Identification of cancerous microscopic colonic images using neural networks

• Main Authors: Gan Lim, Laurence A., Naguib, Raouf N. G., Dadios, Elmer P., Avila, Jose Maria C.
• Format: text
• Published: Animo Repository 2008
• Subjects: Colon (Anatomy)—Cancer—Imaging; Neural networks (Computer science); Bioimaging and Biomedical Optics; Computer Engineering
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/5860
• Institution: De La Salle University

Research has been undertaken over the past two decades in an effort to automate cancer diagnosis. Investigations in the classification of microscopic images of colonic mucosa have shown that textural features derived from grey-level co-occurrence matrices (GLCMs) are very useful. In this paper, the results of applying multi-layer perception (MLP) with back propagation learning in classifying microscopic colonic images are presented. Prior to the application of the MLP, the images were classified by a human expert according to three (3) categories, namely: normal, adenomatous polyp, and adenocarcinoma or cancerous. The images were sorted in order to produce 25 images for each category, totalling 75 images in all. Fifteen (15) images from each classification were used in the training of the network while the remaining 10 images were subsequently used for the validation and testing of the trained neural network.