Lung Nodule Classification of CT images Using Channel and Spatial Attention CNN with Bayesian Optimization

Lung cancer is the highest cause of mortality among cancers related diseases. Early diagnosis and treatment of lung cancer patients can improve their survival rates. Usually, for patients' diagnosis, Computed Tomography (CT) images are manually diagnosed by radiologists, which is a huge burden to them and sometimes leads to inaccuracy. Deep learning techniques have been proved as a popular and influential method in many medical imaging diagnosis areas. This paper proposes an automated computer-aided diagnosis (CAD) algorithm that uses the channel and spatial attention Convolution Neural Network (CNN) with Bayesian optimization to classify benign and malignant lung nodules from chest CT scans. We applied Bayesian optimization to determine the optimal deep CNN architecture by searching hyperparameters values such as number of training epochs, training batch size, number of layers, filter size, and number of filters. We validated the proposed network on CT scan from The Lung Nodule Analysis 2016 (LUNA16) public dataset, which comprises 888 CT scans. Experimental results show that the proposed algorithm gives accuracy, sensitivity, specificity, and ${F}{1}-$ score of 9S.2l%, 96.95%, 99.14%, and 9S.03%, respectively. We found that the proposed system obtained competent results compared with the network algorithms in the literature review.