It’s enabling far better CT and MR images, it’s summarisation capabilities will soon help clinicians take decisions significantly faster
Taha Kass-Hout is a cardiologist and a machine learning expert. He studied to be a doctor, and around the same time, did a Master of Science in biostatistics, during which he developed and validated a computer-based statistical algorithm to ascertain stroke cases, an early instance of what we today know as machine learning. So, he’s able to combine his expertise in the two areas in fascinating ways.
He spent nearly six years at AWS as VP of machine learning, & chief medical office. Since Jan last year, he’s been CTO at GE HealthCare.
We met him recently at GE HealthCare’s massive facility in Bengaluru – the company’s second biggest R&D operation after that in the US. Taha says in Amazon, most of his team was in India. That continues to be so in GE HealthCare. “Here we have software engineers, hardware engineers, experts in machine learning and AI, cybersecurity, 5G, advanced visualisation, medical imaging. To have that all under one roof across any diagnostic component is really rare,” he says.
Taha is using all this expertise now to infuse digital and AI in everything that the company does. Gen AI in particular, with its capacity to digest and analyse humongous quantities of data, has become a big part of his strategy. It’s being used to calibrate machines better, obtain better image quality, bring more applications to the machines faster, to predict the length of stay of patients, which in turn helps better allocate hospital resources.
Improving image quality in CT and MR machines can enormously help doctors and radiologists in making accurate diagnoses. Images are often adversely impacted by things like patients making slight movements during the scan. AI is making a substantial difference here. “Like in many smartphones, if there is some blur, AI algorithms take it out, if there’s some noise, that gets cleaned out. AI makes the image clearer frame by frame by frame. And at the end, as you hit ₹stop the video’, you automatically get really, really high quality video,” Taha says.
New technologies, he says, are giving twice the resolution in half the time, “which means now we can scan twice as many patients, with twice as much resolution.”
When images are being taken, like say for a brain image, it’s also important to position the patient in a particular way. Taha says it typically requires months of training for technicians to get this right. Today, AI tools are helping position the patient accurately – alerting the technician in case the positioning is incorrect.
Taha says such tools not only shave off a lot of time, it also enables standardisation of processes, so that even less trained technicians can do it. This can be a huge advantage particularly in smaller towns where the required talent is not available, and makes telehealth more viable.
Speeding up treatment
AI, Taha says, is also dramatically simplifying the job of those like radio oncologists who have to plan a cancer patient’s treatment. “Traditionally, they would spend hours, sometimes days just analysing the images, hand annotating each one of those, so that when they hand it off for the radiotherapy treatment, they know exactly where to go and zap, and ensure healthy tissues are avoided. Now, we have automated this process down to about 10 to 15 minutes. The tumour is in 3D, but we take a lot of 2D slices (images), and then we reconstruct the whole thing. That way, when the patient is ready for radiotherapy, the therapist knows exactly what to treat, and what to avoid,” he says.
Gen AI is similarly helping in complex cases, where patients might walk in with a suitcase of records. Tradi tionally, the doctor would have to spend hours going through the records, and even then, might miss important information. “Now, with Gen AI type of technology, we’re implementing ways to organise all this information, synthesise it, summarise it. Like for example, this is the last thing this patient had, here are the recommended next actions based on guidelines, a few months ago, he had chemotherapy, he tolerated that, that bone tumour was five centimetres, but has now shrunk. Imagine you can do that in a fraction of a second, and that’s what is becoming possible,” Taha says.
All of these are areas that the India team is working on. Taha says for two years in a row, in the medtech world, GE Healthcare has led with the number of machine learning algorithms that have been authorised by the US Food and Drug Administration (FDA). India teams have contributed substantially to this.