RETRACTED: Lung Nodule Classification Using Deep Learning Algorithms and Feature Fusion

Abstract

Today, lung cancer is the major cause of cancer death. Doctors use the CT scan and manual method for lung nodule detection and diagnosis. However such methods fail to correctly classify benign and malignant nodules. For early, automatic, and accurate detection and classification of lung nodule malignancy, a good lung nodule classification method is required. To solve this problem in the proposed system implementation, feature fusion and DL algorithms such as VGG16 and ResNet50 are used. In this system, two separate architectures, such as Architecture 1 and Architecture 2, are used for finding different features separately regarding the existing IQ-OTHNCCD lung cancer image dataset and real clinical lung CT images. Architecture 1 contains the VGG16 algorithm, and Architecture 2 contains the ResNet50 algorithm, which trains the system. The features obtained from each architecture are fused using the concatenation fusion algorithm. The concatenation fusion algorithm is used to improve classification accuracy and performance. After feature fusion, the softmax classifier layer is used for actual nodule classification. The overall system accuracy achieved is 98.08 percent, which is better than other existing systems.