Data-Driven Gating Boosts PSMA PET Quantification of Prostate Cancer Metastases

Respiratory motion has long been an awkward adversary in PET imaging, blurring the very lesions that physicians need to see and measure with precision. A new retrospective study published in EJNMMI Reports suggests that a hardware-free motion-correction technique called data-driven gating can meaningfully sharpen the picture. Researchers found that the OncoFreeze AI data-driven gating method, applied to PSMA PET/CT scans of patients with metastatic prostate cancer, significantly increased standardized uptake values, improved contrast-to-noise ratios and reduced image noise compared with conventional static reconstructions, and that these gains held for both gallium-68-labelled and fluorine-18-labelled PSMA tracers.

PSMA PET/CT, using either 18F-DCFPyL or 68Ga-labelled tracers, is central to prostate cancer staging. The two tracers show grossly similar detection rates for primary lesions, but detection of low-grade lesions and distant metastases such as lymph node involvement has been reported to be potentially lower with 68Ga-PSMA than with 18F-PSMA. Because 68Ga-PSMA is generator-based and approved as a product in many jurisdictions, it serves as a diagnostic companion for determining patient eligibility for therapeutic agents such as 177Lu-PSMA-617, making optimization of 68Ga-PSMA imaging increasingly important. Respiratory motion can degrade the quantitative accuracy of 68Ga PET images and thereby affect diagnostic accuracy, with the lower thorax and upper abdomen particularly susceptible to motion artefacts that compromise uptake measurements, volumetry and PET/CT registration.

Existing remedies have drawbacks. Breath-hold techniques are unsuitable given the long acquisition durations of PET. External devices such as ANZAI pressure-sensitive belts can track respiratory cycles, but they do not always fully eliminate motion artefacts, and gated reconstruction typically uses only 20 to 50 percent of total counts, which can increase image noise. Data-driven gating offers an alternative: it derives respiratory motion information directly from PET list-mode data without external hardware. The approach has been evaluated mainly for high signal-to-noise tracers such as 18F-FDG, and its effects on 68Ga-PSMA quantitative accuracy in metastatic regions had not been systematically assessed before this work.

The study team, led by Adam Farag and Patrick Veit-Haibach, analyzed patients with biopsy-proven prostate cancer who underwent 68Ga-PSMA or 18F-DCFPyL-PSMA PET/CT on a Biograph Vision scanner, either for primary staging of high-risk disease or for evaluation of biochemical recurrence, and who had metastatic disease. After exclusions for absent contourable lesions, complete response or equivocal findings, 23 patients in the 68Ga-PSMA group and 40 in the 18F-PSMA group remained. These cohorts yielded 133 metastatic lesions for 68Ga-PSMA, comprising 73 bone, 46 lymph node and 14 lung lesions, and 192 lesions for 18F-PSMA, comprising 83 bone, 84 lymph node and 25 lung lesions.

All imaging was performed with continuous bed motion and list-mode PET acquisition with time-of-flight capability, with total PET acquisition times ranging from 560 to 1,270 seconds. The OncoFreeze AI algorithm applies power spectral analysis to the PET data, deriving respiratory frequency and amplitude from a masked three-dimensional time series; the amplitude reflects temporal change in the anterior-posterior distribution of counts, and the respiratory frequency is defined within a range of 0.125 to 0.5 hertz. Unlike the earlier OncoFreeze technique, which required an external waveform source and used only a limited portion of acquired data, the AI version uses the full set of PET events. Static images reconstructed with the clinical standard ordered subset expectation maximization served as the reference.

The quantitative results were consistent across lesion sites. For 68Ga-PSMA, SUVmax increased by 5.3 percent in bone, 10.6 percent in lymph nodes and 14.4 percent in lung lesions compared with static imaging, all statistically significant. For 18F-PSMA, SUVmax rose by 11.6 percent in bone, 9.9 percent in lymph nodes and 32.1 percent in lung lesions. SUVmean biases also favored the gated reconstruction for all lesion types and both tracers, with the greatest gains in lung lesions, the metastases with the highest amplitude of motion. Contrast-to-noise ratio biases for 68Ga-PSMA were 3.2 percent in bone and 21.7 percent in lymph nodes, with corresponding values of 3.9 percent and 7.6 percent for 18F-PSMA. Image noise was significantly reduced with the gated method, although static contrast-to-noise ratios for lung lesions did not differ significantly from the gated values for either tracer.

Qualitative assessment supported the numbers. Two radiologists, each with ten years of experience, independently scored overall image quality, tumor sharpness and noise on a four-point scale in blinded, side-by-side review. Across the 63 patients, data-driven gating produced high overall image quality in 95 percent of assessments, sharp image contrast in 97 percent and low image noise in 57 percent, all better than static images, with an observed inter-reader agreement of 86.1 percent. Inter-reader Cohen’s kappa was 0.66 for overall image assessment, indicating significant substantial agreement, but only 0.22 for lesion-level assessment, a non-significant fair agreement suggesting some variability in subjective lesion-level interpretation. All images fell within the good or acceptable categories, none were rated poor, and lesion counts did not differ between the reconstructions for either tracer.

The findings align with a growing literature. Büther and colleagues reported good correlation between data-driven and belt-derived gating signals and significant SUVmax improvements in 18F-FDG whole-body scans. Dias and colleagues, in a cohort of 592 18F-FDG PET/CT patients, found that automated data-driven gating avoided additional scan time and device failures and yielded SUVmax values roughly 14 percent higher than static images, an observation the authors note mirrors their 18F-PSMA results. Agrigoroaei and colleagues reported SUVmax biases of up to 27 percent in thoracoabdominal lesions, and Ince and colleagues found that combining data-driven gating with elastic motion correction and deblurring produced the lowest noise and highest contrast-to-noise ratios for 18F-FDG and 68Ga-DOTATATE scans. Regression analysis in the new study showed SUV measurements were largely independent of lesion location, and comparisons between tracers suggested the method is generally robust across both.

The clinical stakes extend to theranostics. Because eligibility for PSMA-targeted therapy such as 177Lu-PSMA-617 depends partly on SUV-based thresholds, with the VISION sub-study showing significantly improved outcomes in patients treated when SUVmean exceeded 10, more accurate quantification may influence patient selection. The authors argue that improved image quality reduces false-negative lesions, particularly small or motion-prone metastases near the liver dome, lung bases and nodes, and increases confidence that disease burden is truly PSMA-avid. They also report that the technique simplified workflow and reduced patient set-up time without altering lesion detection.

Important caveats remain. The study was conducted at a single center on a single PET/CT system with a heterogeneous cohort, and the two tracers were evaluated in different patient cohorts, preventing a fully paired comparison. The gated method was compared only with static imaging, not with belt-based correction in this cohort, and CT images were not motion-corrected, so residual PET/CT misregistration may have affected SUV quantification. No correction for multiple comparisons was applied, meaning some findings may reflect chance. A small number of outliers in 68Ga-PSMA SUVmean values, three of 73 bone lesions and one of 46 lymph nodes, did not affect clinical outcomes but suggest the method’s performance stability, while high, is imperfect for lower-uptake tracers. Downstream management changes were not assessed. Even so, the consistency of the quantitative and qualitative gains across two tracers with distinct distribution characteristics marks data-driven gating as a promising, hardware-free route to more reliable metastatic disease imaging.

Subject of Research: Data-driven respiratory gating for quantitative PSMA PET/CT imaging of prostate cancer metastases

Article Title: Effect of data-driven gating on the quantification of prostate cancer metastases in 68Ga-PSMA- and ¹⁸F-DCFPyL-PSMA PET/CT

Article References: Farag, A., Felder, C., Hussey, D., Mirshahvalad, S. A., Harsini, S., Metser, U., & Veit-Haibach, P. (2026). Effect of data-driven gating on the quantification of prostate cancer metastases in 68Ga-PSMA- and ¹⁸F-DCFPyL-PSMA PET/CT. EJNMMI Reports, 10(1), Article 44. https://doi.org/10.1186/s41824-026-00328-1

Image Credits: AI Generated

DOI: 10.1186/s41824-026-00328-1

Keywords: PSMA PET/CT, prostate cancer, data-driven gating, motion correction, 68Ga-PSMA, 18F-DCFPyL, SUV quantification, image noise, metastases, nuclear medicine, OncoFreeze AI, respiratory motion