While cognitive SOC contouring offered the highest specificity rate (96.2 percent vs. 72.1 percent for adjunctive AI and 53.4 percent for hemigland contouring), it also had a significantly lower negative margin rate (1.6 percent) in contrast to adjunctive AI (72.8 percent) and hemigland contouring (86 percent).
“(Adjunctive AI) was significantly more accurate than hemigland and SOC contours, indicating a better balance between sensitivity and specificity. Furthermore, exposure to AI software encouraged physicians to define larger and more patient-specific contours, improving the negative margin rate. When AI software was used in our study population, the negative margin rate rose from 1.6% to 72.8% — a 45-fold increase,” wrote lead author Sakina Mohammed Mota, M.D,. a data scientist with Avenda Health, and colleagues.
(Editor’s note: For related content, see “Study Says AI Mapping More Effective than MRI for Assessing Extent of Prostate Cancer,” “FDA Clears AI ‘Contouring Assistant’ in MRI-Guided Ultrasound Ablation Procedures” and “MRI-Based Deep Learning Algorithm Shows Comparable Detection of csPCa to Radiologists.”)
Three Key Takeaways
1. Increased accuracy with AI-assisted contouring. AI-assisted contouring demonstrated higher accuracy (84.7 percent) compared to hemigland contouring (75.9 percent) and cognitive SOC contouring (67.2 percent). This suggests AI can improve the precision of prostate cancer treatment planning by better balancing sensitivity and specificity.
2. Improved negative margin rates. The use of AI software significantly increased the negative margin rate from 1.6 percent with cognitive SOC contouring to 72.8 percent. This indicates that AI-assisted contouring may lead to more effective and safer surgical outcomes by helping define more accurate tumor boundaries.
3. Enhanced treatment recommendations: The introduction of AI-assisted contouring led to changes in treatment recommendations in over 25 percent of cases. Notably, it resulted in a 9 percent increase in the recommendation for focal therapy, suggesting that AI can support more targeted and less invasive treatment options for prostate cancer.
The researchers also pointed out that the adjunctive AI software’s specificity advantage over hemigland contouring (72.1 percent vs. 53.4 percent) is a significant consideration with respect to targeted, efficient treatment of prostate cancer.
“Though the sensitivity of AI and hemigland contours were comparable, AI contours were smaller and more specific. It is therefore plausible that AI may facilitate faster treatment with fewer side effects, particularly because hemigland ablation entails lethal treatment in close proximity to the urethra, external sphincter, and ipsilateral nerve bundle,” added Mota and colleagues. “Avoidance of these structures, though common in clinical implementation, would reduce hemigland contour size and likely compromise their efficacy for a subset of cases.”
The researchers also noted that adjunctive AI changed treatment recommendations in over 25 percent of cases with a notable 9 percent change in urologists recommending focal therapy (mean of 21.5 percent with adjunctive AI vs. 12.5 percent based on SOC contouring).
“The cancer estimation map improved confidence in recommending focal therapy over whole gland therapy, with urologists trending towards a more targeted approach,” noted Mota and colleagues.
In regard to study limitations, the authors acknowledged the reviewed cases were derived from one institution and entirely drawn from patients who had radical prostatectomy procedures. The researchers noted the performance of hemigland contours may have been inflated due to limiting the cohort to those with apparently unilateral cancer.
