Machine learning predicts PRRT eligibility for neuroendocrine tumors using clinical and lab parameters alone | Research

The takeaway

Mathematical models using clinical parameters (tumor origin, treatments, CK7) predicted somatostatin receptor expression with 70-83% accuracy in 65 metastatic NEN patients, potentially reducing the need for receptor imaging.

70-83% prediction accuracy
Clinical parameters alone can predict PRRT eligibility in NEN patients, potentially reducing the need for expensive receptor imaging

What the researchers found

Models predicted SSTR expression with 70-83% accuracy using clinical parameters. Key predictors: tumor origin, oncological treatments, CK7. Associations: age, grade, disease extent, CEA, CA19-9, AFP. 65 patients, 392 lesions evaluated.

Why it matters

PRRT requires expensive somatostatin receptor imaging for patient selection. Predictive models using routine clinical data could pre-screen patients, reducing costs and accelerating treatment decisions.

How the study worked

Retrospective study of 65 patients with metastatic NENs. SPECT/CT imaging with [99mTc]Tc-EDDA/HYNIC-TOC. Mathematical models built from histological, oncological, immunohistochemical, and laboratory parameters.

What this study cannot tell us

Retrospective single-center. Moderate sample size. Prediction accuracy (70-83%) not perfect. External validation needed. Used SPECT/CT rather than PET/CT.

How to read the evidence

Retrospective single-center study with biostatistical modeling. Good proof of concept needing external validation.

When this study was published

Published in 2025.

The bigger picture

AI-driven patient selection is increasingly important in precision oncology. This approach could democratize PRRT access by enabling centers without expensive imaging to identify likely candidates.

Questions still open

Can these models be validated at other centers?
Would combining clinical predictors with liquid biopsy improve accuracy?
Could this approach reduce time-to-PRRT?

Common questions

How can AI help with cancer treatment selection?

This study used mathematical models to predict whether a patient's tumor would respond to a specific targeted radiation therapy (PRRT) based on routine blood tests and pathology results. This could help doctors quickly identify candidates without waiting for specialized imaging.

What is somatostatin receptor expression and why does it matter?

Neuroendocrine tumors often express somatostatin receptors on their surface. PRRT uses radiolabeled peptides that bind to these receptors to deliver targeted radiation. Patients with high receptor expression are the best candidates—this study shows clinical data can predict expression levels.

Read the original research

Machine Learning Uncovers Novel Predictors of Peptide Receptor Radionuclide Therapy Eligibility in Neuroendocrine Neoplasms.

Cancers, 17(17)

View on PubMed ↗Publisher’s paper ↗

Citation

Sipka, Gábor; Farkas, István; Bakos, Annamária; Maráz, Anikó; Mikó, Zsófia Sára; Czékus, Tamás; Bukva, Mátyás; Urbán, Szabolcs; Pávics, László; Besenyi, Zsuzsanna. (2025). Machine Learning Uncovers Novel Predictors of Peptide Receptor Radionuclide Therapy Eligibility in Neuroendocrine Neoplasms.. Cancers, 17(17). https://doi.org/10.3390/cancers17172935