Development of differential convolutional neural network based image recognition software for diagnosis of lung cancer from CT images

Lung cancer is one of the major causes of mortality in the world, as there are more deaths each year from lung cancer than from any other type of cancer. If detected and identified at the initial stage, the survival rate of a large number of patients can be increased. Therefore, it can improve the accuracy of classification using the latest techniques of machine learning in the field of medical image processing in order to achieve a convenient and instantaneous result. In this work, we use a new convolution technique called Differential Convolution and an updated error back propagation algorithm. DCNN aims to transfer feature maps containing directional activation differences to the next layer. This implementation takes into account the idea of how convolved features change over the feature map. This feature improves the classification performance without changing the number of filters. proposes a differential convolutional neural network model to classify different types of lung cancer, including abnormal and normal computer tomography (CT) images. To train and test the performance of this model, a dataset of 1,000 lung computed tomography (CT) images including abnormal and normal images is used. Experimental results show that the proposed model achieves an accuracy of 97.25%. This study shows that the proposed differential CNN model can be used to facilitate automatic classification of lung cancer.