How can Artificial Intelligence (AI) help automate COVID-19 detection?
Climate change and the COVID-19 pandemic are two significant global challenges that are interrelated in several ways. As temperatures rise and ecosystems are disrupted, the risk of new diseases emerging and spreading from animals to humans increases. For instance, climate change can alter migration patterns of disease-carrying animals and insects, spreading diseases to new areas. The COVID-19 pandemic is a stark reminder of the consequences of neglecting the interconnections between human health, animal health, and the environment.
In this context, we must have increasingly faster ways of detecting diseases to prevent their spread, as with COVID-19. To show how this could be done using artificial intelligence techniques, this text will deal with this subject in general, presenting the results of the study done to test this automation and make the code available in python for anyone who wants to reproduce a sample of this automation process for COVID-19 detection and further develop it.
AI and the COVID-19 diagnosis process
Deep learning models, specifically convolutional neural networks (CNN), can assist in the diagnostic process by detecting COVID-19 from chest radiograph images such as X-rays and computed tomography. This text aims to show how to train those models to detect the presence of COVID-19 from chest radiographs accurately. The study used the transfer learning technique to train the models and a dataset with COVID-19 chest X-rays.
In detail, The methodology of this study involved the use of deep learning techniques, specifically the use of convolution neural networks trained through the supervised learning process. Three different architectures were used for the training process: Xception, Residual Networks (ResNet), and VGG-16. The models were trained using TensorFlow and Keras Python libraries and evaluated based on the accuracy metric. The results were plotted on a confusion matrix and analyzed using a class activation map (Grad-Cam) and visual inspection.
The results showed that the VGG-16 model had the highest accuracy, 97.3%, and the least overfitting. Specifically, with a total of 75, the model correctly classified 73 images. Accurately, the model correctly classified 100% (n = 39) of the images related to COVID-19, 88,2% (n = 17) of the images of normal lungs, and 100% (n = 19) of the images of lungs with other infections. Regarding the errors, the model classified 2,7% (n = 2) of the images referring to normal lungs as other infections.
In conclusion, using deep learning models can assist in the diagnostic process of COVID-19 by detecting the presence of the virus from chest radiographs. The results of this study suggest that the VGG-16 model had the best performance, with high accuracy and consistency in classifying the images.
The availability of codes and data on GitHub aims to encourage other researchers to enhance the models, leading to further progress in facing COVID-19 and other diseases.
