Click here to Skip to main content
14,972,962 members
Articles / Artificial Intelligence / Keras
Article
Posted 19 Feb 2021

Stats

2.8K views
24 downloads

COVID-19 Diagnosis Results with Deep Learning and ResNet50

Rate me:
Please Sign up or sign in to vote.
5.00/5 (1 vote)
19 Feb 2021CPOL2 min read
In this article, we’ll discuss the COVID-19 detection results we’ve achieved with our model and compare these results with those of other models.
Here we’ll compare the results we’ve obtained with those that had been published for alternative models, as well as analyze its performance on a new dataset.

In this series of articles, we’ll apply a Deep Learning (DL) network, ResNet50, to diagnose Covid-19 in chest X-ray images. We’ll use Python’s TensorFlow library to train the neural network on a Jupyter Notebook.

The tools and libraries you’ll need for this project are:

IDE:

Libraries:

We are assuming that you are familiar with deep learning with Python and Jupyter notebooks. If you're new to Python, start with this tutorial. And if you aren't yet familiar with Jupyter, start here.

In the previous article, we trained and tested a ResNet50 model that had been transfer-learned to classify chest X-rays into COVID-19 and Normal images. In this article, we’ll compare the results we’ve obtained with those that had been published for alternative models, as well as analyze its performance on a new dataset.

Result Comparison

As was mentioned in the previous article, our network showed a robust performance when classifying testing images, in which it achieved the accuracy of 95%.

Table 1 shows a comparison of our network’s accuracy with other related competing solutions. You can see that our fine-tuned ResNet50 outperformed several networks in diagnosing COVID-19. This demonstrates the powerful generalization capability of our model.

Reference Pre-trained Model Dataset Testing Accuracy
Panwar (2020) VGG16 COVID-19 and others 88.1%
Albahli (2020) ResNet152 COVID-19 and other chest diseases 87%
Ozturk (2020) DarkCovidNet COVID-19, Pneumonia, and Normal 87%
Ours ResNet50 COVID-19 and Normal 95%

Testing on a New Dataset

To further verify the feasibility of the proposed transfer learning-based COVID-19 diagnosis system, we tested it on a new set of images, collected from another public dataset. Testing a network on a totally new dataset can be challenging as the type and quality of images may be different than those of the first dataset that the network was trained on. We usually test the network performance on a small set of images taken from the same dataset used for training. However, in this project, we attempted to measure the robustness of the network if tested on new images taken from a new dataset. The new dataset contains COVID-19 and No finding images but we only selected 300 COVID-19 images and passed them to our model. First, we load the test images from the new dataset using ImageDataGenerator.

Python
# Testing nb2...ANOTHER DATASET
test_generator2 = train_datagen.flow_from_directory(r'C:\Users\abdul\Desktop\ContentLab\test2', 
                                                   target_size = (224, 224),
                                                   color_mode = 'rgb',
                                                   batch_size = 3,
                                                   class_mode = 'Binary',
                                                   shuffle = True)

After loading the new testing images, we passed them to the model to calculate the accuracy:

Python
Testresults2 = model.evaluate(test_generator2)
print("test2 loss, test2 acc:", Testresults2)

It is clear that our model maintained a relatively good accuracy even when it was run on images from a new dataset.

Next Step

In the next article, we’ll show you how to build a network for Covid-19 detection from scratch. Stay tuned!

License

This article, along with any associated source code and files, is licensed under The Code Project Open License (CPOL)

Share

About the Author

Abdulkader Helwan
Engineer
Lebanon Lebanon
Dr. Helwan is a machine learning and medical image analysis enthusiast.

His research interests include but not limited to Machine and deep learning in medicine, Medical computational intelligence, Biomedical image processing, and Biomedical engineering and systems.

Comments and Discussions

 
-- There are no messages in this forum --