Analytical performance validation of PROMISE criteria with deep learning enabled platform for total prostate tumor burden in 18F-DCFPyL analysis

Introduction: The objective was to validate the performance of automated Prostate Cancer Molecular Imaging Standardized Evaluation (aPROMISE) in quantifying total prostate disease burden in patients with intermediate and high-risk prostate cancer who undergo 18F-DCFPyL for staging purpose.

Methods: Patients with a recent diagnosis of intermediate/high risk prostate cancer according to D’Amico risk classification were consecutively included. 18F-DCFPyL-PET/CT, a prostate-specific membrane antigen (PSMA) ligand, was performed within the context of compassionate use under the approval of the Spanish Agency of Medication and HealthCare Products and after being approved by a multidisciplinary committee and previous patient informed and signed consent.

In addition, the International Society of Urological Pathology (ISUP) grade group (1 to 5) was obtained by histological analysis of multiple biopsy specimens of prostate gland.

Prostate axial slices of ¹⁸F-DCFPyL PET/CT were visually assessed independently by two experienced observers belonging to two investigational groups. In PSMA-positive studies, automated prostate tumor segmentation was performed using aPROMISE and compared to the scientific software package Matlab (R2021b, MathWorks, Natick, Mass) using an in-house semiautomatic-manual guided segmentation procedure developed by the Mathematical Oncology (MOLab) group.

After tumor segmentation, a visual check was performed to exclude physiological urinary activity from the segmentation. In case of extra-prostatic invasion with involvement of seminal vesicles, contiguous bladder, or rectum, all the involved structures were included in the segmented volume. Standardized uptake value (SUV) and volume-based variables: SUVmax, SUVpeak, SUVmean, PSMA tumor volume (PTV) and total lesion activity (TLA) were obtained.

We compared all the obtained variables by the aPROMISE and MOLab packages in the total sample of patients and attending to ISUP and risk categories using intraclass correlation coefficient (ICC) for the concordance analysis and paired sample T-test and ANOVA analysis.

Results: 58 patients were evaluated although only 54 were included in the analysis (¹⁸F-DCFPyL PET/CT was considered negative in 4 cases). 46/54 (85.2%) were high risk and 23/53 (43.4%) ISUP 4 or 5 tumors.

In the global analysis of the total sample, the mean ± SD of SUVmax, SUVpeak, SUVmean, MTV and TLA for MOLab vs aPROMISE were of: 34.31±32.45 vs 34.53±32.65, 18.75±17.39 vs 14.02±12.46, 11.20±7.31 vs 6.09±4.67, 10.27±10.85 vs 24.18±16.04 and 149.65±275.01 vs 182.21±340.51. ICC for the previous semiquantitative variables obtained in both packages was: 1, 0.833, 0.615, 0.494 and 0.950 (p<0.001 in all cases).

No significant differences were observed of SUV and volume based variables for the different ISUP grade groups and risk-categories for any individual segmentation package. However, significant differences were detected between both segmentation packages attending to the different ISUP grade and risk category groups except for the TLA and ISUP grade group.

Conclusions: The analytical validation of aPROMISE shows a good performance for the SUVmax and TLA prostate tumor segmentation in comparison to our inhouse MOLab method in the global sample of patients. However, significant differences exist between practically all the semiquantitative variables for the different ISUP groups and risk categories, facing up the highly procedure-dependence of the segmentation if these division is performed.