Feature Extraction and Nuclei Classification in Tissue Samples of Colorectal Cancer

Abstract

Cancer is considered to be a major health risk and ranked third most important cause of death in the USA. The American Cancer Society (ACS) predicted that by the end of 2020, there will be close to 2 millions new cases and over half million deaths in the USA. In particular, colorectal, breast, lung, and prostate cancers are the most dangerous cancers. This paper aims at providing new solutions for Computer-Aided Diagnosis (CAD) of colorectal cancer, using feature extraction and machine learning algorithms. in this paper, four well-known machine learning techniques have been compared to classify tissue categories. That is, Random Forest, Naive Bayes, Multi-layer Perceptron and Support Vector Machine. In order to measure the performances of these algorithms, we have used Precision, recall and F1-Score. In particular, we have focused on the colors and morphological characteristics in the images and, how they can be useful to improve the classification and diagnosis of colorectal cancer. We believe that such an improvement represents a significant contribution to the state-of-the-art, in both quantitative and qualitative ways.