CEO SUMMARY: Study data provided to the FDA “demonstrated increased diagnostic accuracy” when pathologists used the Paige Prostate system, said Paige’s Medical Director. The artificial intelligence-powered pathology image analysis tool is trained to help pathologists detect even small areas of cancer in biopsied prostatic tissue on digital images thus improving accuracy, a significant development given the high prevalence of prostatic cancer and its heterogeneous nature.
PATHOLOGISTS HAVE A NEW TOOL for diagnosing digital images of prostate cancer.
Last week, the federal Food and Drug Administration (FDA) approved an application from New York-based Paige to market Paige Prostate, the first artificial intelligence (AI)-based software to identify areas that could be cancerous on digitized images of prostate specimens collected via biopsy. FDA clearance of this system allows its use in clinical practice.
FDA officials noted that the prevalence of prostate cancer was one factor in its decision to clear Paige Prostate for clinical use. “The authorization of this AI-based software can help increase the number of identified prostate biopsy samples with cancerous tissue, which can ultimately save lives,” said Tim Stenzel, MD, PhD, Director of the Office of In Vitro Diagnostics and Radiological Health in the federal agency’s Center for Devices and Radiological Health.
As most pathologists know, prostate cancer is the second most common form of tumor among U.S. men and is one of the leading causes of cancer death among men, according to the federal Centers for Disease Control and Prevention.
The potential risks for pathologists include the usual false negative and false positive results, but pathologists could limit the number of false results by using the software as an adjunct to their visual review of slide images, the FDA noted.
In effect, the software assists qualified pathologists reviewing slide images, and those clinicians also would consider the patient’s medical history and other clinical information and may do additional laboratory studies on patients’ specimens before issuing a diagnosis, the FDA said.
“The FDA reviewed data from studies Paige completed on what effect the AI software would have on how pathologists diagnose cancer from digitized images of prostate biopsy specimens,” explained Paige Medical Director Juan Retamero, MD. “Essentially, we demonstrated increased diagnostic accuracy using Paige Prostate. In other words, pathologists may have fewer errors when aided by artificial intelligence.
“The problem for pathologists reviewing digital images is that prostate cancer is a heterogeneous disease,” Retamero commented during an interview with The Dark Report. “What that means is that the same patient may have different cancerous cells with different biological behavior.
“Some of these areas may be low grade, which is a less-aggressive tumor, but also some of the areas may correspond to a high-grade tumor, which would have a very different prognosis,” he noted. “Even within the same patient, there could be regions of cancer cells that are less aggressive, and there could be regions with more aggressive cancer cells.
“The possibility that a patient could have different regions with each type of cancer is why an accurate diagnosis is essential,” Retamero noted. “Pathologists want to be sure that they identify clinically significant cancer, meaning a tumor that will become aggressive, potentially metastasize, and go on to kill the patient.
“Such heterogeneity means it is important to help pathologists identify any areas harboring cancer in a prostatic biopsy specimen,” he added.
Therefore, an AI-powered image analysis tool can assist pathologists in achieving a higher level of diagnostic accuracy. “According to some studies, about 1% to 3% of prostate cancer cases are diagnosed incorrectly,” he explained. “But that means that—out of 100 patients—three might end up with a false negative diagnosis.
“That 3% rate may not sound like a lot, but the incidence of prostate cancer is substantial enough that even a small percentage of incorrect diagnoses can equal a large number of patients when that percentage is applied to the entire population,” Retamero observed. “That means inaccurate diagnoses may affect a significant number of prostate cancer patients.
“This is where artificial intelligence—such as in the form of the Paige Prostate software—can be useful in helping pathologists to do a better job,” he noted.
In its review of the Paige Prostate device, the FDA reported that it evaluated data from a clinical study in which 16 pathologists examined 527 slide images of prostate biopsies that were digitized using a slide scanner. Of the 527 images, 171 contained cancerous cells and 356 were benign. Each pathologist did two assessments of each slide, one was an unassisted read without the software followed by an assisted read using the Paige Prostate software.
For the clinical study, researchers did not evaluate the effect of the software on a final patient diagnosis, which is typically based on multiple biopsies, the FDA added.
Instead, the researchers in this study found that Paige Prostate improved detection of cancer on individual slide images by 7.3% on average when compared with pathologists’ unassisted reads, the agency noted. No effect was reported in the study on the reads pathologists did on benign slide images, the FDA added.
The Benefits of AI
While a 7.3% rate may appear to be relatively low, Retamero explained that researchers in this study reported the average sensitivity level among 16 readers who had different levels of clinical skill. “The data comes from some pathologists who were urological specialist pathologists and from some who were not such specialists,” he noted.
“If you’re a urological specialist in pathology and that’s what you do day in and day out, you have a significant amount of experience assessing prostatic biopsies,” he added. “The AI software will help you, but not as much as it would help pathologists who are not urological specialists. These are the pathologists who end up reporting most prostatic biopsies, and they are the group in the study in which we saw the greatest amount of impact from using artificial intelligence.”
Contact Juan Retamero, MD, at email@example.com; Andy Moye, PhD, at 904-312-4867 or firstname.lastname@example.org.