Image feature extraction for colorectal cancer cells classification / Mohd. Yamin Ahmad

Manual screening of colorectal biopsy tissue under microscope to conform the presence of cancerous cell is difficult, arduous and time consuming. The criteria in diagnosing colorectal cancer cell are gland shape and nucleus size. In this study, we proposed a method of image pre-processing to extract the important feature of colorectal tissue images. Images captured under microscope may vary in colour brightness due to different H&E stain concentration and the size of biopsy tissue. To overcome this problem a method using HSV colour model to remove element outside the area of nucleus is used. A novel method named Pixel Mask Analyzer is proposed to clean the image and remove noises. Meanwhile, the gland boundary tracking and segmentation is proposed to extract the gland shape. By using the result of gland tracking, nucleus size that forms the glands are measured. By combining result of gland shapes and nucleus size, the image classification is performed. The result shows that classification achieves 96.9% accuracy by using the proposed methods. With the high accuracy results and findings of this study, it is hope that the study can contribute a very substantial amount of outcomes that would greatly benefit the research areas especially in image processing and classification of colorectal cancer.