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

M.T. Duong; J.D. Rudie; J. Wang; L. Xie; S. Mohan; J.C. Gee; A.M. Rauschecker

doi:10.3174/ajnr.A6138

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 (ρ = 0.99). Lesion segmentations were accurate across a large range of image-acquisition parameters on >30 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: Brain Intensity Abnormality Classification Algorithm
CNN: convolutional neural network
FDR: false discovery rate
LST: lesion segmentation tool
RMdSPE: root median squared percentage error
RMSPE: root mean squared percentage error
SVID: small-vessel ischemic disease

Footnotes

Michael Tran Duong and Jeffrey D. Rudie contributed equally to this work and are co-first authors.
Disclosures: Jeffrey D. Rudie—RELATED: Grant: National Institutes of Health T32-EB004311-10 (Research Track Radiology Residency), Radiological Society of North America Resident Research Grant (RR1778)*; UNRELATED: Grants/Grants Pending: American Society for Neuroradiology, Radiological Society of North America, Comments: I have applied for additional grants to fund related work.* Suyash Mohan—UNRELATED: Grants/Grants Pending: Galileo Clinical Decision Support, Comments: research grant.* Andreas M. Rauschecker—RELATED: Grant: Radiological Society of North America Resident Research and Education Foundation, Comments: Resident Research Grant RR1778*; Other: National Institutes of Health, Comments: T32 Training Grants: T32-EB004311–10, T32-EB001631–14*; UNRELATED: Support for Travel to Meetings for the Study or Other Purposes: American College of Radiology–Association of University Radiologists Research Scholar Program, Comments: $1000 to support travel to and presentation at the Association of University Radiologists 2018 on this work.* James C. Gee—RELATED: Grant: National Institutes of Health, Comments: National Institutes of Health grants supported the contribution and participation of me and my lab members with respect to development, evaluation, and application of methods for neuroimage processing and analysis*; UNRELATED: Employment: University of Electronic Science and Technology of China, Comments: I receive effort support for my role as Director of Center for Health Innovation at University of Pennsylvania–University of Electronic Science and Technology of China; Stock/Stock Options: mutual funds, Comments: as part of general retirement investment portfolio; Travel/Accommodations/Meeting Expenses Unrelated To Activities Listed: various scientific societies and academic institutions, Comments: for invited scientific lectures and presentations. *Money paid to the institution.
A.M. Rauschecker was supported by a Radiological Society of North America Resident Grant (RR1778). A.M. Rauschecker and Jeffrey D. Rudie were supported by a National Institutes of Health T-32 Training Grant from Penn Radiology for the duration of the project (T32-EB004311-10). A.M. Rauschecker was also supported by a National Institutes of Health T-32 Training Grant from the University of California, San Francisco, Radiology, for a portion of the project (T32-EB001631-14).