InceptionResNetV2 ve Sınıf Aktivasyon Haritaları ile Akciğer Kanserinin Tespit Edilmesi

Abstract

Lung cancer is a life-threatening disease. According to the World Health Organization, cancer is the disease with the highest mortality rate. This disease, which is quite insidious, does not show any symptoms in the early stages. If the disease is diagnosed correctly in the early stages, it can be treated. With computed tomography, masses in the lung region can be detected and diagnosed by experienced doctors. Convolutional neural network, which is one of the deep learning methods, is successfully applied in the detection of many diseases today. When class activation maps are trained with a convolutional neural network, distinctive regions of the image are colored according to their importance, so that the important regions for the target class can be determined. In this study, a total of 1197 lung images consisting of three classes obtained by computed tomography, class activation maps were trained with the InceptionResNetV2 convolutional neural network, and the important regions of the images were determined that the features of these regions were obtained. The Obtained features were classified using support vector machines and classified with an accuracy rate of 95.44%.