Can you trust a diagnosis made with the help of Artificial Intelligence?

Every few seconds in India, someone somewhere is undergoing an MRI, a CT scan, an X-ray or some kind of test that requires the services of a radiologist. If one goes to any doctor with virtually any symptom, they usually ask for some scan and some blood tests. Unlike some years ago, it has become almost the regular protocol.

Although in the metros and bigger cities in India, usually your scan will be examined and the problem diagnosed by a qualified expert, the chances of that happening in a small town – or a remote village – are quite slim. Often the patient will be handed the report and asked to take it to the physician; very rarely will he get a report in hand from a trained radiologist.

Why? High quality medical care in smaller towns and villages is in any case a rarity, but even rarer is the presence of a trained radiologist. As things stand, India only has around 15,000 trained radiologists. One can assess how low that number is by comparing with the US that has around 35,000 for a population that is one-fourth of India’s.

It is to fill this gap that Prashant Warrier, 38, and Pooja Rao, 33, have come together and formed Qure.ai in July 2016.

The Mumbai-headqurtered company has developed software based on artificial intelligence (AI) to examine, read, detect, and diagnose scans of all kinds.
 

The duo has strong technical background in medicine and machine learning. A physician by training, Rao has a PhD in research from the International Max Planck Research School for Neuroscience focused on neuroscience and machine learning. She had previously worked as a bio-informatician at Genomescan, Leiden, and as a data scientist at GoDataDriven, Amsterdam.

Prashant Warrier, a BTech from IIT-Delhi and PhD from Georgia Tech in operations research, worked for five years in the US, came back to India in 2011 and founded an AI-powered personalised digital marketing firm, Imagna Analytics, which was acquired by Fractal Analytics Inc in 2015. Fractal is now financing the new start up, Qure.ai.

The idea is to use artificial “neuro” networks – software that can be plugged into the existing diagnostic machines - that are modelled on the human brain.
 

They work like the human brain and learn to read and understand images like humans do.

Radiologists go through thousands of cases in their eight years of training period. In the process of examining cases, they see scans that are normal and abnormal.

As a result, they learn to detect abnormality by discerning the difference between the two. The only difference is that in the case of the new technology Qure.ai is using, the artificial neuro network will be exposed to millions of cases instead of thousands.

But can one trust a diagnosis made with the help of machines or AI? Human beings would typically trust another human being over a machine.

But Warrier says that that is part and parcel of the diagnosis – human or artificial. Eighty-million chest X-rays happen in India every year, with radiology error rates for chest X-rays at 20-23 per cent. In fact, 19 per cent of lung nodules are missed!

So even now if one goes to a trained radiologist, there are chances he can read or make a mistake in the diagnosis.

But since the software runs through “millions” of scans instead of thousands as a human might, chances of an error are, in fact, lower. So chances of an error are, if anything, minimised.

Secondly, as of now, Qure.ai is not producing a final report but developing a report that is ready for further analysis by a radiologist. “At locations across the country where radiologists are not available, this report can be the one that is physically available. It could go wrong but something is better than nothing,” argues Warrier.

In cases where physicians disagree – often happens in cases like interstitial lung disease - the report produced through AI is “consistent and reproducible” and can be relied upon to arrive at a final conclusion.

In India, Qure.ai has no competitors, but globally Zebra Medical Vision is the nearest competitor. The software has been adopted and deployed in five centres in Mumbai, Delhi, and Bengaluru.

The company is also working with public health programmes in some countries. TB Reach has used it for processing TB cases (chest X-rays) from Nepal and Cameroon. Its solutions are being marketed in North America through its partners EnvoyAI (TeraRecon).

According to Warrier, the global sales team of their parent company, Fractal Analytics, is helping them to reach out to health care providers and possible customers in the US and Europe.

It may be early days yet, but if the technology and product prove as good as they claim, the benefits for patients will be felt for a long time to come.