White matter hyperintensities segmentation using an ensemble of neural networks

Abstract

White matter hyperintensities (WMHs) represent the most common neuroimaging marker of cerebral small vessel disease (CSVD). The volume and location of WMHs are important clinical measures. We present a pipeline using deep fully convolutional network and ensemble models, combining U‐Net, SE‐Net, and multi‐scale features, to automatically segment WMHs and estimate their volumes and locations. We evaluated our method in two datasets: a clinical routine dataset comprising 60 patients (selected from Chinese National Stroke Registry, CNSR) and a research dataset composed of 60 patients (selected from MICCAI WMH Challenge, MWC). The performance of our pipeline was compared with four freely available methods: LGA, LPA, UBO detector, and U‐Net, in terms of a variety of metrics. Additionally, to access the model generalization ability, another research dataset comprising 40 patients (from Older Australian Twins Study and Sydney Memory and Aging Study, OSM), was selected and tested. The pipeline achieved the best performance in both research dataset and the clinical routine dataset with DSC being significantly higher than other methods (p < .001), reaching .833 and .783, respectively. The results of model generalization ability showed that the model trained on the research dataset (DSC = 0.736) performed higher than that trained on the clinical dataset (DSC = 0.622). Our method outperformed widely used pipelines in WMHs segmentation. This system could generate both image and text outputs for whole brain, lobar and anatomical automatic labeling WMHs. Additionally, software and models of our method are made publicly available at https://www.nitrc.org/projects/what_v1.

Higher‐resolution quantification of white matter hypointensities by large‐scale transfer learning from 2D images on the JPSC‐AD cohort

“…The ConvNet research work has often focused on the small sample regime and architecture improvement (Fartaria et al, 2016;Ghafoorian et al, 2017;Guerrero et al, 2018;La Rosa et al, 2020;Li et al, 2018;Li et al, 2021;Liang et al, 2021;Moeskops et al, 2018;Orbes-Arteaga et al, 2018;Valverde et al, 2017). Recently (Kuijf et al, 2019) organized a competition, using 60 (3D or thin-sliced 2D) FLAIR images as training set, and 110 manually labeled subjects for evaluation.…”

Adherence to a healthy lifestyle and brain structural imaging markers

“…Briefly, structural MRI data were processed applying a pipeline to the T1 images that used gradient distortion correction, field of view reduction, registration to the standard atlas, brain extraction, defacing, and finally segmentation. In PRECISE, each T1 weighted images was processed using FreeSurfer default processing pipeline (version 7.0) and WMHV data was summarized applying White matter Hyperintensities Analysis Tools (WHAT) software 31 ; meanwhile, corresponding imaging variables in UKB study were derived from the image-derived phenotypes (IDPs) released by the UK Biobank team 32 . In PRECISE, lacune of presumed vascular origin, as a marker of cerebral small vessel disease, was defined as rounded or ovoid lesion in the subcortical, BG, or brain stem, with diameter ranging from 3 to 15 mm and cerebrospinal fluid signal density on T2 and FLAIR sequences and no increased signal on DWI 33 .…”

Ensemble learning via supervision augmentation for white matter hyperintensity segmentation

“…Li et al ( 2018 ) proposed an ensemble of three U-Net's with different random weight initializations to automatically detect WMH. Li et al ( 2022 ) present a pipeline using deep fully convolutional network and ensemble models, combining U-Net, SE-Net, and multi-scale features, to automatically segment WMHs and estimate their volumes and locations. Sundaresan et al ( 2021 ) achieves ensemble by combining three different planes of brain MR images.…”