Cuckoo lévy flight with otsu for image segmentation in cancer detection

• Main Author: Subki, Luqman
• Format: Thesis
• Language: English
• Published: 2019
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/98284/1/LuqmanSubkiMSC2019.pdf; http://eprints.utm.my/id/eprint/98284/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:144594
• Institution: Universiti Teknologi Malaysia

Detecting cancer cells from computed tomography (CT), magnetic resonance imaging (MRI) or mammogram scan images is a challenging task as the images are black and white and the neighbouring organs tend to be separated by edges with smooth varying intensity. On top of that, medical images segmentation is challenging due to the presence of weakly correlated and ambiguous multiple regions of interest. A few bio-inspired algorithms were developed to efficiently generate optimum threshold values for the process of segmenting such images. Their exhaustive search nature makes them computationally expensive when extended to multilevel thresholding, thus, this research is keen to solve the optimum threshold problems. This research propose an enhancement of image segmentation algorithms based on Otsu’s method by incorporating Cuckoo Search (CS) method for Lévy flight generation while simultaneously modifying and optimizing it to work on CT, MRI or mammogram image scanners, specifically to detect breast cancer. The performance of the proposed Otsu’s method with CS algorithm was compared with other bio-inspired algorithms such as Otsu with Particle Swarm Optimization (PSO) and Otsu with Darwinian Particle Swarm Optimization (DPSO). The experimental results were validated by measuring the peak signal-to-noise ratio (PNSR), mean squared error (MSE), feature similarity index (FSIM) and CPU running time for all cases investigated. The proposed Otsu’s method with CS algorithm experimental results achieved an average of 231.52 of MSE, 24.60 of PNSR, 0.93 of FSIM and 3.36 second of CPU running time. The method evolved to be more promising and computationally efficient for segmenting medical images. It is expected that the experiment results will benefit those in the areas of computer vision, remote sensing and image processing application.