Integrating Generative Adversarial Networks (GANs) with Deep Learning for Improved Lung Cancer Detection from CT Scans

Early and accurate diagnostic methods for lung cancer is still one of the main factors in regard to preventing the death of cancer patients in the entire world. Motivated by recent breakthroughs in deep learning applications of lung cancer detection, this paper proposes a novel architecture of using Generative Adversarial Networks with Convolutional Neural Networks for image analysis in computed tomography (CT) scan images. By integrating GANs, the generation of high quality synthetic medical images can be achieved that can be used to improve the training of the CNN models to help it to detect early stage lung cancer lesions that are hard to find with conventional approaches. We also exploit the advantage of data augmentation using the images generated from GANs, which helps to overcome the difficulties brought by the small amount of annotated datasets in medical imaging. In this work, we compare the proposed method with traditional deep learning models by applying it on a publicly available lung cancer dataset and evaluating the accuracy, sensitivity and specificity of the method. The results demonstrate the superiority of the proposed GAN enhanced deep learning based framework compared to the standard CNN based models in terms of performance in detecting the lung cancer. The advantage of clinical application and diagnosis of lung cancer at timely stage is also feasible by this approach.