Color and morphological features extraction and nuclei classification in tissue samples of colorectal cancer

Cancer is an important public health problem and the third most leading cause of death in North America. Among the highest impact types of cancer are colorectal, breast, lung, and prostate. This thesis addresses the features extraction by using different artificial intelligence algorithms that provide distinct solutions for the purpose of Computer-AidedDiagnosis (CAD). For example, classification algorithms are employed in identifying histological structures, such as lymphocytes, cancer-cells nuclei and glands, from features like existence, extension or shape. The morphological aspect of these structures indicates the degree of severity of the related disease. In this paper, we use a large dataset of 5000 images to classify eight different tissue types in the case of colorectal cancer. We compare results with another dataset. We perform image segmentation and extract statistical information about the area, perimeter, circularity, eccentricity and solidity of the interest points in the image. Finally, we use and compare four popular machine learning techniques, i.e., Naive Bayes, Random Forest, Support Vector Machine and Multilayer Perceptron to classify and to improve the precision of category assignation. The performance of each algorithm was measured using 3 types of metrics: Precision, recall and F1-Score representing a huge contribution to the existing literature complementing it in a quantitative way. The large number of images has helped us to circumvent the overfitting and reproducibility problems. The main contribution is the use of new characteristics different from those already studied, this work researches about the color and morphological characteristics in the images that may be useful for performing tissue classification in colorectal cancer histology