Improve accuracy for automatic acetabulum segmentation in CT images

Abstract

Separation of the femur head and acetabulum is one of main difficulties in the diseased hip joint due to deformed shapes and extreme narrowness of the joint space. To improve the segmentation accuracy is the key point of existing automatic or semi-automatic segmentation methods. In this paper, we propose a new method to improve the accuracy of the segmented acetabulum using surface fitting techniques, which essentially consists of three parts: (1) design a surface iterative process to obtain an optimization surface; (2) change the ellipsoid fitting to two-phase quadric surface fitting;(3) bring in a normal matching method and an optimization region method to capture edge points for the fitting quadric surface. Furthermore, this paper cited vivo CT data sets of 40 actual patients (with 79 hip joints). Test results for these clinical cases show that: (1) the average error of the quadric surface fitting method is 2.3 (mm); (2) the accuracy ratio of automatically recognized contours is larger than 89.4%;(3) the error ratio of section contours is less than 10% for acetabulums without severe malformation and less than 30% for acetabulums with severe malformation. Compared with similar methods, the accuracy of our method, which is applied in a software system, is significantly enhanced.

Deep Learning and Computer Vision Techniques for Automated Total Hip Arthroplasty Planning on 2-D Radiographs

“…These methods can be roughly categorized as supervised or unsupervised approaches based on prior knowledge. Unsupervised approaches that do not use prior knowledge can be further categorized as region growing [24], thresholding [25], [26], graph-cut [27], surface fitting [28], and harmonic field approaches [29]. Supervised approaches using prior knowledge can be further categorized as statistical shape models (SSM) [30]- [32], active shape models (ASM) [33], atlas-based [34], [35], and patch-based approaches [36].…”