AUTHOR=Ramana Kadiyala , Kumar Madapuri Rudra , Sreenivasulu K. , Gadekallu Thippa Reddy , Bhatia Surbhi , Agarwal Parul , Idrees Sheikh Mohammad TITLE=Early Prediction of Lung Cancers Using Deep Saliency Capsule and Pre-Trained Deep Learning Frameworks JOURNAL=Frontiers in Oncology VOLUME=12 YEAR=2022 URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.886739 DOI=10.3389/fonc.2022.886739 ISSN=2234-943X ABSTRACT=

Lung cancer is the cellular fission of abnormal cells inside the lungs that leads to 72% of total deaths worldwide. Lung cancer are also recognized to be one of the leading causes of mortality, with a chance of survival of only 19%. Tumors can be diagnosed using a variety of procedures, including X-rays, CT scans, biopsies, and PET-CT scans. From the above techniques, Computer Tomography (CT) scan technique is considered to be one of the most powerful tools for an early diagnosis of lung cancers. Recently, machine and deep learning algorithms have picked up peak energy, and this aids in building a strong diagnosis and prediction system using CT scan images. But achieving the best performances in diagnosis still remains on the darker side of the research. To solve this problem, this paper proposes novel saliency-based capsule networks for better segmentation and employs the optimized pre-trained transfer learning for the better prediction of lung cancers from the input CT images. The integration of capsule-based saliency segmentation leads to the reduction and eventually reduces the risk of computational complexity and overfitting problem. Additionally, hyperparameters of pretrained networks are tuned by the whale optimization algorithm to improve the prediction accuracy by sacrificing the complexity. The extensive experimentation carried out using the LUNA-16 and LIDC Lung Image datasets and various performance metrics such as accuracy, precision, recall, specificity, and F1-score are evaluated and analyzed. Experimental results demonstrate that the proposed framework has achieved the peak performance of 98.5% accuracy, 99.0% precision, 98.8% recall, and 99.1% F1-score and outperformed the DenseNet, AlexNet, Resnets-50, Resnets-100, VGG-16, and Inception models.