Kidney Cancer Detection from CT Images by Transformer-Based Classifiers

Convolutional neural networks (CNNs) have been used as standard deep neural networks in medical image analysis owing to their ability to automatically extract high-level features from training images. However, detecting kidney cancer from a wide variety of abdominal computed tomography (CT) images remains challenging. Recently, Transformer-based models have demonstrated higher predictive performance in computer vision tasks. Thus, in this study, we applied two Transformer-based models named the Vision Transformer (ViT) and the Swin Transformer to detection of kidney cancer from various types of abdominal CT images. Our experimental results show that our transformer-based models generally performed better than conventional CNNs, such as VGG-16 and ResNet-50, in term of detection accuracy across various types of CT images.