Fractional Crank-Nicolson finite difference method for benign brain tumor detection and segmentation. (July 2020)

• Record Type: Journal Article
• Authors: Chandra, Saroj Kumar; Bajpai, Manish Kumar
• Is Part Of: Biomedical signal processing and control. Volume 60(2020)
• Publication Date: 2020-07
• Subjects: Fractional calculus -- Crank-Nicolson finite difference method -- Boundary based edge detection -- Region based image segmentation; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.102002
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel fractional diffusion method has been proposed in the present manuscript for benign brain tumor detection and segmentation. An unconditional Crank Nicolson finite difference scheme has been used to solve proposed fractional diffusion equation. Alternate direction implicit (ADI) approach has been used to solve Crank Nicolson finite difference scheme. The proposed method has been validated on numerical head phantom of size 1024 × 1024 × 1024 and BRATS dataset (For real brain tumor images). Abstract: Benign brain tumor is early stage of cancer in tumor development life cycle. Its detection is hard and most challenging task due to low variability with its surrounding non-cancerous tumor cells. Image segmentation is used as a primary tool in brain tumor detection algorithms to segment the tumorous region. It has been observed that the available methods such as region-based, watershed-based method, cluster-based method and contour and shape-based methods are not able to find such low-intensity variational regions (i.e. benign brain tumor). Current work proposes a novel fractional method for finding such a low intensity variational region. The proposed method uses alternate direction implicit finite difference scheme. The performance analysis has been done on three-dimensional numerical head phantom and BRATS dataset. Results obtained by the proposed tumor detection and segmentation method have been compared with the popular tumor detection and segmentation methods. Hausdorff distance, Jaccard similarity index and Dice coefficient have been used for quantitative comparative performance analysis.