A Deep Neural Network for Detecting Lung Cancer from CT Images

Lung cancer impairs the respiratory mechanism and has been one of the leading cause of death from cancer in today’s world. Early detection using a computed tomography (CT) scan of the lung can enhance the average survival rate from 14 to 49% among lung cancer patients. Therefore, artificial intelligence-based automated classification and detection systems are required to detect lung cancer in its early stages. For such automated classification, in this paper, we proposed a convolution neural network (CNN) for classifying the benign and malignant stages of lung cancer from CT images. CT images have less noise disturbance compared to MRI, X-Ray. To further enhance the quality of CT images, we use Gaussian filter, thresholding, “open” morphological operations, and dilation in the preprocessing stage and then passed the preprocessed images through proposed CNN. We perform extensive ablation studies on network architecture, image size, and activation functions by using a Kaggle lung cancer dataset which has 613 lung CT images. The experimental results of the proposed CNN shows that it can achieve 99.67% accuracy on the CT image based Kaggle dataset (https://www.kaggle.com/datasets/mohamedhanyyy/chest-ctscan-images).