Brandon Suh, MD, MPH, MBA
With its ability to spot patterns in huge datasets, artificial intelligence can unlock secrets in medical images – and help move the dial on cancer care.
Immunotherapy may be the next step in cancer therapy but matching the right patient to the right treatment remains a challenge.
Brandon Suh, CEO of South Korean med tech company, Lunit, believes artificial intelligence (AI) in medical imaging will help advance precision diagnosis, personalised medicine, and, ultimately, conquer cancer.
“AI is the most advanced technology for pattern recognition. Because it is trained on a massive amount of data, its performance is extremely high and it can catch lesions that are hardly visible to the human eye,” he says, explaining that more effective, rapid diagnosis was crucial to cancer care.
“It’s not just about the treatment side of things,” he says. “We know that if cancer is diagnosed early, the patient does not need as extensive treatment and they have much better outcomes.”
Brandon believes that his company’s technology for mammography, which has been shown to be accurate in studies published in JAMA Oncology1 and Lancet Digital Health2, still has many secrets to unlock, including ways to tailor screening methods based on individual risk of breast cancer.
Paradigm shift in treatment
In recent years, next-generational immunological therapies that kill cancer by targeting immune cells have promised to transform the treatment landscape. But the current standard of care is reliant on a biomarker called PD-L1, which can be inaccurate in predicting responders, so there is still much to learn about who will respond.
“Cytotoxic chemotherapy and targeted therapy were about killing the cancer cells directly, so what we did before — sequencing of DNA and RNA — is no longer as important in immunotherapy,” says Brandon, explaining the focus was now on understanding which tissue structure and cells surround a cancer tumour.
AI is the most advanced technology for pattern recognition. Because it is trained on a massive amount of data, its performance is extremely high and it can catch lesions that are hardly visible to the human eye.
“That’s really complicated because there are so many different types of cells, including immune cells, that surrounds the tumour and we don’t really know how that correlates to treatment response.”
By analysing digitalised versions of tissue slide images magnified down to the cell level, AI can “join the dots” in the data in a way that has not been possible until now.
“Before, we had pathologists reviewing the images through the microscope. Now, we can run software on this complex data and extract insights.
“We are looking at a whole set of data that has not really been taken advantage of before, because it was so complicated for humans to look at.”
Through this work, Lunit is discovering a range of pathological features that are predictive of a person’s response to immunotherapy, and AI has been shown to be between 30 and 40% better at predicting outcome to treatment than PD-L1. The findings has been presented at AACR and ASCO, one of the largest conferences in oncology.
“Although our studies are still preliminary, we now have very robust data and are planning more ambitious prospective studies to clinically validate our product,” he says.
1 https://jamanetwork.com/journals/jamaoncology/article-abstract/2769894 |2 https://www.thelancet.com/journals/landig/article/PIIS2589-7500(20)30185-0/fulltext