Feature extraction using infrared spectroscopy from blood samples related to colon cancer

Nowadays, cancer is one of the most common causes of death. There are many different types of cancer. One of these types is colon cancer. In Turkey, according to the latest published 2013 report of Cancer Department under The Ministry of Health, colorectal cancer type is located in the third place in terms of the incidence in men and women. One of the methods used in the determination of colon cancer is colonoscopy. Colonoscopy is a diagnostic method which is not highly acceptable to many people. It is known that there are lots of disease cases that patients do not take their colonoscopy process (due to shyness and/or being afraid, etc.); thus, their cases get developed and may result in death. Considering that there is no easy way to overcome the above-mentioned feelings, there is a need for simple pre-diagnostic method to show people the needed colonoscopy on the right time. Briefly, it is intended to be able to determine whether situation is risky or not by using a blood sample as a pre-research for people who do not accept a colonoscopy. Therefore, early treatment to a person with a cancer risk will be provided. In this way, when considering a question like "If I have colonoscopy surgery or not?" disease progression and early treatment delay will be prevented. In this thesis, a pattern recognition based on intelligent automatic method has been proposed using routine blood samples to determine the risk of colon cancer. A software was developed to distinguish between colon patients and healthy individuals from FTIR patterns of sign that obtained from their blood samples. The developed software is a novel pre-diagnosis method used before other conventional diagnosis methods and it has a high sensitivity and accuracy rate. With the developed software, information extraction has been done from sub-band signs by applying Wavelet and Wavelet Packet Transform to the FTIR signs. Feature vectors that are generated from these extracted information were classified by Neural Network. Thus, according to the proposed three methods, blood samples of 30 colon cancer patients and 40 healthy individuals were classified with the accuracy rate of %95,65 to %100. This thesis study was made with the permission of Firat University Research Ethics Board of the Non-invasive (25.03.2014/02).