Colorectal Cancer Classification using PCA and Fisherface Feature Extraction Data from Pathology Microscopic Image
Fajri Rakhmat Umbara, Adiyasa Nurfalah, The Houw Liong
Graduate school of Telkom Institute of Technology, St. Telekomunikasi num. 1, Bandung, Indonesia
Colorectal Cancer, Feature Extraction, Fisherface, PCA, Random Tree Algorithm
Colorectal cancer is the one of variant cancer which can kill people on this
earth. World Health Organization, from their website wrote about 608,000
people can get killed every year because of it. The variant of colorectal
cancer such as lymphoma and carcinoma strikes colon from the inside and
outside. Lymphoma can be found in white corpuscle and attack colon
through lymphocytes, whereas carcinoma can attack the outer layer of colon.
Early detection is needed to decrease the number of death because of this
The study about colorectal cancer is to classified lymphoma, carcinoma, and
normal colon. It is doing by using 198 pathology microscopic images data
from Hasan Sadikin Hospital in Bandung, Indonesia. Feature extraction using
PCA and Fisherface and each generate 2, 5, 10, 50, 100 features. The study
compared these two methods and using WEKA to testing the accuracy.
Using 10 folds cross-validation and 3 different classifier in WEKA such as
Random Tree, Multi Layer Perceptron, and Naïve Bayes, Fisherface has
capability for classified colorectal cancer around 84% – 100% for accuracy. It
came from almost all features. Difference result is much visible in PCA.
From this result, Fisherface is better than PCA for feature extraction.
Copyright © 2013 Information Systems International Conference.
All rights reserved.