The tech giant plans to teach its AI to find the traits of cancerous cells that trained pathologists do manually by looking at a slice of the patient’s tissue on a glass slide. SRL’s bank of more than one million such slides from past diagnosis will be used as the training material for the AI. Microsoft has already deployed its AI for early detection of blindness in people suffering from diabetes and assessing cardiovascular health of patients.
The company says this cancer-detecting solution won’t replace the need of an oncologist, but the goal is for the AI to speed up, make more accurate and give valid pointers to the doctor to make the right diagnosis. This will reduce the cost of treatment, better utilise the time of trained specialists to reach more patients and make healthcare in general more patient-centric, it adds.
“If you look at AI in general, it’s very good at taking a set of data, reasoning over it and providing insights. This is the same in healthcare, where access to good data combined with the power of machine learning, AI and cloud computing can provide doctors or healthcare providers with insights to make the right diagnosis,” says Anil Bhansali, Managing Director, Microsoft India (R&D).
Bhansali says the company is looking to partner with quality healthcare providers to acquire the large amounts of data that will be used to train the AI first. After the initial training, cases will be introduced to the AI to gauge its performance, before it can be taken out of the walled garden of high-quality data with which it has been trained. Since AI is always learning, every new or odd case it comes across, it can learn from and help with similar diagnosis elsewhere.
The partnership with SRL will give Microsoft access to a large set of high-quality images of tissue biopsies with the labelled diagnosis of a doctor for each one of those. For the Indian firm, the AI could replace the current practice of placing tissue biopsies on glass slides and manually checking them for cell abnormalities, doing away with one of the most laborious processes it follows today. “Unlike a routine blood test that happens using a machine, which is basically a chemical reaction between the antibodies in your blood and reagents, tissue biopsies have to be physically seen by qualified pathologists. They need to then take a judgment call as to whether there is any abnormalities in the cell structure on a microscope,” says Arindam Haldar, chief executive officer, SRL Diagnostics.
While Microsoft has selected India as a site for several of its AI in healthcare pilots, the company says it is building the capabilities for global consumption, as many of the problems India’s healthcare sector face are prevalent globally. Moreover, the company has begun partnering with global healthcare providers to get access to data of patients from those regions to train its AI system as well.
“One of the key things with AI that we have to be very careful about is that you can train the model, but at the end of the day it will be biased if the data you use to train it is biased. In cardiology one of the examples is the health risk markers which were more or less only from patients in the US or Europe. Hence, we specifically went out to get data on the South Asian population,” added Bhansali.
While neither of the executives believes AI will do healthcare diagnosis by itself anytime in the near future, they are of the opinion that given the constraint of resources in the Indian healthcare market, AI tools will help doctors make many more diagnosis in any day.