A novel deep learning approach for the detection and classification of lung nodules from CT images

In a conventional Computer-Aided Detection (CAD) system, complexity is seen in the classification procedure of Lung Nodule Detection (LND). Lower classification accuracy along with a high False-Positive Rate (FPR)is caused since the classification outcome extremely relies on the performance of every step in LND. The work proposed a new Deep Learning (DL) approach for detecting and classifying Lung Nodules (LNs) from Computer Tomography (CT) images to address these difficulties. Initially, the input lung image is pre-processed, and then the non-informatics blocks are removed using Step Deviation Mean Multilevel Thresholding (SDMMT). After that, the lung image’s contrast is enriched and the earliest event-Net classifier is utilized to detect the LN parts. From the identified LN portion, the features are retrieved and the important features are chosen using an optimization algorithm called Minkowski Distance-based Horse herd optimization Algorithm (MD-HHOA). The selected features are fed into the Crossover Swap-Displacement and Reversion-based Jaya-Weight Hinge Generative Adversarial Network (CSDR-J-WHGAN) classifier for classifying as nodule or non-nodule. This study utilizes publicly accessible Lung Image Database Consortium image collection (LIDC-IDRI) datasets.The experiential result shows that the proposed method attains 97.11% accuracy, 96.98% sensitivity, and 94.34% specificity for detecting nodules when compared with the existing methods.