3197 - Clinical Correlation of Physician and Auto-Generated Contours on Post-Implant CT Following Low Dose Rate Brachytherapy

Purpose/Objective(s): Post-implant CT-based dosimetry following low dose rate (LDR) brachytherapy for prostate cancer correlates with biochemical control and toxicities. Historically, dosimetric evaluation relied on physician contours, but artificial intelligence (AI) contouring is now more available. We performed a validation study comparing AI and physician contours in relation to clinical outcomes.

Materials/Methods: One hundred patients with prostate cancer who underwent LDR brachytherapy in 2009 at a single institution were retrospectively reviewed. Prescription dose was 144 Gy. Dosimetric variables (prostate volume, D90, V150, rectal V100) were calculated using physician contours and two AI methods: Commercially available software atlas-based segmentation and TheraPanacea (TP) ART-Plan contours. Patients with incompatible post-implant CTs or fewer than two follow-up PSAs were excluded. Clinical outcomes included biochemical failure (bF), clinical failure (cF), and late grade =2 GI and GU toxicities. Kaplan-Meier analysis was performed for biochemical recurrence free survival (bRFS) and clinical recurrence free survival (cRFS). Cox regression identified predictors of bF, cF, and toxicities.

Results: Eighty-three patients were eligible for analysis. Median age was 66 years, and median follow-up was 16 years. Seventy patients (85%) had low/favorable-intermediate risk disease, and 14 patients (17%) received hormone therapy. Median ultrasound-based prostate volume at the time of implant was 33cc. Using physician, MIM, and TP contours, median post-implant prostate volumes were 28cc, 35cc, and 65cc, respectively (p < 0.01); median D90’s were 143 Gy, 141 Gy, and 111 Gy, respectively (p < 0.01); and median V150’s were 39%, 41%, and 34%, respectively (p < 0.01). The 10-year bRFS and cRFS were 87% and 90%, respectively. Eleven patients (13%) experienced late grade =2 GU toxicity; none had GI toxicity. On univariate analysis, initial PSA (p = 0.03), Gleason score 7 vs 6 (p = 0.04), Gleason score 8-10 vs 6 (p = 0.01), risk group (p < 0.01), and hormone therapy (p < 0.01) were significantly associated with bF, but multivariate analysis found no associations. Notably, physician-, MIM-, and TP-derived prostate volume, D90, and V150 were not associated with bF. On univariate analysis, Gleason score 7 vs 6 (p = 0.04), Gleason score 8-10 vs 6 (p = 0.05), risk group (p = 0.02), hormone therapy (p = 0.02), and TP-derived D90 (p = 0.04) were significantly associated with cF; TP-derived D90 trended toward significance (p = 0.08) on multivariate analysis. No dosimetric variables were associated with late toxicities.

Conclusion: To our knowledge, this is the first study correlating AI contours with clinical outcomes following LDR brachytherapy. Despite dosimetric differences among physician, MIM, and TP contours, no significant associations with clinical outcomes were identified. TP contours yielded lower D90 values, with a trend toward association with cF. Larger studies are needed for further validation.