Polyp diagnostic with heuristic algorithm techniques on small bowel images

The small bowel is the longest organ of the digestive tract, starting after the stomach and ending with the large bowel at the end of about 5-6 meters. In the small bowel which has the most processing in the digestive system, the rate of digestion is 90%. Also it is the main center for digestion and absorption, the largest endocrine organ of the body and the biggest part of the immune system. While the small bowel has such important functions, the proportion of polyps rooted in any layer of the small bowel and pushed outwards in the form of tumors is much lower than the other organs of the digestive track. For this purpose, this study was started to detect polyps in the small intestine and images were pre-processed using various filters in the matlab program. Subsequently, the images were digitized using LBP feature extraction method in matlab program and the results were recorded in a file in excel format. After excel file is converted to csv file with the help of converter program, this file is opened in weka program. The data in the file were classified by Support Vector Machines, Random Forest Algorithm, Decision Tree Algorithm and Naive Bayes algorithm and achieved success rates of 70%, 71%, 63% and 67% were obtained respectively. Afterwards, these data were examined in detail by using PCA feature extraction method and the data were re-classified by the algorithms after being cleared of overlaps and conflicts. Results showed success rates of 73%, 73%, 68%, and 69% respectively. On the other hand, being sparse polyp rate seeing in small bowel, cause small bowel images with polyp be more less than small bowel images without polyp. Unbalanced data that adversely affect the detection rate of polyps were made more stable by SMOTE method and reclassified by algorithms. Results showed success rates of 79%, 76% ,72% 68%, respectively. These final success rates are planned to provide support for doctors in deciding whether polyps are present in the small bowel. In addition, besides the traditional algorithms, data has been reclassified and the success rate has increased to %83 within the scope of Resnet Architecture, which has been used recently and based on deep learning.