In this paper, we propose a rapidly testing method which has a high productivity in a short time. In details, we will apply deep learning neural networks, e.g. ResNet50 and VGG19 to solve this problem. After that, we will proceed analysing pros and cons of those models for a thorough vision about applying artificial intelligence in COVID-19 rapid testing.
@article{detect-covid19-dl,
title={Detecting COVID-19 and Pneumonia with Chest X-Ray images using Deep Convolutional Neural Networks},
author={Vo, An and Pham, Tan Ngoc},
journal={Introduction to Computer Vision},
month={June},
year={2021},
url={https://github.com/DTA-UIT/Detect-COVID19}
}
In this research, we use COVIDx dataset [2] - which is a widely used dataset in recent research about COVID-19 nowadays. COVIDx Datset is a dataset synthesized from many a different source, which are in details: Cohen et al. [3], Chung [4], Chung [5], Radiological Society of North America [5],and Radiological Society of North America [6]. Additionally, this dataset also provides an image extension transfer tool: from .mri into .jpg. And the author moreover provide a code to support data pre-processing and getting rid of unnecessary part for synthesized data.
Method
We proposed using diagnostic imaging, which is an approach using Chest X-ray (CXR) image. This is due to its lower cost and faster testing time in comparison with Real-time Polymerase Chain Reaction (RT PCR) or Computed Tomography (CT) Image.
Model approach
Throughout this research, we use 2 different approaches which are ResNet50 and VGG19 to solve this problem. VGG19 is a deep neural network architecture under-using residual design principals, it is also a compact architecture which has a low diversity of architectures. On the other hand, ResNet50 is a deep neural network harnessing residual design principles and it has a moderate diversity of architectures. This network brings many a high productivity in a large number of researching in classifying X-ray images. Despite each approach has its own benefits and drawbacks, both are proved their productivity through real application.
2. Prerequisites
Python >= 3.6
Sklearn >= 0.24.2
NumPy >= 1.13.3
Tensorflow >= 2.6
3. Repo structure
Model: deep convolutional neural network architectures
VGG19
ResNet50 - 14 epochs
ResNet50 - 50 epochs
Plot: data plot for each architecture
VGG19
Loss for training
Model accuracy
Confusion matrix
ResNet50 (14 epochs)
Loss for training
Model accuracy
Confusion matrix
ResNet50 (50 epochs)
Loss for training
Model accuracy
Confusion matrix
Accuracy on training set
Accuracy on validation set
Loss for training set
Accuracy on validation set
Report: detailed report for this research
src: LaTeX source for report
Detect COVID-19 with Pneumonia via DCNNs.pdf: publication's name
Presentation.pptx: presentation file
Demo.ipynb: Demo to prototyping on jupyter notebook
[8] Krizhevsky, A., Sutskever, I. & Hinton, G. E. (2012). ImageNet Classification with Deep Convolutional Neural Networks. In F. Pereira, C. J. C. Burges, L. Bottou & K. Q. Weinberger (ed.), Advances in Neural Information Processing Systems 25 (pp. 1097--1105) . Curran Associates, Inc. .
[9] K. Simonyan et al., Very deep convolutional networks for large-scale image recognition, 2015.