Taguchi Design for Setting EHO Variants Parameters: Application in Brain Image Segmentation Using HMRF

Approximation methods are employed to remedy the significant computational time required by exact methods. They provide an approximate solution, close to the optimum, for complex problems (NP-hard problems). The performance guarantees of these methods vary depending on the chosen parameter values. However, selecting parameter values that yield good performance is a challenging task. Since the 1950s, Genichi Taguchi, a Japanese engineer, has developed robust design techniques to improve the quality of manufactured goods. Recently, these techniques have found applications in various domains, including engineering and biotechnology. In this paper, we investigate the application of the Taguchi design for the first time to simplify the process of selecting suitable parameter values for Elephant Herding Optimization (EHO) variants. These later are recent and known by their simplicity and ease implementation. A study case of using EHO variants is the segmentation of Brain Magnetic Resonance (MR) images using Hidden Markov Random Fields (HMRF) to help physician get the right decision. HMRF is a powerful model widely used for segmenting brain MR images which can be formulated as a problem of minimization an objective function. Subsequently, we compare the obtained results with those of well-known brain images’ segmentation tools such as FSL (the Functional Magnetic Resonance Imaging of the Brain Software Library) to demonstrate the usefulness of the Taguchi design. The quality of the segmentation is measured and tested using the Dice coefficient criterion and established on BrainWeb and IBSR images. Our findings indicate that the Taguchi design is effective, and Sine Cosine EHO (SCEHO) and Enhanced EHO (EEHO) methods yield excellent results based on the conducted tests.