With an accuracy rate well above 95%, the system can speed up triage and treatment amid record-breaking infections and backlogs.
The Indian Armed Forces’ technology wing, Defence Research and Development Organization (DRDO), has come up with an AI algorithm that detects COVID-19 infections quickly amid the record numbers of infections and deaths in the country.
Analyzing X-rays quickly can be a very useful alternative in India, especially in the smaller rural towns, owing to limited CT scanning equipment. A DRDO spokesperson noted that the AI-enhanced usage of X-rays in triaging cases with suspected COVID-19 is fast, cost-effective and efficient: “The novel coronavirus affects the lungs early even before the patient starts showing significant symptoms. Thus, compared to the RT-PCR test which takes many days, an X-ray based AI diagnostic tool can detect the infection in early stages with processing time in seconds.”
The tool called ATMAN AI was developed by DRDO’s Centre for Artificial Intelligence and Robotics (CAIR), in collaboration with a startup medical diagnostics firms 5C Network. The latter firm has been conducting scans of over 4,500 cases of potential coronavirus-infected patients every day. Its AI-powered algorithms have proven useful in helping radiologists improve patient triage.
Useful for urgent triage
The diagnostic algorithm can screen chest X-rays as a triaging tool to rapidly identify and assess lung infections requiring COVID-19 treatment. The AI-driven web-based application can detect and classify conditions into normal, COVID-19 and pneumonia using limited numbers of sample images.
By depending on X-rays, hospitals can reduce the existing burden on radiologists and reserve the limited CT equipment for use in diagnosing other illnesses. The ATMAN software is easy to navigate and can be easily accessed over the Internet through a range of smart devices, including mobiles, tablets, laptops and computers.
The most heartening aspect of ATMAN is that the algorithm has an accuracy of 96.73%. According to an official statement from DRDO, “a novel feature called Believable AI, along with existing residual neural network (ResNet) models, has improved the accuracy of the software. Being a machine learning tool, the accuracy will improve continually.”
Said Dr U K Singh, Director, CAIR, DRDO: “The development of ATMAN is a part of DRDO’s effort to help clinicians and partners on the frontline to have the tools they need to rapidly diagnose and effectively treat COVID-19 patients. There is a rush to develop AI tools for quick analyses of X-rays. ATMAN will help in automatically detecting radiological findings indicative of COVID-19 in seconds, enabling physicians and radiologists to more effectively triage the cases, especially in an emergency environment.”
In addition to diagnostic use, the system can be deployed for research and statistical reference. The chest X-rays of RT-PCR-positive hospitalized patients in various stages of disease can be retrospectively analyzed with Deep Learning & Convolutional Neural Network models by an indigenously-developed Deep Learning application by the CAIR.