Detecting and Visualising the Infectious Regions of COVID-19 in X-ray Images and CT scans Using Different Pretrained-Networks in Tensorflow 2.x.

The Corona Virus Disease 2019 (COVID-19) pandemic continues to have an upsetting effect on the health and well-being of the global population. A critical step in the fight against COVID-19 is effective screening of infected patients, with one of the key screening approaches being radiological imaging using chest radiography. It was found in early studies that patients present abnormalities in chest radiography images that are characteristic of those infected with COVID-19 as discussed in a recent [Paper]. Therefore, in this Repository I have tested different Pretrained Networks for the detection of COVID-19 cases from chest radiography images that is open source and available to the general public. Finally a Novel Method will be introduced to get the maximum of Positive Predicted Values from these Pretrained Networks.

Table of Contents

• Datset Description

  • Dataset 01-Chest XRays
  • Dataset 02-Chest XRays
  • Dataset 03-CT Scans

• Pretrained Networks for COVID-19 Detection and Visualization of Infectious Regions for Dataset 01-Chest XRays

  • VGG-16
  • VGG-19
  • DenseNet-121

• Pretrained Networks for COVID-19 Detection and Visualization of Infectious Regions for Dataset 02-Chest XRays

  • MobileNet-V2
  • DenseNet-169

• Pretrained Networks for COVID-19 Detection and Visualization of Infectious Regions for Dataset 03-CT Scans

Dataset Description:

Dataset 01-Chest XRays:

Dataset 01 is the first version introduced by the authors of the [Paper]. This dataset is comprised of a total of 5949 posetrior chest radiography images across 2839 patients. I have write a python script to categorize each X-ray image in the raw data folder to a separate Class as per corresponding label. The original data distribution was

Category	Number of Patients	Number of X-rays
0: Normal	1203	1583
1: Bacteria	931	2786
2: Viral	660	1504
3: COVID-19	45	76

I have split the original data to train/test folders with following distrubtuions. This categorized dataset can be downloaded at Here. I have also write a python Script to make the NumpyFiles for the train/test images but without any normalization so that one can normalize the data as per his/her need. These NumpyFiles can also be accessed at Here.

Category	Train	Test
0: Normal	1349	234
1: Bacteria	2540	246
2: Viral	1355	149
3: COVID-19	66	10

Dataset 02-Chest XRays:

Dataset 02 is the second version introduced by the authors of the [Paper]. One can follow the steps given Here to generate this dataset. The NumpyFiles for this data can also be accessed at Here. The distribution of these Numpy files are as followed.

Category	Train	Test
0: Normal	8751	100
1: Pneumonia	5945	100
2: COVID-19	229	31

Dataset 03-CT Scans:

Dataset 03 is a generated version of three different datasets. Two of these datasets were origginally developed for segmentation tasks and can be found at Here and one is developed for identification/diagnosis task and can be found at Here. One can follow this simple Script to generate this dataset.

Pretrained Networks for COVID-19 Detection for Dataset 01-Chest XRays:

The core idea behind the Pretrained Networks and Transfer Learning can be studied in detail at my another GitHubRepository at Here. I have used the following Pretrained Networks for COVID-19 Detection. Each Pretrained Network is also described with its train/test History and a Confusion-Matrix for better visualization of the Generalization of the Trained Network.

VGG-16:

Code: Google Colab Notebook

Statistical Results:

Pretrained-VGG16 Accuracy-Graph	Pretrained-VGG16 Loss-Graph

Pretrained-VGG16 Confusion Matrix	Pretrained-VGG16 ROC

Visualization:

VGG-19:

Implementation:

Code: Google Colab Notebook

Statistical Results:

Pretrained-VGG19 Accuracy-Graph	Pretrained-VGG19 Loss-Graph

Pretrained-VGG19 Confusion Matrix	Pretrained-VGG19 ROC

Visualization:

DenseNet-121:

Code: Google Colab Notebook

Statistical Results:

Pretrained-DenseNet121 Accuracy-Graph	Pretrained-DenseNet121 Loss-Graph

Pretrained-DenseNet121 Confusion Matrix	Pretrained-DenseNet121 ROC

Visualization:

Pretrained Networks for COVID-19 Detection for Dataset 02-Chest XRays:

MobileNet-V2:

Code: Google Colab Notebook

Statistical Results:

Pretrained-MobileNetV2 Accuracy-Graph	Pretrained-MobileNetV2 Loss-Graph

Pretrained-MobileNetV2 Confusion Matrix	Pretrained-MobileNetV2 ROC

Visualization:

DenseNet-169:

Code: Google Colab Notebook

Statistical Results:

Pretrained-DenseNet169 Accuracy-Graph	Pretrained-DenseNet169 Loss-Graph

Pretrained-DenseNet169 Confusion Matrix	Pretrained-DenseNet169 ROC

Visualization:

Pretrained Networks for COVID-19 Detection for Dataset 03-CT Scans:

Code: Google Colab Notebook

Statistical Results:

Pretrained-DenseNet121 Confusion Matrix	Pretrained-DenseNet121 ROC

Visualization: