136

Bioinformatics of the Brain

A new threshold approach to segment brain tumors was proposed by Ilhan

and Ilhan [28]. There are three stages to the suggested approach. The first

stage is pre-processing, where the image was improved and made ready for

analysis using morphological and pixel subtraction operations. The second

stage is segmentation, where a novel thresholding technique was suggested

to distinguish the tumor region from the enhanced image. In the proposed

threshold method, average gray value, which was used to transform a grayscale

image into a binary image, was computed by dividing the sum of unique pixel

values—excluding zeros—to the count of unique pixel values. In the last stage,

they applied a median filter to remove the noise from the segmented image.

A study analyzing the performance of different edge detection techniques

on brain MRI images was carried out by Yıldız and Yıldız [61]. The study

revealed that the Roberts, Prewitt, and Sobel methods, when used with a

threshold value of 0.03, produced better results in comparison to other meth-

ods.

Pooja et al. [62] conducted a research on brain tumor detection, where they

examined the performance of a variety of segmentation techniques. The tech-

niques analyzed in their study encompassed threshold-based, edge detection,

region growing, watershed, and k-means segmentation. The proposed system

encompassed the following stages of skull stripping; preprocessing; segmenta-

tion; and comparative analysis. The results of the study indicated that region

growing and k-means segmentation techniques were more effective than the

other segmentation techniques.

Zotin et al. [19] suggested a methodology to determine the location of

tumor borders in MRI brain images. They utilized Fuzzy C-means clustering

method to segment the images and Canny edge detector to identify fine edges.

The performance of the suggested techniques was evaluated against Classic

Canny, Prewitt, Roberts, Sobel, and LoG methods. The proposed approach

obtained an average of 3–7% more accuracy.

Aslıyan and Atbakan [20] utilized a variety of methods like k-means, Fuzzy

C-means, Self Organizing Maps, Otsu, and hybrid k-means+Otsu methods to

identify the regions of tumors in brain MRI images. They proposed an algo-

rithm to identify and remove the skull from the image. The success of the

different segmentation systems were examined with and without the skull

stripping step. Hybrid k-means+Otsu method was reported as the most suc-

cessful with an accuracy rate of 94% in the skull-removed images and 84% in

the non-removed images.

Mahdi et al. [63] applied Sobel, Prewitt, Roberts and Canny edge detection

methods to brain, bone, and liver MRI images. Among the edge detection

methods, it has been stated that the Canny and Sobel edge detectors perform

better than the other techniques.

Traditional edge detection techniques and eight-direction Sobel edge de-

tection algorithm are compared by As and Gopalan [64] in their work and

identified that 8-Sobel is the most appropriate method to analyze the MRI

images of brain tumors.