Harnessing insights from existing data and using machine-learning to assist with COVID-19 diagnoses.
The Department of Veteran Affairs is the federal agency in the United States, tasked with providing lifelong services for military veterans, their families, and all eligible beneficiaries. The Veterans Health Administration, under the purview of the VA, is the largest health care system in the US – with 172 official Medical Centers and close to 1200 outpatient clinics through the US. Over 9 million veterans are enrolled in the program.
With the unprecedented onset of the COVID-19 pandemic in 2020, health care service providers were struggling to meet the challenges around containing the spread of the virus and expediting the availability of reliable diagnostic methods. The call to action was more dire for the veteran community in the United States, with 37% of its population over the age of 65 – the age bracket most susceptible to the serious implications brought along with the illness.
Researchers at Veteran Affairs were looking to explore the use of AI & Computer Vision to detect COVID-19 infections from chest X-Rays in their efforts to expedite care & treatment. InterKnowlogy was brought in to provide technical expertise around the effort.
Microsoft’s CustomVision – an automated image classification and object detection system, part of Microsoft’s Azure Cognitive services – and publicly available Chest X-Ray images from patients with COVID-19 pneumonia, and those with pneumonia originating from other health issues, were used to train the detection platform. The training data set consisted of close to 500 images, each, of COVID-19 pneumonia, non-COVID-19 pneumonia, and normal lungs.
The model created produced astounding results – precisely predicting the presence of COVID-19 pneumonia with 93% accuracy. The research clearly demonstrated the value of utilizing technologies such as Computer Vision and Artificial Intelligence as a means to manage future health crises and their impact on society.
A publicly available website was also built as part of the effort to showcase the possibilities, where users can upload chest x-ray images and receive the model’s readings – for research use only. The details of the exercise are documented in a publication on the National Library of Medicine.
