Lung Cancer Diagnosis via Gabor Filters and Convolutional Neural Networks

In our study, we present a lung cancer diagnosis system to classify the cancerous and healthy lung images using the latest deep learning techniques. Most of the studies have been proposed so far are based on Convolutional Neural Networks (CNN). Our model uses CNN deep learning model which aims to increase the accuracy rate by passing the images through an image pre-processing stage before training them with deep learning. In this stage, Gabor Filter method was applied with different orientation values to the images. Then the modified images were given to the deep learning architecture. Gabor Filter transform was provided with the orientation values of 60 and 90 separately for comparison. The accuracy value in case the orientation is 60 is 96.48% and the loss rate is 23.69%. On the other hand the accuracy value in case the orientation is 90, is 98.37% and the loss rate is 8.35%. The high accuracy results of the proposed study could enable experts in the field to make more accurate and faster disease detection. Thus, it is expected that the patient treatment initiation process would be minimized.