Automated Machine Learning Vision Detection for COVID-19

Background

Although the number of new cases of COVID-19 per day are dropping in recent times, hospitals are still crowded and overloaded with patients who have COVID-19 or COVID-19-like symptoms. Many hospitals struggle to find beds for all the afflicted.

Source: https://www.nytimes.com/interactive/2020/us/coronavirus-us-cases.html

Source: https://covidtracking.com/data/charts/us-currently-hospitalized

Introduction

Goal: Develop a model that can identify COVID-19 in chest x-ray images given data including lung images of healthy patients and patients with COVID-19 or Pneumonia that will aim to increase diagnosis accuracy and decrease the time taken to diagnose a patient, ultimately increasing the efficiency in hospitals to reduce overcrowding.

​

Healthy

COVID-19

Pneumonia

Source & Data: https://www.kaggle.com/tawsifurrahman/covid19-radiography-database

Modeling Approach

About the Dataset

• The samples are sourced from Kaggle’s COVID-19 Radiography Database. (https://www.kaggle.com/tawsifurrahman/covid19-radiography-database?select=COVID-19+Radiography+Database)
• The dataset included X-ray images of COVID-19 positive cases along with Normal and Viral Pneumonia images. The three types samples have almost equal numbers to ensure there is no class imbalance for ML modeling process.

Source: https://www.linkedin.com/pulse/covid-19-detection-using-google

-cloud-automl-vision-gopi-kudaravalli/?articleId=6645597631752036352

Google’s Cloud AutoML Vision (https://cloud.google.com/)

​

• AutoML Vision enables us to train high-quality machine learning models to classify the images according to our own defined labels and help us achieve faster performance and more accurate predictions.

​

#1: Enable AutoML Vision Cloud on GCP

#2: Select Image Classification under AutoML Vision

#3: Create New Dataset for Modeling

#4: Import Covid-19 X-ray Images for Labeling

Modeling Approach

#5: X-ray Images Labels for Different Samples

#6a: Train Covid-19 Detection Model (start training)

#6b: Train Covid-19 Detection Model (training complete)

​

#7: Evaluate the Covid-19 Detection Model Performance

​

Results (Web App at Confidence Threshold: 0.5)

Precision and Recall Score:

Confusion Matrix:

Results (Cloud Based at Confidence Threshold: 0.5)

​

Precision and Recall Score:

Confusion Matrix:

Cloud Based Model

​

Covisure Predicting Model

​

Conclusion

We were successful in developing a model that can distinguish between x-ray images of healthy lungs and lungs affected by COVID-19 and Pneumonia

Our experiment could be improved if we trained the model using x-ray images of a greater variety of lung-affecting diseases. This would greatly increase our model’s usefulness and allow medical professionals to identify diseases at a much faster rate.

Reference

• Narin, A., Kaya, C., & Pamuk, Z. (2020). Automatic detection of coronavirus disease (covid-19) using x-ray images and deep convolutional neural networks. arXiv preprint arXiv:2003.10849.
• Maguolo, G., & Nanni, L. (2020). A critic evaluation of methods for covid-19 automatic detection from x-ray images. arXiv preprint arXiv:2004.12823.
• Minaee, S., Kafieh, R., Sonka, M., Yazdani, S., & Soufi, G. J. (2020). Deep-covid: Predicting covid-19 from chest x-ray images using deep transfer learning. Medical image analysis, 65, 101794.
• https://cloud.google.com/vision/automl/docs/prepare
• https://towardsdatascience.com/google-cloud-automl-vision-for-medical-image-classification-76dfbf12a77e#f2b1
• https://www.linkedin.com/pulse/covid-19-detection-using-google-cloud-automl-vision-gopi-kudaravalli/?articleId=6645597631752036352

​