Chest_XRay_Disease_Detection

COVID and Lung Disease Detection from Xray images

• Intro: Due to the quick spread of COVID-19, many patients were unable to access required treatments in time. Therefore, we are motivated to develop deep learning models that can quickly and accurately classify lung diseases and address them with the appropriate treatments in a timely matter.
• Goal: Find the best model with lowest misclassification rate for certainly contagious disease such as COVID and Viral Pneumonia, and highest test set accuracy. Utilize transfer learning on Pytorch pretrained Convolutional Neural Networks (CNN): ResNet, VGG, AlexNet, and a benchmark model with Logistic Regression to predict correct class labels Covid-19, Viral Pneumonia, normal(healthy) and Lung Opacity(non covid, non viral pneumonia infections) from X-ray images.
• Tech used: Pytorch, pandas, numpy, opencv, matplotlib
  

Results

(1) Grad-CAM visualization on CNN models

Model	Original COVID	AlexNet	VGG19_bn	ResNet50
Layer	None	Layer 5	Layer 16	Layer 49
Grad-CAM

(2) Train, Test Accuracy

Model Name \ Evaluation	Train Accuracy	Test Accuracy
AlexNet	97.87 %	93.86 %
VGG19 with batch norm	98.12 %	94.33 %
ResNet 50	98.59 %	95.46 %
Logistic Regression	60.84 %	62.24 %

(3) Multi-class: Precision, Recall, F1 score

Class label \ Model Name	Logistic Regression	AlexNet	VGG19_bn	Resnet50
COVID	25.86 / 28.75 / 27.23	96.55 / 94.65 / 95.59	97.7 / 94.44 / 96.05	98.28 / 97.71 / 97.99
Lung Opacity	53.85 / 70.59/ 61.09	88.78 / 94.22 / 91.42	90.38 / 94.31 / 92.31	90.71 / 96.42 / 93.48
Normal	88.04 / 67.14 / 76.19	95.65 / 93.17 / 94.39	95.36 / 93.96 /94.65	97.43 / 93.99 / 95.68
Viral Pneumonia	0.0 / NA / NA	96.97 / 95.52 / 96.24	96.21 / 96.95 / 96.58	95.45 / 96.92 / 96.18

Dataset

• COVID-19 Radiography Database (COVID-19 Chest X-ray Database)
• M.E.H. Chowdhury, T. Rahman, A. Khandakar, R. Mazhar, M.A. Kadir, Z.B. Mahbub, K.R. Islam, M.S. Khan, A. Iqbal, N. Al-Emadi, M.B.I. Reaz, M. T. Islam, “Can AI help in screening Viral and COVID-19 pneumonia?” IEEE Access, Vol. 8, 2020, pp. 132665 - 132676. Paper link
• Rahman, T., Khandakar, A., Qiblawey, Y., Tahir, A., Kiranyaz, S., Kashem, S.B.A., Islam, M.T., Maadeed, S.A., Zughaier, S.M., Khan, M.S. and Chowdhury, M.E., 2020. Exploring the Effect of Image Enhancement Techniques on COVID-19 Detection using Chest X-ray Images. Paper Link

Files

• jupyter notebooks:

  1. Logistic Regression
  2. AlexNet
  3. VGG19 with batch norm
  4. ResNet50

• Report: STAT_453_Report.pdf

• Presentation slide: group07.pdf

• Presentation recording: group07.mp4

• helper_functions: dataset, evaluation, train, plotting are written by Professor Sebastian Raschka

• helper_GradCAM: code borrowed from Kaggle Notebook by Debarshi Chanda