Geometric Deep Learning Using Vascular Surface Meshes for Modality-Independent Unruptured Intracranial Aneurysm Detection

Early detection of unruptured intracranial aneurysms (UIAs) enables better rupture risk and preventative treatment assessment. UIAs are usually diagnosed on Time-of-Flight Magnetic Resonance Angiographs (TOF-MRA) or contrast-enhanced Computed Tomography Angiographs (CTA). Various automatic voxel-based deep learning UIA detection methods have been developed, but these are limited to a single modality. We propose a modality-independent UIA detection method using a geometric deep learning model with high resolution surface meshes of brain vessels. A mesh convolutional neural network with ResU-Net style architecture was used. UIA detection performance was investigated with different input and pooling mesh resolutions, and including additional edge input features (shape index and curvedness). Both a higher resolution mesh (15,000 edges) and additional curvature edge features improved performance (average sensitivity: 65.6%, false positive count/image (FPC/image): 1.61). UIAs were detected in an independent TOF-MRA test set and a CTA test set with average sensitivity of 52.0% and 48.3% and average FPC/image of 1.04 and 1.05 respectively. We provide modality-independent UIA detection using a deep-learning vascular surface mesh model with comparable performance to state-of-the-art UIA detection methods.

I. Introduction

An intracranial aneurysm (IA) is a focal bulging of the vessel wall in the brain. They have a prevalence of approximately 3% in the general population [1]. If an IA ruptures it leads to subarachnoid haemorrhage which can be fatal or lead to long-term disability [2], [3]. It is important that unruptured IAs (UIAs) are detected early to allow for clinicians to make informed rupture and preventative treatment risk assessments [4]. A radiologist usually diagnoses UIAs by visually inspecting Time-Of-Flight Magnetic Resonance Angiographs (TOF-MRA) or contrast-enhanced Computed Tomography Angiographs (CTA) but this inspection can be time-consuming and unreliable. Visual inspection for UIA detection has been found to have a large variability in sensitivity across different studies, varying from as low as 28% for small UIAs to as high as 96% [5], [6], [7], [8]. Automatic detection methods would remove observer variability and speed-up the detection procedure. Various (semi-) automatic detection methods for UIAs exist, including those developed for TOF-MRAs as part of the Aneurysm Detection and segMentation (ADAM) Challenge [9]. All methods submitted to this challenge were voxel-based deep learning methods, including nnU-net [10] and nnDetection [11] with a varying range of sensitivity and false positive count(FPC)/image (top 10 methods: sensitivity = 76% - 59%, FPC/image = 0.18 - 9.37). Voxel-based methods are often limited by their sensitivity to modality and scan acquisition parameters. Geometric deep learning methods operating on vessel surface meshes generally would not have these limitations and could be used instead. In general, geometric deep learning methods have less parameters than voxel-wise deep learning methods. This paper investigates the use of a mesh convolutional neural network for modality-independent UIA detection.