Improving the diagnosis of pancreatic cancer based on image processing and machine learning techniques

摘要

In the current age, pancreatic cancer is one of the worst forms of cancer. The complications of pancreatic include five types of pancreatitis, benign tumors, malignant tumors, benign cysts and malignant cysts. This cancer has a few clinical symptoms than other cancers. Also, if not treated in a timely manner, it also causes other organs of the body and the patient' chance of survival is greatly reduced. One of the ways to detect this disease is to use CT scan images. But the appearance of pancreatic complications is very different in a similar category, and their tissue is very similar to healthy abdominal tissues. For this reason, it's very difficult to identify the range of complications. Materials and Methods: In this study, the data contained 151 CT scan images. These images are divided into five classes of pancreatitis, malignant tumors, benign tumors, malignant cysts, benign cysts and a healthy class. The pancreatic complications are varied and different, if the diagnostic system is based on simple experts; the possibility of achieving high detection accuracy is not possible. According to the results of this study, lonely no classification can detect all diseases and combining these methods is the best option. Therefore, in this study we have achieved high accuracy in prediction (96.09%) by combining the perception, convolution and SVM neural networks.