Automatic lung cancer detection from CT image using optimized Robust Deformed Convolutional Neural Network with TriHorn-Net

• Published in: Expert systems with applications Vol. 276; p. 127124
• Main Authors: Lokanath Reddy, C, Prasad, M.V.D.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2025
• Subjects: Automatically Weighted Binary Multi-View Clustering; Robust Deformed Convolutional Neural Network; Second-Order Synchroextracting Wavelet Transform; Sub Aperture Keystone Transform Matched Filtering; TriHorn-Net; Wader Hunt Optimization Algorithm; SS; CNN; SOSWT; DTOA; DF; DNN; Wader Hunt Optimization Algorithm; CSViT-IMFO; MS; SAKTMF; WHOA; AW-BMVC; Robust Deformed Convolutional Neural Network; Second-Order Synchroextracting Wavelet Transform; MAE; CT; SNN; RDCNN; TP TN FP FN; AEALOA; TL; TriHorn-Net; Sub Aperture Keystone Transform Matched Filtering; Automatically Weighted Binary Multi-View Clustering
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2025.127124

Lung cancer is a leading cause of death for both men and women, requires accurate and early detection to improve treatment outcomes. The inability of traditional approach to handle intricate nodule formations, subpar segmentation methods, and low-quality CT images results in inaccurate predictions. To overcome these complications, Automatic Lung Cancer Detection from CT image using optimized Robust Deformed Convolutional Neural Network with TriHorn-Net (RDCNN-TriHorn-Net-WHOA-ALCD) is proposed. The input CT images, sourced from the Chest CT-Scan Images Dataset and Formatted and Augmented Chest CT-Scan images dataset, undergo preprocessing via Sub Aperture Keystone Transform Matched Filtering (SAKTMF) to reduce noise and improve image quality. Automatically Weighted Binary Multi-View Clustering (AW-BMVC) is used to segment the affected regions, and the Second-Order Synchroextracting Wavelet Transform (SOSWT) is used to extract spectral features. Classification is conducted using Robust Deformed Convolutional Neural Network (RDCNN) models across three strategies. In Strategy 1, RDCNN-TriHorn-Net with Wader Hunt Optimization Algorithm (WHOA) outperformed other models in detecting lung cancer. In Strategy 2 showed RDCNN-ResNeXt-50 with Adaptive Elite Ant Lion Optimization Algorithm (AEALOA) yielded better results, while Strategy 3 highlighted RDCNN-CoAtNet with Dipper Throated Optimization Algorithm (DTOA). The RDCNN-TriHorn-Net-WHOA-ALCD method achieved superior classification of lung CT images, like Large Cell Carcinoma (LCC), Adenocarcinoma, Normal, and Squamous Cell Carcinoma (SCC) in the Chest CT-Scan images dataset, as well as normal and Squamous in the Formatted and Augmented Chest CT-Scan images dataset. The proposed RDCNN-TriHorn-Net-WHOA-ALCD technique is implemented in Python. The effectiveness of the RDCNN-TriHorn-Net-WHOA-ALCD approach attains 13.67%, 27.55% and 14.67 dice similarity coefficient and 22.23%, 24.11% and 25.56% logarithmic loss compared with existing techniques respectively.