Averting from Convolutional Neural Networks for Chest X-Ray Image Classification

This paper attempts to survey newer approaches that do not use convolutional neural networks (CNNs) conventionally to the evolving field of medical image classification. While analyzing, firstly, an all feed-forward architecture MLP-Mixer and secondly, the inverted convolutional kernels coined as Involution with the baseline ResNets, both models yield comparable results in detecting Covid19 and pneumonia using Chest X-ray images. On top of that, merging Involution kernels into ResNet architectures can produce promising performance while training on roughly 40% fewer parameters. This paper further compares these two architectures with various CNN-based models. We hope this survey further helps the research community to utilize the capabilities of these newly introduced architectures in the medical field. [Code: https://github.com/Vrushank264/Averting-from-CNNs]