Analyzing all tumor lesions across multiple scans with a predictive model achieved 88% accuracy in identifying which neuroendocrine tumor patients would respond to PRRT.
AUC 0.88The model correctly predicted PRRT response in 88% of cases by analyzing every individual tumor lesion across multiple PET scans
What the researchers found
A multivariate linear regression model using comprehensive, lesion-level, longitudinal features from Ga-68 DOTA-TATE PET scans achieved a concordance index of 0.826, AUC of 0.88 (85% specificity, 81% sensitivity), and significant patient stratification into good and poor responders (PFS ≥25 months cutoff, p < 0.00001). This approach outperformed all single-feature predictors in a benchmark study.
Why it matters
PRRT is expensive and not all patients respond equally. Being able to predict who will benefit — with 88% accuracy — could prevent unnecessary treatment for likely non-responders and fast-track therapy for likely responders. It also demonstrates the power of comprehensive peptide-receptor imaging analysis.
The numbers in context
Models used comprehensive (all lesions), longitudinal (temporal), and lesion-level imaging features from 68Ga-DOTA-TATE PET scans.
How the study worked
Retrospective study of 36 NET patients treated with 177Lu-DOTA-TATE and imaged with 68Ga-DOTA-TATE PET at baseline and post-therapy. All individual lesions were segmented, anatomically labeled, and longitudinally matched between scans. Patient-level features were engineered from lesion-level data and selected for multivariate modeling. Validated via concordance index, ROC analysis, and Kaplan-Meier survival analysis.
Who was studied
Neuroendocrine tumor patients receiving 177Lu-DOTA-TATE PRRT
What this study cannot tell us
Small cohort of only 36 patients — needs external validation in larger populations. Retrospective design. Comprehensive lesion segmentation is labor-intensive and not yet automated. Model performance may not generalize across different institutions or scanner types.
How to read the evidence
Rated moderate: innovative methodology with strong predictive performance, but limited by small sample size (36 patients) and single-center retrospective design.
When this study was published
Published in 2024. Represents state-of-the-art in imaging-based prediction for peptide-targeted cancer therapy.
The bigger picture
This represents the 'precision' in precision medicine for peptide-targeted therapy: using the very same somatostatin receptor imaging that selects patients for treatment to also predict how well they'll respond. Better prediction models help allocate this specialized therapy to those most likely to benefit.
Questions still open
- Can this lesion-level analysis be automated using AI for routine clinical use?
- Would this approach work for predicting PRRT response in other somatostatin-receptor-expressing cancers?
- Could early post-treatment scans predict response before completing all PRRT cycles?
Common questions
Can doctors predict who will respond to peptide cancer therapy?
How does this prediction method work?
Read the original research
Models using comprehensive, lesion-level, longitudinal [68Ga]Ga-DOTA-TATE PET-derived features lead to superior outcome prediction in neuroendocrine tumor patients treated with [177Lu]Lu-DOTA-TATE.
European journal of nuclear medicine and molecular imaging, 51(11), 3428-3439
Citation
Santoro-Fernandes, Victor; Schott, Brayden; Deatsch, Ali; Keigley, Quinton; Francken, Thomas; Iyer, Renuka; Fountzilas, Christos; Perlman, Scott; Jeraj, Robert. (2024). Models using comprehensive, lesion-level, longitudinal [68Ga]Ga-DOTA-TATE PET-derived features lead to superior outcome prediction in neuroendocrine tumor patients treated with [177Lu]Lu-DOTA-TATE.. European journal of nuclear medicine and molecular imaging, 51(11), 3428-3439. https://doi.org/10.1007/s00259-024-06767-x