A content-based 3D shape retrieval system for Abdominal Aortic Aneurysm rupture risk prediction

Abstract

In this work, we present a novel content-based 3D shape retrieval system for Abdominal Aortic Aneurysm (AAA) rupture risk prediction. The algorithms incorporate shape context, RANdom SAmple Consensus (RANSAC) and thin plate spline (TPS) to achieve a reliable AAA rupture risk assessment system. Pre-labeled unruptured and ruptured cases ('-1' for unruptured and '1' for ruptured cases) are built and given a new shape, and the geometric distances between the new shape and all shapes in a 3D shape library are measured. The rupture risk index (RRI) is calculated from the distances and labels of the library. The correspondences between the new shape and the library are built based on shape context and RANSAC. In addition, optimized non-rigid transformations are computed using TPS. The distances are defined as the transformation energies of TPS, which are the linear combination of rigid transformation energies and non-rigid (or bending) energies. The RRI is derived from the weighted sum of labels and the weights are calculated from the distances.