Intracranial vessel wall segmentation with deep learning using a novel tiered loss function incorporating class inclusion

Abstract

Purpose To develop an automated vessel wall segmentation method on T1‐weighted intracranial vessel wall magnetic resonance images, with a focus on modeling the inclusion relation between the inner and outer boundaries of the vessel wall. Methods We propose a novel method that estimates the inner and outer vessel wall boundaries simultaneously, using a network with a single output channel resembling the level‐set function height. The network is driven by a unique tiered loss that accounts for data fidelity of the lumen and vessel wall classes and a length regularization to encourage boundary smoothness. Results Implemented with a 2.5D UNet with a ResNet backbone, the proposed method achieved Dice similarity coefficients (DSC) in 2D of 0.925 ± 0.048, 0.786 ± 0.084, Hausdorff distance (HD) of 0.286 ± 0.436, 0.345 ± 0.419 mm, and mean surface distance (MSD) of 0.083 ± 0.037 and 0.103 ± 0.032 mm for the lumen and vessel wall, respectively, on a test set; compared favorably to a baseline UNet model that achieved DSC 0.924 ± 0.047, 0.794 ± 0.082, HD 0.298 ± 0.477, 0.394 ± 0.431 mm, and MSD 0.087 ± 0.056, 0.119 ± 0.059 mm. Our vessel wall segmentation method achieved substantial improvement in morphological integrity and accuracy compared to benchmark methods. Conclusions The proposed method provides a systematic approach to model the inclusion morphology and incorporate it into an optimization infrastructure. It can be applied to any application where inclusion exists among a (sub)set of classes to be segmented. Improved feasibility in result morphology promises better support for clinical quantification and decision.

VWI‐APP: Vessel wall imaging–dedicated automated processing pipeline for intracranial atherosclerotic plaque quantification

“…34 In the vessel lumen and wall segmentation module,we adopted our recently proposed segmentation method based on deep learning to capture and incorporate the inclusion relationship among classes. 19 Typical lumen and vessel wall segmentation methods either fit active contours to the inner and outer wall boundary based on image intensities and gradients [35][36][37] or infer the pixelwise class membership for the lumen and the vessel wall with multiple output channels with deep networks. 10,38 None of the existing methods accounts for the inclusion relationship between the inner and outer boundaries of the vessel wall and can result in clinically infeasible segmentation solutions such as isolated class pixels or lumen outside of the vessel wall.…”

“…In the vessel lumen and wall segmentation module, we adopted our recently proposed segmentation method based on deep learning to capture and incorporate the inclusion relationship among classes 19 . Typical lumen and vessel wall segmentation methods either fit active contours to the inner and outer wall boundary based on image intensities and gradients 35–37 or infer the pixel‐wise class membership for the lumen and the vessel wall with multiple output channels with deep networks 10,38 .…”

“…Vessel wall and lumen segmentation are performed using our recently developed deep learning method that particularly models the inclusion relationship among the segmented classes with a multiple level-set approach. 19 The network structure adopts a UNet structure with a ResNet backbone 20,21 and has a single output channel with sigmoid activation, as illustrated in Figure 2, for a level-set inference.…”