An improved capuchin search algorithm optimized hybrid CNN-LSTM architecture for malignant lung nodule detection

Abstract

For the early diagnosis of lung cancer, radiologists assisted computer-aided detection (CAD) systems are used. The false-positive reduction (FPR) is important in feature representation and classification based on lung nodule CAD. The region of interest (ROI) in lung computer-aided detection comprises an extra imbalance between negative and positive samples, as well as false positives. To tackle image recognition challenges, specific machine learning or deep learning models are utilized in the existing research. This study presents novel and effective methods for classifying lung nodule abnormalities. In the original computed tomography (CT) image, the binary operation is done first for pre-processing. The lung nodules are then located using an entropy-based K-means clustering approach, and these nodules are segmented using an automated active contour level set. Finally, the Improved Capuchin Search Algorithm (ICSA) optimized hybrid convolutional neural network (CNN) based long and short term memory (LSTM) is used to classify abnormalities of lung nodules into Juxtapleural pulmonary nodules, Juxtavascular pulmonary nodules, Ground-glass opaque (GGO) pulmonary nodules, and Small pulmonary nodules categories. The Opposition based learning and chaotic local search strategy are used in the ICSA algorithm to minimize the complexity of the hybrid CNN-LSTM architecture by optimizing the hyperparameters. The overall pulmonary nodule identification accuracy is improved and it is measured using different metrics such as accuracy, sensitivity, and precision. F1-score, dice, Jaccard, and Hausdorff. The simulation results show that the proposed method outperforms the existing state-of-the-art methods.