The project aims at building a robust CNN Architecture which is able to classify a given chest xray image into one of the three classes:
- Covid19
- Other Pneumonia
- Non Pneumonia
This project was carried out as a part of the project work of the course Deep Learning Foundations and Applications (AI61002), IIT Kharagpur under Prof. Debdoot Sheet sir.
The model used is relatively light weight. Weights of pretrained Resnet18(trained on ImageNet dataset) downloaded from Pytorch were used. These weights were further tuned during training.
Pretrained ResNet18(Pytorch, trained on ImageNet dataset) Input Dimensions:(channels = 3(RGB), 224, 224) Output Dimensions: (1000)
Fully Connected Layer 1 (in=1000, out=200)
Fully Connected Layer 2 (in=200, out=40)
Fully Connected Layer 3 (in=40, out=10)
Fully Connected Layer 4 (in=10, out=3)
A ReLU is placed after each Fully Connected Layer except the last layer which has a softmax.
Total Parameters = 11907815
The dataset used in training this model has been obtained from sources mentioned in the References section(1-7).
All images were converted to dimensions of (224x224) and into 3 channel RGB images. Data augmentation was carried out using techniques like colorjitter, random rotations (<15 degrees) and normalisation.
This model was trained on Google Colab using the provided GPUs. Due to hardware limitations, the training was done in an unorthodox fashion. The hardware available could handle only a dataset of the size of 500 images for training hence, at regular intervals the train loader had to be deleted and a new train loader was loaded with a fresh set of 500 images from the dataset. Framework : Pytorch Learning rate = 1e-3 Optimizer : Adam optimizer
[1] https://twitter.com/ChestImaging/status/1243928581983670272
[2] https://www.sirm.org/category/senza-categoria/covid-19/
[3] Irvin, Jeremy, Pranav Rajpurkar, Michael Ko, Yifan Yu, Silviana Ciurea-Ilcus, Chris Chute, Henrik Marklund et al. "Chexpert: A large chest radiograph dataset with uncertainty labels and expert comparison." In Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, pp. 590-597. 2019. (link: https://stanfordmlgroup.github.io/competitions/chexpert/ )
[4] Radiological Society of North America, RSNA https://www.kaggle.com/c/rsna-pneumonia-detection-challenge pneumonia detection challenge.
[5] Joseph Paul Cohen and Paul Morrison and Lan Dao, “COVID-19 image data collection”, arXiv:2003.11597, 2020 https://github.com/ieee8023/covid-chestxray-dataset.
[6] Linda Wang, Alexander Wong, Zhong Qiu Lin, James Lee, Paul McInnis, Audrey Chung, Matt Ross, Blake VanBerlo, Ashkan Ebadi, “FIgure 1 COVID-19 Chest X-ray Dataset Initiative”, https://github.com/agchung/Figure1-COVID-chestxray-dataset
[7] Kong, Weifang, and Prachi P. Agarwal. "Chest imaging appearance of COVID-19 infection." Radiology: Cardiothoracic Imaging 2, no. 1 (2020): e200028. https://pubs.rsna.org/doi/full/10.1148/ryct.2020200028
[8] Deep Residual Learning for Image Recognition, Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun, 2015