LCD-Capsule Network for the Detection and Classification of Lung Cancer on Computed Tomography Images

• Published in: Multimedia tools and applications 2023-10, Vol.82 (24), p.37573-37592
• Main Authors: A.R., Bushara, R.S., Vinod Kumar, S.S., Kumar
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Classification; Computed tomography; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Deep learning; Image classification; Lung cancer; Machine learning; Medical imaging; Multimedia Information Systems; Neural networks; Pattern classification; Spatial data; Special Purpose and Application-Based Systems; Tomography
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-14893-1

Lung cancer is the second most prominent cancer in men and women, and it is also the leading cause of cancer-related mortality.If lung cancer is diagnosed early, when it is minuscule and has not spread, it is preferable to be adequately treated. A non-invasive low-dose Computed Tomography (CT) scan can detect abnormal patches in the lungs that could be cancerous. It is proposed that machine learning and pattern classification be used to identify and categorize lung cancer from CT scans. Pattern classification algorithms like deep learning can categorize input data into various classes based on the input’s characteristic features. A novel deep learning framework formulated by the encapsulation of a Convolutional Neural Network (CNN) and a Capsule Neural Network (CapsNet) called LCD-CapsNet, leveraging the capabilities of these networks to minimize vast amounts of data and achieve spatial invariance for lung cancer detection and classification using CT images is proposed. The primary objective of the proposed method is to create algorithms that would classify and examine images from a dataset to determine whether or not a patient had or posed a danger of developing lung cancer. The Lung Image Database Consortium (LIDC) datasets are utilized to assess this deep learning model, from which 4335 images were collected for the training and testing pipeline. The results demonstrated that LCD-CapsNet outperforms CapsNet, with an average Precision of 95 %, Recall of 94.5 %, F1-Score 94.5 %, Specificity 99.07 %, Area Under the Curve of 0.989, and Accuracy 94 % of benign and malignant data.