Researchers aim to ‘harness the power’ of AI to detect, treat prostate cancer
Click Here to Manage Email Alerts
Researchers at The Ohio State University have harnessed artificial intelligence machine learning processes to inform and support timely treatment decision-making in prostate cancer.
Although these tools are not meant to replace trained pathologists, they are becoming a valuable means of confirming cancer diagnoses and identifying high-risk disease, according to Anil Parwani, MD, PhD, MBA, director of the division of anatomic pathology at The Ohio State University Comprehensive Cancer Center — Arthur G. James Cancer Hospital and Richard J. Solove Research Institute.
In the future, these artificial intelligence (AI) resources may help address a global shortage of pathologists.
“This is an opportunity for cancer diagnostics to benefit from AI,” Parwani told Healio. “We are very cautious as we launch these algorithms and are testing them extensively through clinical trials, just to make sure that when we expose patient materials to these algorithms, the output we are getting is what we expect. Ultimately, it’s someone’s life at the end of the line.”
Healio spoke with Parwani about his institution’s use of digital pathology and the potential for AI to be a support tool for cancer diagnosis and treatment decision-making.
Healio: How did you begin using AI in prostate cancer pathology?
Parwani: Historically, pathologists reviewed glass slides from a biopsy, then reviewed the slides under a microscope and made a diagnosis. At Ohio State, we implemented digital pathology, which converts a glass slide into a digital format. Now we don’t have to look at glass slides; we can look at images on our monitors. We can look at patient information, review the case and collaborate with another expert. I can quantitate cancer more accurately. That’s the first part of this equation. Our institution went live with digital reads of all biopsies in 2018. We have scanned about 3.5 million slides, and each slide is about 1.2 gigabytes. That’s a lot of storage, so our next step was AI.
Healio: How do you use AI to help with prostate cancer pathology?
Parwani: About 2 years ago, we started doing clinical trials with commercially available prostate cancer algorithms, and we began to validate them for our patients. We found quite a bit of interesting data, especially around the fact that we were able to accurately detect cancer using only computer-aided diagnosis. The lab processed the slides and, as the slides were scanned, the computer algorithm was running in the background, confirming the diagnosis of cancer and quantitating it.
This can help a patient in several ways. A clinician can use this as a screening tool to detect cancer and present it to a pathologist for final review. It can be used to train future pathologists, showing them highlights of where the cancer is and how it was detected. It can be used as a quality check — meaning, once I review the images on my monitor and release them into a patient’s chart, it could undergo second review by a computer. This could help a pathologist who doesn’t read many prostate biopsies, or serve as “training wheels” for new pathologists.
Healio: How is this AI software being used at your institution?
Parwani: Only one software is FDA approved, and we are testing it in a community setting to see if it can help pathologists. We’ve also done a study on nearly 400 biopsies to show that the digital read with AI is not inferior to a digital read alone. There is a lot of promise to this technology and the goal is to integrate it into the workflow. When I’m reviewing my cases, potentially I will have a button that will launch the AI algorithm, and it will screen and detect the case. This is very exciting as an additional tool for cancer diagnostics and patient care.
After digital pathology and AI, the next step is predictive methods to triage and stratify patients. Two prostate cancers that look the same may behave differently. For the first time, we have predictive assays that allow us to stratify these patients and determine which one will have a better prognosis. This predictive algorithm goes beyond diagnosing cancer. It provides clues as to what the prognosis might be. We also are looking at algorithms that can help design better therapies.
Healio: How might AI address the shortage of pathologists?
Parwani: There is a shortage of pathologists, both in the United States and around the world. Some countries only have one pathologist per million patients. Additionally, the number of cases we’ll be seeing likely will increase. If AI can reduce 15% to 20% of my workload, I will be able to take care of more patients. As workloads increase, this has the potential to free physicians to do the things they went to medical school to learn.
AI is not meant to replace a pathologist because, at the end of the day, a pathologist is responsible for the diagnosis. This is a decision support tool to help pathologists. A pathologist with AI is better than a pathologist without AI, so we have to embrace the new tools, understand them and harness their power.
Reference:
- AI tools see beyond the human eye to better diagnose cancer (press release). Available at: https://cancer.osu.edu/news/ai-tools-see-beyond-the-human-eye-to-better-diagnose-cancer. Published Jan. 19, 2024. Accessed Jan. 31, 2024.
For more information:
Anil Parwani, MD, PhD, MBA, can be reached at OSUCCC-James, 410 W. 10th Ave., Columbus, OH 43210; email: anil.parwani@osumc.edu.
Collapse
Healio Interviews
Click Here to Manage Email Alerts