APPLICATION OF GENERATIVE ADVERSARIAL NEURAL NETWORKS FOR LUNG CANCER CT IMAGE SEGMENTATION

Abstract

Lung cancer remains a leading cause of cancer-related mortality, necessitating advancements in early detection and diagnostic tools. This study explores the application of Deep Convolutional Generative Adversarial Networks (DCGANs) to augment CT imaging datasets for lung cancer segmentation. Using a combination of local Kazakhstani and re-labeled LIDC-IDRI data, DCGAN generated realistic synthetic images, improving segmentation performance. The UNet model, evaluated with the DICE metric, showed enhanced accuracy, with scores improving from 0.3708 to 0.4191. While DCGAN demonstrates strong potential in addressing data scarcity, its high computational demands remain a significant challenge.