IANNwTF Final Project: Comparison of CheXNet and ResNet binary classification performance on chest X-ray images

Pneumonia is a medical term describing the infection of the lungs that is caused either by bacteria or viruses. Despite being seemingly treatable, the long-term consequences of the disease are, in fact, grave. As such, it can be claimed that early diagnosis increased the likelihood of improved long-term health outcomes, and prevents early mortality.

Recent advances in deep learning have provided evidence that neural networks could be successfully used as an alternative to a diagnosis made by medical professionals, which would alleviate the burden on healthcare specialists.

In this project, we have compared the performance of a modified DenseNet121, that is one of the state-of-the-art models in pneumonia detection (CheXNet) with an adapted residual neural network (ResNet). For that, we have reimplemented the model following the methods provided in the original study and then compared it with ResNet, which will both be trained on the Chest X-ray 2017 dataset on the binary classification task, differentiating between pneumonia-ridden and healthy individuals.

Link to the Chest X-ray 2017 dataset: https://data.mendeley.com/datasets/rscbjbr9sj/2

Link to the original paper: https://arxiv.org/pdf/1711.05225v3.pdf