Enhanced Robustness in Lung Cancer Classification from CT Images Using Noise-Resistant Dynamic Integral Neural Networks

The lung tumor histology is crucial in respect to prognosis and the treatment, in which standard represented by tissue biopsy is used. But modern developments in deep learning point to the fact that radiologic data such as computed tomographs can be of great value in the characterization of diseases. This work uses Computed Tomography (CT) images originally taken from the Shenzhen hospital set with added noise reduced by the Modified Hamilton Filter. Segmentation is also done using the Deep Fuzzy Curriculum Clustering and extract only the lung image regions of interest out from the whole images. With the help of the proposed Hunger Game Search Optimization Algorithm (HGSOA), the efficiency of the Anti-Interference Dynamic Integral Neural Network (ADINN) framework is improved for lung cancer classification. ADINN-LC-CTI performs much better than the previous methods, LCD-CT-CNN, HSC-LCD-CT, and ALC-DNN by up to 34.78%, 22.89%, and 26.78% gains in accuracy, up to 29.82%, 17.47% and 21.66% gains in precision, and up to 21.78%, And those findings prove that ADINN-LC-CTI has a high potential to become an effective and reliable tool for the complete computer-aided lung cancer classification, thus giving a landmark contribution in increasing diagnostic accuracy with the minimal use of invasive diagnostic procedures.