FDA approves new pathology AI software, launches AI-enabled device database
Posted on September 22, 2021 | Through
The United States Food and Drug Administration (FDA) has cleared software designed to help pathologists detect prostate cancer in slides scanned from prostate biopsies. In addition, the agency announced Wednesday that it has made available a list of devices using artificial intelligence and machine learning.
The newly licensed software, called Paige Prostate, is the first artificial intelligence (AI) tool approved by the FDA for in vitro diagnostic use for the detection of prostate cancer in biopsies. The AI Pathology Tool is “designed to identify an area of interest on the prostate biopsy image with the highest probability of harboring cancer so that it can be further examined by the pathologist if the area of concern was not identified during the initial examination, “according to FDA Tuesday Press release announcing the authorization of the software through the de novo path.
Paige Prostate, a product of the company Paige.AI, is used by pathologists on scanned slide images. Authorization of the software was based on a study involving 16 pathologists who reviewed 527 slide images scanned from prostate biopsies. Of these, 171 had cancerous tissue and 356 were benign.
Each pathologist performed an unassisted and assisted read of each slide image. Cancer detection in slides containing cancer cells was improved by 7.3% with the help of Paige Prostate compared to unassisted reads. Use of the AI tool had no impact on pathologists’ readings of slide images of benign prostatic tissue.
The FDA has pointed out that the use of Paige Prostate carries a risk of false negative and false positive results; this risk can be mitigated when a pathologist qualified to read prostate biopsy slides reads the slides and includes relevant clinical information in diagnostic decision making. Pathologists may also choose to use other diagnostic techniques on biopsy samples before making a diagnostic decision.
Since prostate cancer is typically diagnosed based on multiple biopsy and slide sites, the FDA did not consider the impact of Paige Prostate on the patient’s final diagnosis.
“Pathologists examine daily biopsies of tissue suspected of disease, such as prostate cancer. Identifying areas of concern in the biopsy image can help pathologists make a diagnosis that informs appropriate treatment, ”said Tim Stenzel, director of the Bureau of In Vitro Diagnostics and Radiologic Health at the Center for Devices and Radiological Health from the FDA, in the press release.
“Authorizing this AI-based software may help increase the number of prostate biopsy samples identified with cancerous tissue, which may ultimately save lives,” added Stenzel.
With this authorization, the FDA is also implementing special controls “related to labeling and performance testing” for devices of this new type; subsequent devices of this type can be authorized through the 510 (k) pre-market process, using Paige Prostate as the predicate device.
New AI / ML compatible database
On Wednesday, the FDA also announcement he launched a database of medical devices that use AI and machine learning (AI / ML).
In the announcement, the FDA noted that the agency has seen an increase in recent years in submissions of AI / ML compatible devices for use in many therapeutic areas and entering through all of its submission routes.
Describing the list of IA / ML compatible devices as “a resource for the public on these devices and the work of the FDA in this area,” the FDA presented its methodology for building the list: first, it moved on to comb through the FDA’s public information about the device. approvals and authorizations. Then he turned to other publicly available databases from sources such as Nature, STAT, and the American College of Radiology, among others. Finally, the agency referred to “publicly available documents published by specific manufacturers”.
The list is a public resource, but is not meant to be “exhaustive or complete,” the FDA said. Currently, the list has 343 entries and is up to date until mid-June 2021.
© 2021 Society of Regulatory Affairs Professionals.