Breast DCE-MRI segmentation for lesion detection using Chimp Optimization Algorithm

The high prevalence of breast cancer in women has increased dramatically in recent times. Physician’s knowledge in breast cancer diagnosis and detection using computerized algorithms for extraction and segmentation of features can help. Image segmentation is a critical component of image analysis that has a direct impact on the quality of the results. This article presents Kapur’s entropy-based multilevel thresholding using Chimp Optimization Algorithm (ChOA) to estimate optimal values for the lesion segmentation of breast DCE-MRI. An improved ChOA is also developed by incorporating Opposition based-learning (OBL) in it, termed as ChOAOBL, and applied to solve the same problem. The proposed methods are evaluated using 200 Sagittal T2-Weighted fat-suppressed DCE-MRI images of 40 patients. The proposed methods are compared with Improved ChOA (IChOA), Particle Swarm Optimization (PSO), Multi-verse Optimizer (MVO), Slime Mould Algorithm (SMA), Arithmetic Optimization Algorithm (AOA), Tunicate Swarm Algorithm (TSA), Multilevel Otsu Threshold (MLOT), Conventional Markov Random Field (CMRF), Hidden Markov Random Field (HMRF), and Improved Markov Random Field (IMRF). The high sensitivity, accuracy, and Dice Efficient Coefficient (DSC) level of the proposed ChOA-based method are achieved at 90.75%, 98.24%, and 87.09% respectively. The accuracy value of 99.02%, sensitivity 95.73%, and DSC 93.25% are achieved using another proposed ChOAOBL-based segmentation method. The results are analyzed using a one-way ANOVA test followed by Tukey HSD, and Wilcoxon Signed Rank Test. We have also analyzed the overall performance using Multi-Criteria Decision Making based on accuracy, precision, specificity, F-measure, sensitivity, false-positive rate, Geometric-Mean (G-mean), and DSC. The proposed methods outperform other compared methods, according to both quantitative and qualitative outcomes.