Amalgamation of Optimization Algorithms with Driven Deep Learning for Severity Level Classification of Lung Cancer and Survival Timeline Prediction of Lung Cancer

Abstract

Lung cancer is one of the most commonly diagnosed deadly disease. It is classified as small and large cell carcinoma, adenocarcinoma, and cell carcinoma. However, accurate lung cancer discovery with severity levels represent major concern faced by several techniques. This research devises three contributions that relies on lung cancer detection, severity level detection and survival time prediction. The first contribution is developing a technique for lung cancer detection. The goal is to develop an automatic discovery scheme for detecting the lung cancer based on optimization driven deep model. Here, Salp-Elephant Herding Optimization Algorithm based Deep Belief Network (SEOA-DBN) is proposed for lung cancer detection. The Deep Belief Network (DBN) training is carried out with SEOA, and is newly devised by combining Elephant Herding Optimization (EHO) and Salp Swarm Algorithm (SSA). The second contribution is to devise a technique for evaluating the levels of severity considering Computed Tomography (CT) images. In this technique, Shape Local Binary Texture (SLBT) is adapted which is obtained by integrating Local Directional Pattern (LDP) and Linear Binary Pattern (LBP). The SLBT feature is obtained from nodules. The features extracted are fed to the Adaptive-SEOA-DBN, which is the incorporation of Adaptive-SEOA in DBN for effective training of the model parameters. The proposed Adaptive-SEOA is developed by combining the self-adaptive concept in the SEOA. Finally, severity level classification is done to predict the severity level of the patient affected with lung cancer. The third contribution is to devise EHMO-based Deep ResNet for predicting the survival time suffering from adenocarcinoma cancer. For lung cancer detection, the developed SEOA based DBN classifier offered enhanced performance with highest accuracy of 96 %, highest TPR of 0.96, smallest False Positive Rate (FPR) of 0.05, and smallest False Discovery Rate (FDR) of 0.0002 respectively. For severity level prediction, the proposed Adaptive-SEOA-DB