Contrastive Learning with Self-reconstruction for 3D Intracranial Aneurysm Detection
Year: 2024
Type: conference-paper
Abstract: Despite its wide success, 3D point cloud has not yet been exploited for intelligent disease detection and recognition. Also, most current disease detection methods are supervised and require labels for training. In this paper, we propose an unsupervised method for intracranial aneurysm detection based on 3D medical point cloud data. In particular, we devise an unsupervised dual-branch network which consists of contrastive learning and reconstruction tasks, namely CORE. Our method enables the effective extraction of both integral and point-wise features of point clouds. We specifically design a two-part reconstruction decoder. The first part is a fully-connected decoder to predict a coarse point set that represents the global geometry. The second part is a folding-based decoder to approximate the surface of the local geometry. Then we employ a dual-optimisers strategy so that the parameters of each branch are optimised separately to avoid the interaction between contrastive learning and reconstruction that may cause the local optimum issue. We conduct two evaluation schemes on the public IntrA dataset: unsupervised linear classification and fine-tuning. CORE achieves the best unsupervised learning results and state-of-the-art fine-tuning results. In addition, the generalization ability of CORE is also successfully validated on ModelNet40/10 and ShapeNetPart which are the general 3D point cloud benchmarks.
Institutions: Deakin University, UNSW Sydney, La Trobe University
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4403941925