4 directions and 4 magnitudes) are extracted from 5 × 5 × 5 3D ROIs and used as features as in study done by Klöppel et al [3].…”

“…In many cases, ones have to test several different feature extraction methods before finding suitable features for the task. For this study, three different sets of features from three different studies by Klöppel et al [3], Leite et al [4] and Ithapu et al [5] are used for segmenting WMH using conventional machine learning algorithms. We use the same set of features that proved relevant for this task in previous studies and are implemented in publicly available tools, such as the W2MHS toolbox.…”

“…Before the emergence of deep learning algorithms, most of the works were based on support vector machine (SVM) and random forests (RF) for which some image features need to be extracted first. Some notable works were done by Klöppel et al [3,4] and Leite et al [5] where several feature extraction methods and learning algorithms were evaluated to find the best possible combination for WMH segmentation. However, these studies cannot be compared to each other directly because they use different feature extraction methods and MRI datasets.…”

“…SET-2 contains histogram statistic, GLCM, GLRLM and gradient image statistic from T2-FLAIR that is used in [4]. The feature vector size is 44.…”