INTRODUCTION
The outbreak of COVID 19 brings one of the most difficult challenges to humankind recently. The infection spread rate is very high and infection testing time is very important. In some countries it has been observed that the infection rate is doubling every 3 days. There are several ways where we can detect. As of now, various countries are using real-time polymerase chain reaction (RT-PCR) as a standard diagnostic method. However, this is a very time-consuming process and many countries cannot afford RT-PCR on such a large scale. Deep Learning could play an important role in this scenario to classify the infected patient. Several researchers have used CT images to train Deep Learning algorithms. Authors in [2] used chest CT for COVID-19 screening for 51 patients and authors in [3] used machine learning on these CT images for classification. However, the high quality of CT images is required for training and testing.
In contrast to [3], several authors and projects focused on X-Rays instead of CT images. In this survey we have identified few studies which used X-Rays images as their datasets and classify COVID 19 patients.
A live API demo is setup at https://api.alpes.ai/covidprediction/
A pre print of our paper is available at https://www.researchgate.net/publication/340870379_Precise_Prediction_of_COVID-19_in_Chest_X-Ray_Images_Using_KE_Sieve_Algorithm
Project 1:
In [1], the authors proposed to use the X- Rays of COVID 19 infected patients instead of CT images. X-ray is the cheapest and most common imaging technique which could be an effective tool for training and classification. Authors in [1] presented a deep learning model consist of
- Backbone: An 18-layer residual pre-trained imageNet based convolutional neural network is used as a backbone.
- Classification Head: It’s an added convolutional layer a multi-layer perception, which contains a 100-neuron hidden layer, a one-neuron output layer, and the sigmoid activation.
- Anomaly detection head: this head generates the scalar anomaly scores and, accordingly, detects anomaly images.
Experimental Results:
100 chest X-ray images acquired on 70 subjects with confirmed COVID 19
1431 chest X-ray images diagnosed as pneumonia
Results:
| KPIs | Classification | Anamoly detection | |
| True Positive | 95.13% | 95% |
Project 2:
In this work [4], the Residual neural network (ResNet) model which is the improved version of the convolutional neural network (CNN) is used. ResNet50 is a 50- layer network trained on the ImageNet dataset. ImageNet is an image database with more than 14 million images belonging to more than 20 thousand categories created for image recognition competitions.
Experimental Results:
Pre-trained models such as ResNet50, InceptionV3, and Inception ResNetV2 have been trained and tested on chest X-ray images. Chest X-ray images of 50 COVID-19 patients have been obtained from the open-source GitHub repository shared by Dr. Joseph Cohen [5]. Firstly, the InceptionV3 pre-trained model classified 10 of the COVID-19 as True Positive for fold3 and classified 10 of the normal as True Negative. Secondly, the ResNet50 model also classified 10 of the COVID-19 as True Positive for fold-3 and classified 10 of the normal as True Negative. Lastly, Inception ResNetV2 classified 10 of the COVID-19 as True Positive for fold3 and classified 9 of the normal as True Negative.
Project 3:
In this work [6], authors extracted the deep feature of nine pre-trained CNN model and fed to SVM classifier individually. They have used ResNet50 plus SVM and compare with other eight models (AlexNet, DenseNet201, GoogleNet, Inceptionv3, VGG16, XceptionNet, Inceptionresnetv2, VGG19)
Experimental Results:
The experimentation used the One-Vs-all approach and linear SVM as the SVM classifier parameter. The result is based on the data available in the repository of GitHub, Kaggle, and Open-i as per their validated X-ray images. ResNet50 plus SVM is statistically superior compare to the other eight models. The proposed classification model for the detection of COVID-19 is achieved 95.38% of accuracy.
Project 4:
The author in [7] used CNN with the VGG-16 model and has fine-tuned the last few layers. The author argued that the addition of these layers reduced false- positive and false negative cases.
Experimental Results:
Total params: 14,747,650
Trainable params: 2,392,770
Non-trainable params: 12,354,880
Results
| precision | recall | f1-score | support | |
| COVID-19 | 0.94 | 0.94 | 0.94 | 16 |
| PNEUMONIA | 0.95 | 0.95 | 0.95 | 20 |
| accuracy | 0.94 | 36 | ||
| Macro average | 0.94 | 0.94 | 0.94 | 36 |
| Weighted average | 0.94 | 0.94 | 0.94 | 36 |
Project 5: Our Proposed Model
- Datasets[1][2]: We have used available data sets of X-Rays for COVID-19. These X-rays were taken in a specified angle with and Poster anterior view was very frequently used. Since the symptoms of pneumonia are very similar to COVID-19, therefore, we have obtained the datasets of X-ray images with pneumonia and healthy persons as shown in figure 1 below:
Figure 1: X-ray images of COVID, Normal and pneumonia
- Feature extraction: Feature Extraction aims to reduce the number of features in a dataset by creating new features from the existing ones (and then discarding the original features).
- CNN is used for Feature Extraction.
- CNN Input – 224 x224 image
- Dimension – 1000 (CNN Feature Extraction output)
- Classification Technique: KE SNN a non-iterative classification approach is used to classify normal, pneumonia and COVID-19 X-Rays
- Results:
- Training data: Counter ({Pneumonia: 10286, Normal: 3762, COVID: 92}) – 14140
- Testing data- Counter ({Pneumonia: 2533, Normal: 987, COVID: 16}) – 3536
- Output Obtained:
- No.of planes- 39
- Train time- ~30 seconds
- Accuracy – 99.83%
- Metrics
| TABLE 1: RESULTS | ||||
| precision | recall | f1-score | support | |
| COVID-19 | 1.00 | 1.00 | 1.00 | 16 |
| NORMAL | 1.00 | 1.00 | 1.00 | 987 |
| PNEUMONIA | 1.00 | 1.00 | 1.00 | 2533 |
| TABLE 2: RESULTS (ACCURACY, MACRO AVG, AND WEIGHTED AVG) | ||||
| precision | recall | f1-score | support | |
| ACCURACY | 1.00 | 3536 | ||
| MACRO AVG | 1.00 | 1.00 | 1.00 | 3536 |
| WEIGHTED AVG | 1.00 | 1.00 | 1.00 | 3536 |
THE OVERALL ACCURACY OF THE PROPOSED MODEL IS– 99.83031674208145%
References:
[1] Zhang, Jianpeng & Xie, Yutong & Li, Yi & Shen, Chunhua & Xia, Yong. (2020). COVID-19 Screening on Chest X-ray Images Using Deep Learning-based Anomaly Detection.
[2] Ophir Gozes, Maayan Frid-Adar, Hayit Greenspan, Patrick D Browning, Huangqi Zhang, Wenbin Ji, Adam Bernheim, and Eliot Siegel. Rapid AI development cycle for the coronavirus (COVID-19) pandemic: Initial results for automated detection & patient monitoring using deep learning CT image analysis. arXiv preprint arXiv:2003.05037, 2020. 1
[3] Yicheng Fang, Huangqi Zhang, Jicheng Xie, Minjie Lin, Lingjun Ying, Peipei Pang, and Wenbin Ji. Sensitivity of chest CT for COVID-19: comparison to RT-PCR. Radiology, page 200432, 2020. 1 [7] T Franquet. Imaging of pneumonia: trends and algorithms. European Respiratory Journal, 18(1):196–208, 2001. 1
[4] Automatic Detection of Coronavirus Disease (COVID-19) Using X-ray Images and Deep Convolutional Neural Networks: https://arxiv.org/ftp/arxiv/papers/2003/2003.10849.pdf
[5] Open database of COVID-19 cases with chest X-ray or CT images https://github.com/ieee8023/covid-chestxray-dataset
[6] Sethy, P.K.; Behera, S.K. Detection of Coronavirus Disease (COVID-19) Based on Deep Features. Preprints 2020, 2020030300 (doi: 10.20944/preprints202003.0300.v1).
