Urban India might be witnessing COVID19 waning, but the disease is now snaking its way into India's rural pockets, where access to quality healthcare is a challenge. 

Earlier this month, DRDO's Centre for Artificial Intelligence & Robotics (CAIR) developed ATMAN AI, an AI-based-COVID detection application software using chest X-rays (CXRs) which can classify the images into normal, COVID-19 and pneumonia classes. This secure, web-based solution was developed on the premise of optimising chest X-rays for rapid detection of COVID19-triggered lung abnormalities. Along with HCG Academics and Bangalore-based deeptech startup 5C Network, DRDO CAIR aims to take this novel solution to hospitals, clinics and primary healthcare centres. 

The Genesis of the Idea:

In December 2020, a group of doctors from HCG Cancer Hospital, Ankh Lifecare, Vydehi Institute of Medical Sciences, Rajarajeshwari Medical College & Hospital, Tata Memorial Hospital, St Philomena's Hospital, Bhagwan Mahaveer Jain Hospital, with inputs from researchers in DRDO CAIR, Maastricht University Medical Centre in The Netherlands & The Department of Immunology, School of Biological Sciences at the University of Edinburgh, Association of Healthcare Providers India, Indian Medical Association among others tested and validated the use of AI to screen chest x-rays of patients with lung disease and damage. 

Using Indian datasets from COVID-19 positive patients, the dataset comprised of 1,324 COVID-19, 1,108 Normal, and 1,344 Pneumonia CXR images. Transfer learning was carried out on Indian dataset using popular deep neural networks, which included DenseNet, ResNet50, and ResNet18 network architectures to classify CXRs into three categories. The model was retrospectively used to test CXRs from reverse transcriptase-polymerase chain reaction (RT-PCR) proven COVID-19 patients to test positive predictive value and accuracy.

A total of 460 RT-PCR positive hospitalized patients CXRs in various stages of disease involvement were retrospectively analysed. The commonly observed alterations included lung consolidations, ground glass opacities, and reticular–nodular opacities. Of the 460 CXRs analysed, the model reported 445 CXRs as COVID19 with an accuracy of 96.73%.

Dr. Vishal Rao, Department of Head & Neck Surgical Oncology at Bangalore's HCG Hospital was part of the group of doctors that validated this method. He says, "The use of AI to glean off chest X-ray data can be a game changer. We're now seeing COVID19 makes its way into villages and rural areas with limited healthcare access. Chest X-rays are more economical than CT scans, so this solution doesn't increment the challenges PHCs and smaller healthcare facilities already face. In addition, this solution is well suited to detect damage to the lungs caused by pneumonia as well. COVID-related pneumonia was a commonly seen symptom in this wave."

Scaling Technology For Mass Outreach 

Based on the initial screening results, the solution is recommended for mass testing by the doctors part of the study. This is where a scalable solutions deployment platform was essential. Bangalore-based healthcare deeptech startup 5C Network was roped in. 5C Network enables access to radiologists anywhere through its public cloud and AI-based infrastructure, effectively addressing the challenge of a skewed patient to radiologist ratio using technology.

When the HCG team reached out to 5C Network, they were looking for a scalable digital health solution into which ATMAN could integrate and have a wide impact. 5C’s client base, with over 1000+ hospitals/diagnostic centres across 27 states represented a great opportunity.

Cofounder Kalyan Sivasailam adds, "5C Network sees 4000+ Covid scans on a daily basis. A key element of ensuring ATMAN is effective is to ensure deployment is done through a scalable digital health solution. Our prior experience with multiple AI partners across modalities has allowed us to streamline and optimise our solutions for ATMAN. This benefit is two-fold, the turnaround time for deployment of CAIR’s solution is a matter of days and the reach can be increased many fold due to the remote, accessible nature of the 5C Network Platform."

Today, ATMAN is being deployed at zero cost directly on 5C Network’s platform. Additionally, 5C Network is actively working with state governments to increase impact and reach. An MoU has been signed with the government of Karnataka to deploy ATMAN across 194 PHCs across the state, other such rollouts are being targeted in the near future.