Enhancing lung cancer detection and classification using two-stream conditional GANs for accurate computed tomography scan image analysis

Lung cancer is a leading cause of cancer-related death due to its late detection. The existing diagnostic methods face some limitations like low sensitivity, high false-positive rates, time-consuming manual analysis, which hinder the early lung cancer detection. To overcome these limitations, an Enhancing Lung Cancer Detection and Classification Using Two-Stream Conditional GANs forAccurate Computed Tomography Scan Image Analysis (LCDMIA-TCGAN) is proposed in this paper. Initially, input data is collected from LIDC-IDRI dataset. The CT images are pre-processing using the Adaptive Tracking Dual Nested Kalman Filter (ATDNKF) to effectively remove noise. The key morphological features like Area, Perimeter and Eccentricity are extracted using the Lifted Euler Characteristic Transform (LECT). Furthermore, a Two-Stream Conditional Generative Adversarial Networks (TCGAN) is employed to categorize CT images into benign, malignant and normal classes. TCGAN often lacks effective optimization techniques to identify optimal parameters, therefore, the Educational Competition Optimizer (ECO) is introduced to optimize the TCGAN model, thereby enhancing classification accuracy by reducing computational time. The experimental results demonstrate that the proposed method achieves improvements of 29.08 %, 30.70 %, and 16.26 % in accuracy, and 29.78 %, 27.87 % and 25.78 % in recall evaluated with existing techniques.