Radiology diagnosis gets AI edge in pvt hospitals

Several private hospital chains in Bengaluru are now assessing artificial intelligence (AI) tools to deploy in their radiology departments for more efficient diagnoses. AI can instantly point out abnormal areas in X-rays and CT scans with over 90% accuracy, making the final diagnosis easier for radiologists.

Bengaluru has around 10 startups that offer these AI products, and private hospitals are gradually picking these up.

Manipal Hospitals is currently integrating AI software from multiple vendors, including from Bengaluru-based startup Synapsica, into their radiology workflow. “We assessed many software solutions and liked some. The vendors will complete software integration in another six months,” says Dr Sudarshan Rawat, HOD - Radiology.

Also Read | China draft law proposes 'security assessment' for new AI products

“AI will mark areas on the image that are abnormal. After that, the radiologist will check this and give the final diagnosis,” he says, adding it will take time for the technology to mature enough to give diagnosis also by itself.

Other chains like Fortis and Aster are assessing multiple products.

“We are evaluating products on a trial basis. Introducing them into actual practice may take a couple of more years,” says Dr Sudhirkumar Kale, Lead Consultant - Radiology at Aster CMI Hospital.

“AI will be especially useful in case of emergencies. For example, it can read a brain CT and raise an alert simultaneously to the emergency and radiology departments if there is hemorrhage. More so in smaller or peripheral hospitals that don’t have radiologists round the clock.”

Dr S Pradeep, Senior Consultant - Radiology at Fortis Hospitals, says, “AI can definitely help in a hospital like ours that gets 300-400 X-rays a day. We are currently assessing accuracy and how much of our time it can save. We are checking our existing images against the software to see if they miss any case or give wrong interpretations,” he says, adding that issues of data privacy also have to be sorted out.

Dr Phaneendra Yalavarthy, who has been collaborating with Bengaluru hospitals to improve the technology, says, “Diagnosis has two steps - identifying abnormal areas on the image, and then characterising it. Currently, AI tools can do the first, but can only give some characteristics of the abnormality. So, it can’t give a final diagnosis.”

To reach this stage, a large number of annotated images are needed from radiologists, showing what diagnosis each abnormality corresponds to. This data would be used for machine learning, so that AI can give exact diagnoses in future, he says.

Kishor Joshi, Chief Business Officer at Teleradiology Solutions, one of the earliest remote radiology companies, says the industry has been steadily growing though hospitals are still cautious about adoption.