Convolutional Neural Network for Automated FLAIR Lesion Segmentation on Clinical Brain MR Imaging

Abstract

BACKGROUND AND PURPOSE: Most brain lesions are characterized by hyperintense signal on FLAIR. We sought to develop an automated deep learning-based method for segmentation of abnormalities on FLAIR and volumetric quantification on clinical brain MRIs across many pathologic entities and scanning parameters. We evaluated the performance of the algorithm compared with manual segmentation and existing automated methods. MATERIALS AND METHODS: We adapted a U-Net convolutional neural network architecture for brain MRIs using 3D volumes. This network was retrospectively trained on 295 brain MRIs to perform automated FLAIR lesion segmentation. Performance was evaluated on 92 validation cases using Dice scores and voxelwise sensitivity and specificity, compared with radiologists' manual segmentations. The algorithm was also evaluated on measuring total lesion volume. RESULTS: Our model demonstrated accurate FLAIR lesion segmentation performance (median Dice score, 0.79) on the validation dataset across a large range of lesion characteristics. Across 19 neurologic diseases, performance was significantly higher than existing methods (Dice, 0.56 and 0.41) and approached human performance (Dice, 0.81). There was a strong correlation between the predictions of lesion volume of the algorithm compared with true lesion volume (ϭ 0.99). Lesion segmentations were accurate across a large range of image-acquisition parameters on Ͼ30 different MR imaging scanners. CONCLUSIONS: A 3D convolutional neural network adapted from a U-Net architecture can achieve high automated FLAIR segmentation performance on clinical brain MR imaging across a variety of underlying pathologies and image acquisition parameters. The method provides accurate volumetric lesion data that can be incorporated into assessments of disease burden or into radiologic reports. ABBREVIATIONS: BIANCA ϭ Brain Intensity Abnormality Classification Algorithm; CNN ϭ convolutional neural network; FDR ϭ false discovery rate; LST ϭ lesion segmentation tool; RMdSPE ϭ root median squared percentage error; RMSPE ϭ root mean squared percentage error; SVID ϭ small-vessel ischemic disease A pproximately 36 million MR imaging studies are performed annually in the United States, and this number is rising. 1 Approximately 65% of these MRIs are used to assess the central nervous system. The FLAIR sequence is universally used to identify and characterize imaging abnormalities in terms of location, size, and extent, due to its broad utility across many pathologies and lesion appearances. Specific applications of FLAIR include, among numerous others, primary and metastatic brain tumors; demyelinating, autoimmune, infectious, and inflammatory conditions; and ischemia. 2-4 Because of its general utility, FLAIR is acquired on nearly every clinical brain MRI. There is a growing need to develop fully automated, rapid, precise, quantitative assessments of FLAIR abnormalities to standardize quantitative descriptions of pathology. A quantitative lesion-burden assessment has the potential to reduce er...

Comparison and validation of seven white matter hyperintensities segmentation software in elderly patients

“…It uses a U-net convolutional neural network ( Ronneberger et al, 2015 ) architecture. Notably, Duong and colleagues ( Duong et al, 2019 ) and four of the top ten challengers of this challenge used this architecture, but they have not released a user-friendly version. We also did not include the promising LesionBRAIN ( Coupé et al, 2018 ) (Supplementary Table 4 ) because it does not provide the required segmentation mask.…”

RatLesNetv2: A Fully Convolutional Network for Rodent Brain Lesion Segmentation

“…Deep learning, and more specifically convolutional neural networks (ConvNets), has become increasingly popular due to its competitive performance in medical image segmentation. Literature on brain lesion segmentation in MR images with ConvNets is dominated by approaches tested on human-derived data (e.g., Duong et al, 2019 ; Gabr et al, 2019 ; Yang et al, 2019 ). Despite using ConvNets, typical brain lesion segmentation approaches are multi-step, i.e., they rely on preprocessing procedures, such as noise reduction, registration, skull-stripping and inhomogeneity correction.…”

A multi-path 2.5 dimensional convolutional neural network system for segmenting stroke lesions in brain MRI images

“…Indeed, another source of brain pathology that changes over time and to which CNN models have been applied with some success is tumor. Studies such as that of (Duong et al., 2019) suggest that training on numerous types of brain pathologies imaged with FLAIR, including tumor, can yield Dice coefficients between human and model segmentations on the order of 0.79. This apparently high level of overlap points to the potential of a multi-pathology approach.…”