Covid19-Detection

The dataset we use in this project is COVID-19 Radiography Database.
Download the dataset, unzip and put it under /input. You can specify the dataset path in func covidata.py/readData.

Structure of this project

• covidata.py: Read data and crate datasets and dataloaders
• evaluation.py: Script for trained model evaluation
• simplecnn.py: A simple CNN model
• transferlearning.py: Major code for model creation, transforms and set parameters to update
• utils.py: func train, test etc.
• vggmodels.py: VGG11 and VGG16 implementation

How to start training

specify the model_name in transferlearning.py To train the network, run

python transferlearning.py

These 4 files will be automatically saved during the training process:

• saved.pt: the best model
• record.csv: training loss and accuracy on test set of every epoch
• Xtest.npy and ytest.npy: test dataset

Result

AlexNet

	COVID	Normal	Viral Pneumonia
COVID	681	42	0
Normal	44	2005	2
Viral Pneumonia	2	35	220

VGG11-BN

	COVID	Normal	Viral Pneumonia
COVID	713	12	2
Normal	61	1970	7
Viral Pneumonia	4	17	245

VGG16

	COVID	Normal	Viral Pneumonia
COVID	670	58	1
Normal	6	2042	6
Viral Pneumonia	2	19	227

InceptionV3

	COVID	Normal	Viral Pneumonia
COVID	679	57	2
Normal	17	1996	2
Viral Pneumonia	7	59	212

ResNet18

	COVID	Normal	Viral Pneumonia
COVID	667	63	1
Normal	11	2026	5
Viral Pneumonia	2	33	223

ResNet50

	COVID	Normal	Viral Pneumonia
COVID	673	62	1
Normal	9	2016	4
Viral Pneumonia	6	38	222

VGG16-Transfer Learning

	COVID	Normal	Viral Pneumonia
COVID	597	106	0
Normal	14	2036	7
Viral Pneumonia	0	13	258

ResNet50-Transfer Learning

	COVID	Normal	Viral Pneumonia
COVID	556	147	4
Normal	67	1976	5
Viral Pneumonia	9	52	215