Navigating the Transformation of AI Cancer Detection and Diagnosis

Stuart J. Schnitt, MD

Read the first half of the interview here.

In an interview with Targeted Oncology^TM, Stuart J. Schnitt, MD, chief of breast oncologic pathology at Dana-Farber Brigham Cancer Center, associate director of the Dana-Farber Cancer Institute/Brigham and Women’s Hospital Breast Oncology Program, senior pathologist at Brigham Cancer Center and Women's Hospital, and a professor of pathology at Harvard Medical School, considered the current and future role of artificial intelligence (AI) in cancer care to assist pathologists and oncologists in diagnosing patients.

Targeted Oncology^TM: What role do you feel like AI will play in early cancer detection? How can pathologists and oncologists prepare for that shift for patient referrals and staging?

Stuart J. Schnitt, MD: Already, AI is being used in some centers to interpret mammograms. There have been a number of studies that have documented the fact that the use of AI can increase the detection of breast cancer, so that is important. But these studies have also shown that the combination of the radiologist and AI is better than either alone.

I think one of the things that patients are afraid of is that their mammogram or other radiologic finding or pathology will be interpreted by some machine and no human eyes will ever be set on it. That is a misperception; radiologists and pathologists will have the ultimate responsibility for what the AI algorithm does, and they need to use the AI algorithm as an adjunct and not a replacement for anything. Patients [hopefully will] understand that at least in the immediate future, there is no plan to phase out the diagnostic team, including radiologists and pathologists, and replace them with machines.

How do you feel like AI can help oncologists or pathologists synthesize larger or more complex data with actionable insights without adding to clinician burnout?

Algorithms like chatGPT can be very useful in this setting. An oncologist could use chatGPT, or some similar AI-based algorithm to say, "I have a patient with breast cancer who has a tumor of this size, this histologic type, this grade, this receptor status, what's the most appropriate treatment?" From a clinical point of view, the AI algorithms can bring physicians up to a level that would be difficult to obtain for a lot of people because of time constraints and those types of issues.

Then again, with diagnostic issues, I think that there will be more reproducible diagnoses. We know, for example, that there's a subtype of breast cancer, the second most common type of breast cancer, called invasive lobular carcinoma, and there's a fair amount of interobserver variability in the diagnosis of invasive lobular carcinoma. But when you apply an AI algorithm to a breast cancer and ask, “Is this ductile, is it lobular, or is it something else?” the accuracy and the area under the receiver operating curve for diagnosing invasive lobular carcinoma is very high. For difficult diagnoses, problematic diagnoses, and controversial diagnoses, AI algorithms can help standardize the diagnostic criteria.

What are some ethical or regulatory concerns that oncologists should be aware of as AI becomes more embedded in cancer care?

That's an excellent point. Not [only is] digitizing slides a major barrier, but there are also ethical and financial issues, etc. Having pathology department go digital is very expensive. For a big pathology department like ours, we do about 80,000 cases a year, which means that there are hundreds of thousands if not millions of slides that would need to be scanned on a routine basis. The number of scanners that we would need and the personnel to do that is very expensive and you need a lot of space for that.

From an ethical point of view, there's an issue of who's ultimately going to be responsible for a diagnosis that is aided by AI. Is it the pathologist? Is it the AI company? I think most people agree that, at least for the present time, the pathologist whose name goes on the report is going to be ultimately responsible. Pathologists are not going to just blindly depend on AI algorithms, knowing that they're going to be ultimately responsible for the diagnosis.

Then there are issues with digital slides. If they're being sent around among institutions, how is patient privacy protected? This is [the type of concern] that needs to be worked out. The other things that need to be worked out are the regulatory issues: what algorithms can be used with FDA approval, without FDA approval, or as off-label use? Those areas are in their infancy, and people are working very hard to try to define those. I think it will become clearer in the next several years how this is going to play out, both in terms of the ethical issues and the regulatory issues.

Do you feel that AI will reduce or exacerbate disparities in cancer care globally, and what steps can pathologists and oncologists take to ensure equitable adoption?

The limitation in applying AI is digitizing slides. Having said that, I saw something several years ago that was done by one of the people in our department, Faisal Mahmood, PhD, who showed the ability of an AI algorithm to be run on a slide that was placed on a microscope stage, and then the pathologist puts their iPhone to the ocular [lens] of the microscope, captures the image, and it gets digitized that way, and then the AI algorithm gets run on that. So that's interesting, and maybe that would happen in low-resource countries. If a low- or middle-income country can't do immunohistochemistry, it's hard to believe that they're going to be able to buy slide scanners. Although, in the long run, that may be a less expensive way of doing things than sending cases out or bringing up the technology to do other things.

In terms of diagnostic accuracy and leveling the playing field, the AI algorithms would give people with less experience the experience of an expert when using the AI algorithm. But again, they're going to have to know about false positives, false negatives, and all that. I think from a global medical care point of view, AI can do a lot to standardize things across countries and across levels of experience.

Is there any advice you'd give for oncologists to be able to work together with their pathologist on these new technologies?

This has to be a partnership. I view my role as a pathologist as part of the patient's care team, not isolated. The way we function here is that we work very closely with our medical oncologists, our surgical oncologists, and our radiation oncologists. As pathologists, we never implement something new without discussing it with our clinical colleagues, so they understand what they're doing and what we're doing, and they know the potential impact. I know here our medical oncologists are very enthusiastic about moving toward digitized slides and possibly using AI algorithms as well. So I think it's key that pathologists keep their clinical colleagues in the loop of what they're doing, what they're planning on doing, and even seek advice from their clinical colleagues about what they'd like to see done. For example, there are a number of AI algorithms out there that predict outcome in patients with breast cancer, but I would never bring one on without discussing it with my clinical colleagues and say, there are these algorithms out there, this is the literature on this, and what do you think we should do? It's critically important to maintain communication with your clinical colleagues.

What do you think is coming up in the future of AI in the cancer field?

It's critically important that people understand that AI is not going to replace pathologists. It's not going to replace radiologists, and it's going to be a huge step forward. I think the next big step forward in technology [will be] introduced into pathology, and I think it's going to fundamentally change the way pathologists practice. I can envision a scenario where a pathologist comes in in the morning and an AI algorithm has been run on all their cases; the algorithm has prioritized their cases and said, these are benign, and these have cancer in them. Then they will probably give preference to the ones that have [detected] cancer, so those diagnoses can get out, and cases could get moved along in terms of further workup.

I think it's going to fundamentally change the way we practice. In general, people are resistant to change regarding everything. But also, a lot of people who are advanced in terms of the way they think about things would welcome a change like this and embrace it, understanding that there are going to be growing pains and it's going to take a while to change how we do things in routine practice. Ultimately, everybody's going to benefit, particularly patients.