Nailing it Down: What Will AI’s Role Be in Cancer Care?

Krishnansu S. Tewari, MD

As artificial intelligence (AI) is further developed, the applications in oncology will be plentiful but human oversight will remain a crucial component, according to Krishnansu S. Tewari, MD.

“AI will be making significant inroads in oncology through genomics, proteomics, transcriptionomics, epigenomics, and metabolomics. [Additionally,] with machine learning and deep learning neural networks, AI is going to get more integrated and help clinicians and patients through their cancer journey,” Tewari said in an interview with OncLive®. “[Individuals] should not be afraid of AI. In medicine, it’s not going to take away anyone’s jobs but [will rather] make things easier.”

Recent notable findings on AI in cancer care include those from a study evaluating a deep learning radiomics model based on PET/CT scans which predicted PD-L1 expression in patients with non–small cell lung cancer. Findings from the validation dataset of 42 patients with PD-L1–negative disease and 94 patients with PD-L1–positive disease showed that the area under the curve of the fusion PET/CT scan model (0.910, 95% CI: 0.779–0.977) was higher than that of the radiomics model (0.785, 95% CI: 0.628–0.897) and the deep learning model (0.867, 95% CI: 0.724–0.952).1

In the interview, Tewari detailed the current and potential future applications of AI in cancer care as well as the current limitations of the technology. Tewari is director of the division of Gynecologic Oncology, Obstetrics & Gynecology School of Medicine, at the University of California, Irvine. He is also a professor and the Philip J. DiSaia, MD, Chair in Gynecologic Oncology at the Obstetrics & Gynecology School of Medicine.

OncLive: Please define AI as it relates to cancer care.
Tewari: AI is basically the software that allows machines to participate in decision-making with limited human intervention. Robots are the hardware. AI can relate to cancer care in many ways, including [aiding] operating room efficiency, tracking and identifying the source of infections among hospitalized patients, identifying/prioritizing abnormal CT scans, identifying/prioritizing abnormal pathology on microscopic examination, and aiding in drug development.

AI has not yet been rolled out at our institution, [UCI Health], but there are discussions with different AI companies on how to best integrate AI at our institution. It’s being used in several other institutions around the country.

How is AI currently used for oncologic drug modulation? What can it do better in this area vs standard approaches?
AI can be very helpful in determining the objective response rate on serial imaging studies for patients on clinical trials with novel drugs. AI can also [identify] patterns in response to drugs and assist in statistical analyses on clinical trials.

[There are] no data yet in clinical trials, but AI has the ability to triage abnormal scans, pathology, and drug response faster than humans, but at this point human intervention and oversight are critical.

What should a community oncologist know about AI?

AI’s applications in oncology are going to be transformative. Community oncologists should understand that the applications of AI in oncology are going to be very broad. There are going to be applications in clinical trial design, treatment response, triaging, imaging, pathology, genomics, transcriptomics, proteomics, and metabolomics.

How far along is AI when it comes to contributing to clinical trial design?

I’m not sure how far along we’ve come with it, [but] we need to be very cautious. There are different types of AI, [such as] machine learning and deep learning through neural networks. AI may recognize patterns that are not necessarily apparent to us as human beings, but because AI is built up using ‘illness scripts’ and doesn’t have any true knowledge of pathophysiology, there still needs to be human intervention and oversight because AI could potentially lead to false results by over generalizing or overfitting the data.

Before we roll AI out into clinical trial design, where I do believe it will have an impact, we need to do so cautiously. AI was developed for applications outside of the world of medicine and was not designed [specifically] for medicine. We have to move carefully and cautiously so that we can integrate what we know about pathophysiology and even human behavior as we use AI more in the field of medicine. That’s very important in clinical trial design and also in [examining] treatment response with respect to some of the drugs that are being evaluated in the oncology arena.

What are the current limitations of AI in drug modulation?
The major limitation of AI in drug modulation and in medicine in general is to remember that AI was designed for interfacing with disciplines outside of medicine. To be certain we don’t attempt to fit a square peg into a round hole we need to have human oversight as AI gets further integrated in medicine.

The most important shortcoming to be aware of with AI in oncology, particularly in clinical trials and drug development, is that AI has been loaded up with ‘illness scripts’ but does not understand pathophysiology, therefore human intervention and oversight are essential.

How should clinicians consider AI as it stands today and begin to use it in practice?

We hear a lot of things about AI, [such as] it’s going to take jobs, [but] I don’t believe that’s going to happen—at least in medicine it’s going to be more of a tool that’s going to help us. Right now, radiologists have to read so many images per minute to be able to get through an 8-hour day and AI can help radiologists because it can triage and find the abnormal scans that need the most attention.

AI should be [viewed] as a tool. It’s a tool that’s certainly going to need human engagement. As far as we know, there’s not yet any self-conscious AI [and] the ultimate step would be for AI to be self-conscious or generalized. We should embrace AI and use it as a tool to help us and not be afraid of it. We can try to use it to help us spend more face-to-face time with our patients who are struggling with cancer, rather than sitting at the computer doing our charting because AI can do that for us; AI can make great notes and very well quantify the percentage of disease that’s responding to treatment on scans. [It has] lots of applications, and, in terms of genomics, AI can identify important aberrations or variants. It’s going to be a great tool for us to use.

Think of AI as software. When we’re thinking of robots taking over the world, the robots are just the hardware. AI is the software.

Reference

1. Li B, Su J, Liu K, Hu C. Deep learning radiomics model based on PET/CT predicts PD-L1 expression in non-small cell lung cancer. Eur J Radiol Open. 2024;12:100549. doi:10.1016/j.ejro.2024.100549