A More Stable and Accurate Genetic Algorithm (GA) for Segmentation of 3D Medical Images
Yi Wang,Yilong Niu,Yun Tian,Jianyuan Dong,Chongyang Hao
DOI: https://doi.org/10.3969/j.issn.1000-2758.2007.03.026
2007-01-01
Xibei Gongye Daxue Xuebao/Journal of Northwestern Polytechnical University
Abstract:Aim. In our opinion, the two shortcomings in traditional searching algorithms are: (1) the simple genetic algorithm (SGA) is deficient in stability and accuracy, (2) the Complete Search (CS) algorithm is time-consuming. We now propose a GA that we believe is better. In the full paper, we explain our better 3D medical image segmentation algorithm in detail. In this abstract, we just add some pertinent remarks to listing the two topics of explanation. The first topic is: the traditional optimal entropy multi-threshold method. The second topic is: the improvement on the 3D image segmentation searching GA using two thresholds. Its two subtopics are the selection of parameters and operators of the GA (subtopic 2.1) and the improvement on the SGA (subtopic 2.2). In topic 2, we, using Simulated Annealing idea for reference, stretch the optimal entropy function of 3D medical images to construct the new fitness function of the improved genetic algorithm that adopts the elitist strategy to accelerate the convergence. Finally, we conducted experiments, whose results are summarized in one table and one figure in the full paper. These results show preliminarily that our better algorithm can: (1) find good thresholds; (2) reduce the time required for searching thresholds by 60% as compared with the traditional CS algorithm; (3) most importantly, as compared with SGA, be more stable and accurate, since the relative error is improved from 9.2297 × 10-4 to 1.7621 × 10-4 and the variance is improved from 195.29 × 10-6 to 2.2207 × 10-6.