Lutetium-177 PSMA Therapy: Role of Imaging AI in Patient Selection

Lu-177 PSMA (lutetium-177 vipivotide tetraxetan, brand name Pluvicto) received FDA approval in March 2022 for metastatic castration-resistant prostate cancer, based on the VISION trial showing 4.0-month improvement in overall survival compared to standard of care. The approval included a companion imaging requirement: patients must demonstrate PSMA-positive disease on Ga-68 PSMA-11 or F-18 DCFPyL PET/CT before treatment. This creates a direct linkage between nuclear medicine imaging interpretation and therapeutic eligibility - and a direct role for AI in standardizing how that interpretation is performed.

The VISION Trial Imaging Eligibility Criteria

The VISION trial eligibility criteria required at least one PSMA-positive lesion on Ga-68 PSMA-11 PET/CT, defined as uptake higher than liver, with no PSMA-negative metastases larger than 1 cm (or 2.5 cm for bone lesions). The second criterion - excluding patients with PSMA-negative disease in any significant lesion - is the clinically challenging one, because PSMA expression is heterogeneous in advanced prostate cancer, and the definition of "PSMA-negative" on visual inspection is not precisely standardized.

The PSMA-negative criterion was introduced to exclude patients with dedifferentiated prostate cancer that has lost PSMA expression - a subset that would not respond to PSMA-targeted therapy and would face toxicity without benefit. A single large liver metastasis with low PSMA expression in a patient with otherwise PSMA-avid bone disease creates a clinical dilemma: does that patient meet eligibility? The VISION protocol's answer was no, but the threshold for "low" versus "negative" expression was not quantitatively defined - left to reader judgment.

In practice, this reader-dependent eligibility determination has measurable variability. A study comparing eligibility assessments across five nuclear medicine readers for the same set of 50 borderline PSMA PET cases found agreement in 72% - meaning in 28% of borderline cases, reader judgment alone would determine whether a patient received Lu-177 PSMA therapy. Quantitative AI-assisted PSMA expression assessment, providing an objective SUV-based assessment of each lesion's PSMA expression level, reduces this variability by giving readers a consistent quantitative anchor rather than relying on categorical visual assessment.

Quantitative PSMA PET Assessment: What AI Measures

AI-assisted PSMA PET quantification identifies all PSMA-avid lesions in the whole-body study, segments each lesion, and calculates SUVmax, SUVmean, PSMA lesion volume (PSMA-LV), and total PSMA tumor volume (TPTV) - the volumetric analogue of TLG for PSMA-targeted disease. These metrics provide a comprehensive picture of whole-body PSMA-positive lesion burden that manual reads cannot practically generate for patients with 20-50 metastatic sites.

The clinical importance of total PSMA tumor volume is emerging from post-hoc analysis of the VISION trial and from the TheraP trial data. Higher TPTV at baseline correlates with lower PSA response to Lu-177 PSMA and shorter progression-free survival. If this relationship is confirmed in prospective analysis, TPTV could become a continuous biomarker for treatment selection - identifying the high-burden patient subset where combination therapy (Lu-177 PSMA plus chemotherapy or PARP inhibitor) might be appropriate versus the lower-burden patient where single-agent Lu-177 PSMA is sufficient.

Manual calculation of TPTV for a patient with 30 bone metastases and multiple lymph node stations involved is not feasible in clinical practice. Automated segmentation makes it a reportable metric that can be calculated as part of the routine PSMA PET read - changing it from a research tool to a clinical measurement available for every patient selection decision.

Heterogeneous PSMA Expression: The Treatment Response Problem

After Lu-177 PSMA treatment, serial PSMA PET imaging at 6-12 week intervals monitors treatment response. The response pattern in PSMA-targeted therapy differs from FDG PET response patterns: effective treatment reduces PSMA expression (receptor downregulation and tumor cell death), but newly progressing lesions may show increased PSMA uptake as treatment selects for higher-expressing clones. A post-treatment scan with mixed response - some lesions with reduced uptake, others with increased uptake - requires lesion-by-lesion analysis that is time-consuming manually and benefits from automated multi-lesion tracking.

PROMISE MiTNM, the standardized reporting framework for PSMA PET in prostate cancer, provides a structured reporting vocabulary for disease extent and PSMA expression level by lesion category. AI tools that generate PROMISE MiTNM-structured reports from PSMA PET/CT studies - automatically populating lesion count, localization, and PSMA expression category by site - create consistent documentation that enables longitudinal comparison across treatment cycles without relying on reader recall of the prior study.

PSMA-FDG Discordance: Identifying Non-Responders

A clinically important scenario in advanced prostate cancer is PSMA-FDG discordance: disease that is FDG-avid but PSMA-negative, indicating dedifferentiated adenocarcinoma or treatment-emergent neuroendocrine differentiation. This population does not respond to Lu-177 PSMA and has poor prognosis from conventional therapies. Identifying discordant disease requires comparing PSMA PET and FDG PET at the lesion level - a multi-study, multi-lesion registration problem that is difficult to do accurately by visual comparison of separate read sessions.

AI-assisted co-registration of PSMA PET and FDG PET studies, with automated lesion matching and discordance classification, provides a structured output that identifies PSMA-negative/FDG-positive lesions for physician review. This is not a problem that can be solved by reading each study separately and mentally comparing - the spatial precision required to determine whether a specific lymph node shows concordant or discordant expression needs image-level co-registration, not study-level comparison.

The treatment implication of identifying discordant disease before Lu-177 PSMA initiation is the prevention of unnecessary treatment in patients who will not respond - and the redirection of those patients toward trials or standard therapies appropriate to their biology. As we discuss in our article on automated dosimetry for radiopharmaceutical therapy, the precision medicine model for RPT requires both patient selection based on imaging and dose optimization based on dosimetry - PSMA PET quantification addresses the first half of that equation.

The Community Oncology Gap

Lu-177 PSMA therapy is approved and commercially available, but most treatments are administered at academic or large community cancer centers. The PSMA PET required for eligibility assessment is performed at nuclear medicine departments across the country - many of which do not have nuclear medicine physicians with specific expertise in prostate cancer imaging. AI tools that generate structured quantitative PSMA PET reports, standardizing the PROMISE MiTNM documentation and flagging potential discordant disease for physician review, extend consistent eligibility assessment quality beyond the academic center environment where specialized expertise is concentrated.

This is not a minor operational point. As Lu-177 PSMA treatment access expands from academic centers to community oncology networks, the quality of imaging-based eligibility assessment becomes a determinant of how well the trial evidence generalizes to real-world outcomes. Patients incorrectly excluded from treatment due to non-standardized PSMA negativity assessment, or incorrectly included due to missed discordant disease, represent the population where the gap between trial and real-world outcomes is created. AI standardization addresses this gap at the imaging step that drives the eligibility decision.