With ultrasound lung scans, IIT Palakkad develops COVID-19 diagnosis sytem
When Vinod A. Prasad, IIT Palakkad dean, joined hands with principal investigator Mahesh R. Panicker for a project to automate analysis of ultrasound lung scans to identify COVID-19 cases, they knew fully well the hurdles in the path that lay ahead. In pulmonary assessment, CT scans are considered the gold standard; ultrasound scans are not extensively used, especially in countries like India. “There are reasons for that. They could be misused in the identification of gender of foetuses,” says Prasad.
To automate analysis of lung scans using machine learning, the first step was to build up a database. That was the hardest part. After contacting doctors in Italy and Spain, Prasad and his team got 60,000 images of the lungs of COVID-19 patients. A Spanish doctor in La Paz Hospital, who himself was diagnosed with COVID-19, had taken daily ultrasound scans of his lungs. He had taken it for a total of 14 days, and the scans were shared with the IIT Palakkad team.
The analysis works in multiple levels. In the first stage, one can identify if their lungs are normal, or if there is a viral or bacterial infection. In the second, it identifies the intensity of infection. “With different gradients, we are able to identify the percentage of infection. This is a very novel idea,” said Prasad. The system, he says, provides for 96-98 per cent accuracy and sensitivity.
The system is open-sourced and available here. Anyone can upload their lung scans and get their results in real-time.
So, what is the relevance of this project? For a COVID-19 patient in isolation, it is almost impossible to do a CT scan; they would need to be moved to the scan room and this would involve mobility and contact. An additional danger of overuse of CT scans and X-Rays is radiation exposure. “Plus, CT scans are not really recommended for younger patients,” says Prasad. An ultrasound scan is a portable, low-cost, zero-radiation alternative.
But, the problem remains that the project needs more data to work with. Says Prasad, “Kottayam Medical College has given us scan results, but they are only that of pneumonia patients. Then there is the issue of data transfer. The ultrasound scan machines used in our hospitals overwrite data for the next patient, and the scans cannot be transferred to a different system. There are higher-cost models, which provide for the same.”