Custom Lightweight CNN and Data-Efficient Models for Efficient and Fast Convergence in Lung Cancer Classification

Considering that lung cancer is the primary cause of cancer-related death across the globe, there are significant differences in patient outcomes as well as the subsequent clinical management if early detection, prediction, and diagnosis are done correctly. Machine learning techniques, particularly some studies, have received significant attention due to their effectiveness and accuracy in predicting, diagnosing, or treating lung cancers. This research presents a new direction using the data-efficient image transformer (DeiT). The paper also evaluates a custom lightweight convolutional neural network (CNN) and several other pre-trained models.The custom lightweight CNN obtained testing accuracies of 99.71%. DeiT appeared to be very efficient as it successfully achieved 99.89% testing accuracy with almost 16 times faster convergence rate than other models, such as ResNet50, InceptionV3, VGG16 and many more.