Evaluating PSMA PET/CT Scans Using the AI-Based aPROMISE Platform

By Emily Menendez - Last Updated: June 28, 2023

The use of imaging biomarkers prior to prostate-specific membrane antigen (PSMA) radioligand therapy (RLT) plays a crucial role in patient selection and prognostic evaluation. The approval of 177Lu-PSMA-617 PSMA RLT by the US Food and Drug Administration has led to new treatment options for the management of metastatic castration-resistant prostate cancer (mCRPC) and, in combination with 18F-fluorodeoxyglucose (FDG)-positron emission tomography and computed tomography (PET/CT), can improve the assessment of tumor heterogeneity and burden in patients.

The aPROMISE platform is a picture archiving and communication system developed by EXINI Diagnostics AB that has been evaluated by a team of researchers as an artificial intelligence (AI) application for the evaluation of PSMA PET/CT scans. The platform offers quantitative analysis and standardized reporting of PSMA PET/CT image assessments and utilizes deep learning technology on PSMA images to enhance efficiency and consistency.

A research team developed a proof-of-concept study with the aim of implementing aPROMISE for the evaluation of both PSMA and FDG-PET/CT scans that are acquired prior to PSMA RLT.

The study cohort included 7 patients with mCRPC who had received PSMA RLT between December 2021 and December 2022. The aPROMISE software, which is currently validated for use with PSMA PET/CT scans only, was used to semiautomatically extract lesions with quantitative and anatomical data from pretreatment PSMA and FDG-PET/CT.

The aPROMISE platform utilizes deep learning to extract anatomical data from CT images, detect candidate lesions, and combine anatomic information from CT scans with PET-identified lesions to group candidate lesions based on the anatomic site. A workflow is created to vet the final list of true lesions, and the software then summarizes disease burden in an aPROMISE score, which represents the interaction of lesion volume and uptake for the identified lesions. Separate scores are provided based on the location of the metastatic lesions.

The aPROMISE scores obtained from the PSMA and FDG-PET/CT scans were used for PSMA and FDG-PET/CT comparison. No additional modifications have been implemented on aPROMISE for the FDG analysis at this point.

All 7 patients included in the study underwent 18FDG-PET/CT (FDG PET) 75.7±23.0 days after receiving 18F-DCFPyl PET/CT (PSMA PET). PSMA PET scores were recorded 82.4±24.0 days before the first treatment cycle, while FDG PET scores were taken 6.7±8.0 days prior. A total of 4 patients completed 6 courses of treatment, while 1 patient showed significant disease progression and did not undergo further RLT cycles. Six patients had a favorable prostate-specific antigen (PSA) response to treatment, with an average PSA of 178.8±258.1 ng/mL and 3.3±3.5 ng/mL before starting PSMA RLT and after the last cycle, respectively (−95%±0.06).

The average PSMA score was 1341.86 (±1705.02), while the average FDG score was 140.76 (±181.83). A comparison of manual delineation and aPROMISE for obtaining tumor burden assessment on FDG PET and PSMA PET is currently in progress, and a larger cohort will be used to further conduct comparisons of the correlation of the PSMA/FDG aPROMISE scores and treatment response.

The study demonstrated that the aPROMISE platform is an effective option for automated tumor delineation and the quantification of whole-body tumor heterogeneity on FDG-PET/CT. Additional training is necessary to further evaluate AI-based methods of assessing PSMA and FDG PET prior to PSMA RLT.