Skip to main content
Premium Trial:

Request an Annual Quote

AI Tool Could Help Doctors Deliver Focal Therapy in Prostate Cancer

MRI showing tumor in the prostate gland

NEW YORK – A multimodal artificial intelligence tool could improve doctors' ability to deliver focal therapy to prostate cancer patients and potentially increase the number of intermediate-risk patients who can avoid a full prostatectomy.

Focal therapies involve removing a small tumor in one area of the prostate using methods such as cryoablation, high-intensity ultrasound, and laser ablation as an alternative to complete surgical removal of the prostate gland. Patients whose tumors are of low-to-intermediate aggressiveness and are fully contained within the prostate gland may be candidates for such therapy. This approach preserves the prostate gland and minimizes the adverse effects of surgery, such as urinary incontinence and erectile dysfunction.

However, tumor margins must be well delineated for focal therapy to succeed. MRI is commonly used to map the tumor margins, but MRI can underestimate the extent of the cancer, leading the surgical team to miss parts of the tumor and increasing the risk of its recurrence.

Researchers from the University of California, Los Angeles Jonsson Comprehensive Cancer Center and Avenda Health have developed an AI tool called Unfold AI that uses MRI scans to build a 3D model that delineates the tumor within the prostate. In December 2022, Culver City, California-based Avenda received 510(k) clearance from the US Food and Drug Administration for Unfold AI, and it is in clinical use at some institutions. Avenda has plans to expand the use of Unfold AI to major teaching hospitals and urology practices nationwide, bolstered by a new pair of studies validating its clinical utility. Then, in 2023, Avenda reported results from a study that showed the AI model was more accurate at predicting tumor margins than MRI.

In a study published this month in the Journal of Urology, researchers from UCLA and Avenda compared Unfold AI's performance to standard-of-care methods for tumor delineation in 50 prostate cancer patients who were eligible for focal therapy. Seven urologists and three radiologists manually delineated the patients' tumor margins using MRI scans and biopsy reports. Four or more weeks later, the same group of doctors again contoured the margins for the same cases using Unfold AI. Neither set of contours was used to guide therapy. All patients underwent prostatectomy, and the tumor specimens were used to evaluate the manual and AI predictions.

When assisted by the AI model, the doctors' predictions had an accuracy of 84.7 percent, compared to 67.2 percent for the manual method. The sensitivity of AI-assisted tumor delineation was 97.4 percent compared to 38.2 percent for the standard-of-care approach. Further, manual tumor delineation systematically underestimated the extent of the cancers, with only 1.6 percent of samples having negative margins, or no cancer cells at the outer edge of the contour, compared to 72.8 percent for the AI-assisted contours — a 45-fold improvement. Ultimately, doctors changed their decision on how they would define the tumor margins for focal therapy in nearly a third of cases, said senior author Shyam Natarajan, an adjunct professor of urology at UCLA and CEO of Avenda Health.

Natarajan said the study succeeded in clinically validating the algorithm, calling the results "clinically actionable, meaningful, and superior to standard of care." Only about 10 percent of prostate cancer patients are currently offered focal therapy, but Natarajan speculated that Unfold AI could increase those numbers by enabling better margin delineation for patients.

In another study, Natarajan's colleagues led by Wayne Brisbane, a urologic oncologist at UCLA, retrospectively applied Unfold AI to a clinical trial of hemi-gland cryotherapy. In the trial, UCLA investigators enrolled 118 men with treatment-naïve prostate cancer to assess clinical response and quality of life six months after cryotherapy. Using margins predicted by Unfold AI, Brisbane compared the probability of complete tumor ablation to the presence of prostate cancer in post-cryoablation MR-guided therapies.

As they reported at the American Urological Association conference in May, the researchers found that Unfold AI had a 70 percent sensitivity and 68 percent specificity for predicting focal therapy success and concluded that its use could even be more important than baseline parameters such as tumor stage. The group is now preparing a paper describing that study for publication.

According to Natarajan, Avenda is continuing to study Unfold AI's clinical benefit and ultimately hopes to validate it in prospective clinical trials.

Natarajan said that as AI develops as a healthcare tool, Avenda hopes to implement it in a way that engenders confidence. To that end, the company has focused on building and training models that account for physician-to-physician bias by using the actual pathology as the "ground truth," rather than simply comparing the tool's predictions to human predictions.

"It's incumbent on us and other AI developers in the space to keep the patient always in mind by tying everything that we're doing back to the outcome of actually demonstrating clinical benefit," Natarajan said.