Using Convolutional and Capsule Neural Networks for Coeliac Disease Diagnosis

The paper explores the potential of deep learning methods for precise and accessible coeliac disease detection. Addressing challenges such as limited datasets and methodological variability, we investigate model architectures and sampling strategies to enhance classification accuracy using histopathologi-cal images. We compare the performance of a convolutional neural network trained exclusively on histopathological image data, ResNet18 and VGG16 models trained using transfer learning, and a capsule neural network. Our findings highlight the crucial role of architecture and sampling, with ResNet18 outperforming others. Tailored sampling proved to be efficient in mitigating the negative effect of class imbalance on prediction performance.