AI COVID-19 Detection
Working with researchers at Veteran Affairs, //IK explored the use of AI & computer vision technology to diagnose COVID-19 on chest X-Rays. Leveraging Microsoft Azure Cognitive Services, the project demonstrates the potential of using artificial intelligence in the diagnoses of COVID-19 and similar future health crises. With InterKnowlogy’s Rodney Guzman named in the patent, //IK, helped write the paper “Using Artificial Intelligence for COVID-19 Chest X-ray Diagnosis” which has been published medical journals and can be reviewed here.
Medical researchers at the James A. Haley Veteran’s Hospital, working in their fight against the COVID-19 pandemic, were exploring solutions to improve diagnostic detection of the coronavirus.
Wanting to explore the potential of using Artificial Intelligence and Machine Learning to improve diagnostic capabilities, they turned to InterKnowlogy to help:
- Explore the potential of using AI to diagnose patients with COVID-19 using only Chest X-Rays (CXR)
- Employ AI as a means to screen and reliably diagnose emerging health emergencies, such as COVID-19C
- Create a publicly available website that could augment COVID-19 CXR diagnosis
IK created a website application allowing users to submit Chest X-Ray images and have results returned, complete with confidence values for each diagnosis. Health care providers can upload images, and the machine learning algorithm provides the probability of COVID-19 pneumonia, non-COVID-19 pneumonia, or normal lung diagnosis.
Design & Discovery
Arriving at a quick but accurate diagnosis was critical for the success of this platform. IK leaned into the expertise of doctors and industry specialists to understand the intricacies of reading a Chest X-Ray for this particular use case and building a product that they could use with confidence.
InterKnowlogy worked with researchers to build a dataset using a variety of publicly available CXR images in order to train Microsoft CustomVision — an image classification and object detection system — to reliably detect COVID-19 from Chest X-Rays. The CustomVision training dataset included 484 images of COVID-19 pneumonia; 500 images of non-COVID-19 pneumonia; and 500 images of normal lungs. All images were classified and tagged appropriately.
The platform showed 92.9% precision and recall overall, with precision and recall rates of:
- 98.9% and 94.8% for COVID-19 pneumonia
- 91.8% and 89% for non-COVID-19 pneumonia
- 88.8% and 95% for normal lung diagnosis
With this successful implementation, the possibilities of leveraging AI and similar technologies to combat other health crises is highly feasible.
Partner With Us
Get in touch to learn more about our work with Artificial Intelligence, Computer Vision, and how InterKnowlogy can help your organization.
760.930.0075 Ext. 3