Polyp Detection in Endoscopy videos using cascaded classification

Cancer is one of the hazardous disease which cannot be cured when certain stage is reached. Early detection and treatment with disease has a good chance of recovery. Endoscopy is a process involved in analyzing the upper gastro intestine in search of abnormal tissues which is root cause of cancer. In endoscopy process, a long and tine tube with camera is inserted into body to witness the internals of the body. The endoscope is also used to perform some minor surgeries. The medical practitioner or expert uses this endoscope to analyze the internal tissues of intestinal tract, there may be a chance to skip the observation where some malignant polyp may be unidentified. To overcome said drawback, always a computer aided techniques are efficient along with medical practitioner's knowledge and experience. The HaaR based cascade classifier is a machine learning based classifier used for object detection. In this technique the trained is done for cascade classifier by using positive and negative images. Haar feature considers contiguous rectangular areas in the localization region to identify the object, the pixel intensities in each rectangular region are summarized and difference between the regions are calculated among these quantities. The proposed work is conducted on endoscopy videos to support computer aided system to assist medical practitioners in identifying and classifying the polyps. The research work is conducted on a dataset, which consists of sequence of 10 videos with 7894 total number of images. From the input dataset the training dataset has been derived. The positive images which represent the polyp identification are derived. The negative images representing background and no polyp region are derived. The Haar-like classifier is trained with the said positive and negative images for certain amount of time and then same dataset is given as input to measure the efficiency of classifier.