A novel context-sensitive attitude entropy-based multiclass segmentation method for brain MR images using enhanced flow directional algorithm
Manoj Kumar Naik,Bibekananda Jena,Rutuparna Panda,Aneesh Wunnava,Ajith Abraham
DOI: https://doi.org/10.1007/s11042-024-19461-9
IF: 2.577
2024-06-19
Multimedia Tools and Applications
Abstract:Computer vision techniques, aided by artificial intelligence, in medical image analysis, are currently encouraged for precise and speedy diagnosis of brain disorders. The brain MR image shows an intensity variation, which attracts multiclass segmentation for a meaningful representation. The existing entropy-based methods use the image histogram. Nevertheless, the 1D histogram-based technique has a low complexity. However, it lacks the contextual information. Further, these entropy-based methods suffer low accuracy when the gray value distribution is non-uniform. To facilitate a reliable and efficient thresholding process, a novel multiclass segmentation objective function is investigated along with an efficient optimizer to obtain the optimal threshold values. To bridge this research gap, we explore a context-sensitive attitude entropy (CSAE) based multiclass segmentation objective function, an energy curve is used together with the human decision-making ability. Context-sensitive attitude entropy (CSAE), in contrast to classical entropy, which considers each pixel separately, is a measure of uncertainty or randomness in an image considering the spatial relationships and context between adjacent pixels. This can be achieved by calculating the attitude entropy from the energy curve instead of the image's histogram. To maximize the CSAE, we suggest a new enhanced flow directional algorithm (EFDA), which improves the flow directional algorithm (FDA) by adding randomization to both the velocity and the non-performer agents. The experimental effort begins with the EFDA performance validation utilising 23 well-known classical and 10 CEC-C06 2019 benchmark functions, which produce better results than cutting-edge optimizers. The EFDA's supremacy over other optimizers is ensured by Friedman's mean rank and Wilcoxon's mean rank test. The suggested CSAE-EFDA method is tested using T2-weighted brain images from Harvard Medical School's AANLIB database and Brain MR images for tumor classification from the Kaggle dataset. The proposal outperforms most of the standard and well-known entropy-based thresholding techniques, which are also verified by the ANOVA statistical test. The satisfactory improvement of the proposed method in Average PSNR, SSIM, and FSIM makes it suitable for use in a range of imaging modalities, such as satellite, color, grayscale, and biological images.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering