INTRACRANIAL VESSELWALL SEGMENTATION WITH DEEP LEARNING USING A NOVEL TIERED LOSS FUNCTION TO INCORPORATE CLASS INCLUSION

The goal of this study is 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. 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 levelset 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. 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 mm, 0.345 +/- 0.419 mm, and mean surface distance (MSD) of 0.083 +/- 0.037 mm and 0.103 +/- 0.032 mm for the lumen and vessel wall, respectively, on a test set; compared favorably to a benchmark UNet model that achieved DSC 0.924 +/- 0.047, 0.794 +/- 0.082, HD 0.298 +/- 0.477 mm, 0.394 +/- 0.431 mm, and MSD 0.087 +/- 0.056 mm, 0.119 +/- 0.059 mm.

Automated summary

This study aims to develop an automated method for vessel wall segmentation 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. A novel method is proposed, which simultaneously estimates the inner and outer vessel wall boundaries using a network with a single output channel resembling the levelset function height. The method achieved promising results on a test set, outperforming a benchmark UNet model in terms of Dice similarity coefficients, Hausdorff distance, and mean surface distance.