How AI is reshaping cancer care and diagnosis

Artificial intelligence could have profound implications for the field of oncology, concluded panelists speaking to journalist and moderator Katie Couric at The Constellation Forum 2023 at Northwell Health last week.

Dr. Richard Barakat, physician in chief and executive director of cancer services and research division at Northwell Health, noted that using AI in imaging will help radiologists and serve as a clinical “copilot” designed to help avoid errors, such as false negative mammograms.

“The key we have to focus on with artificial intelligence is providing those backup systems,” he said. “But I think the role of AI is a lot more than that.” 

Barakat said another place his team is looking at using AI is to help with clinical trial matching in cancer. He said AI could also help predict the side effects of some of these treatments, allowing oncologists to try to mitigate them proactively.

Andy Moye, CEO of Paige.AI, an AI-enabled diagnostic platform for oncologists and pathologists, agreed that AI is really beneficial in making better diagnoses and reducing human error.

“[Oncologists] have to start with the right diagnosis and get it right the first time,” he said. “What we endeavor to do is to take these glass slides, these analog instruments, and digitize them, and once they’re digital, you’re able to unlock this huge world of machine learning and AI and all of the things that come with it.”

The challenge is how this vast volume of data can be stored and analyzed. 

“Every slide can hold up to two gigabytes worth of data, and 30 or 40 million slides are produced every year, maybe more than that,” Moye said. 

“We think about genomic information, clinical lab data, your clinical notes–you take all of that data, and you can build models that then have predictive values to them and really start to parse out upstream the population health side of this,” he explained.

That helps determine who in a population might be at higher risk for breast or prostate cancer.

“But then downstream, if you do get that mammogram, you can have better predictive outcomes,” he said. “These are the kinds of things where we see a really bright future.”

Daisy Wolf, an investing partner at venture capital firm Andreessen Horowitz, says AI can help manage spiraling healthcare costs by reducing the number of tasks clinicians currently perform.

“The very high cost of healthcare is driven by a lot of labor shortages, and AI is going to help us with that very soon by taking work off the human’s plate,” she explains. “And then every patient is going to have an amazing AI doctor and nurse in their pocket supplementing their real doctor.”

She added that even though ChatGPT wasn’t explicitly trained for medicine, from her perspective, it’s still “better than an average person with Google,” and she was impressed with the progress being made.

“I’m very optimistic about what technology and AI are going to do for human health,” Wolf said.

Still, Moye addressed the issue of implicit bias in AI, noting every clinician and every patient should have access to what he called the “nutritional label” for an AI model—the data sets on which the model has been trained.

“If you have this model that comes out, especially these large language models that are built on billions and billions and almost trillions of parameters–there’s implicit bias in human nature, and these large language models are built on that,” Moye said. “It’s going to mirror a lot of that stuff, unfortunately.”

Taking the other perspective, Barakat pointed out that AI could help with the bias incurred in many clinical trials.

“The reality is that the patients who are getting the most advanced cutting-edge novel therapeutics are those who know how to get to the right places, and underserved and minority patients are not getting the most advanced clinical trials,” he said.

He indicated what would help is when clinical trials are opened to everybody, and healthcare professionals can learn from everybody because there are “clearly” genetic reasons that differentiate certain patients.

“One of the most lethal forms of brain cancer, glioblastoma, is almost unheard of in African Americans–there’s a reason for that,” he said. “There’s a genetic reason for that. Let’s learn that and apply that to other populations. This is bidirectional. We must learn from everyone.”

Barakat added that despite the promise of AI, it’s essential to understand that only some have the ability to access generative AI tools, accentuating the importance for medical professionals to understand the technology.

“We can’t assume that all this tremendous technology is available to everyone,” he said. “My advice is let us be the best that we can be so that we can guide you and let us understand the AI to get the patients to where they belong.”