3D U-Net Application for White Matter Hyperintensity Analysis and Brain Lobe Identification

Kaue Tartarotti Nepomuceno Duarte, David Gobbi, Richard Frayne

Keywords: U-Net, semantic segmentation, white matter hyperintensity (WMH), Fluid attenuated inversion recovery (FLAIR) imaging

TL;DR: 3D U-Net Application for WMH Analysis

Abstract: White-matter hyperintensity (WMH) is associated with many disorders including small vessel diseases where it is suggestive of underlying vascular abnormalities. The fluid-attenuated inversion recovery (FLAIR) magnetic resonance (MR) imaging sequence is commonly used to visualize WMH, since it provides good image contrast not only between WMH and normal tissue, but also between WMH and cerebrospinal fluid (CSF). Manual segmentation of these lesions in brain volumes, on a slice-by-slice basis, is time-consuming, however, this process remains the broadly accepted gold standard. 3D U-shaped convolutional neural networks (3D U-Net CNNs) are a type of semantic segmentation that processes images volumetrically rather than slice by slice. We used a 3D U-Net model to segment the WMH in brain volumes obtained from twenty presumed healthy participants. VGG16, VGG19, ResNet and Efficientnet were used as feature extractors. Results were grouped by brain lobe (frontal, insula, occipital, parietal, temporal) to identify regions that are more affected by the WMH. In general, the predicted volumes had an intersection over union (IoU) measure $>95\%$ compared to manual segmentation. This metric demonstrated that 3D U-Net CNN models are reliable for WMH identification. Assessments of these architectures have shown that the correspondence between WMH lesions and the brain lobes was meaningful and might provide future insights about presumed abnormalities.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: methodological development

Primary Subject Area: Segmentation

Secondary Subject Area: Detection and Diagnosis

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.