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

Area of Science:

• Nuclear medicine
• Oncology
• Radiology

Background:

• Somatostatin receptor (SSTR) imaging is crucial for predicting treatment outcomes in neuroendocrine tumor (NET) patients undergoing peptide receptor radionuclide therapy (PRRT).
• Previous studies focused on limited imaging features, neglecting comprehensive, longitudinal, and lesion-level analyses which are vital for understanding treatment heterogeneity.

Purpose of the Study:

• To evaluate the predictive power of comprehensive, longitudinal, lesion-level ⁶⁸Ga-SSTR-PET features combined with a multivariate linear regression (MLR) model for PRRT outcomes in NET patients.
• To assess if this combined approach can capture disease heterogeneity and improve outcome prediction compared to existing methods.

Main Methods:

• Retrospective analysis of 36 NET patients treated with [¹⁷⁷Lu]Lu-DOTA-TATE and imaged with [⁶⁸Ga]Ga-DOTA-TATE at baseline and post-therapy.
• Segmentation, anatomical labeling, and longitudinal matching of all metastatic lesions were performed.
• Lesion-level uptake and variation were measured, and patient-level features were engineered for MLR modeling of progression-free survival (PFS), with validation using concordance index, ROC analysis, and survival analysis.

Main Results:

• The MLR model achieved a concordance index of 0.826 and an AUC of 0.88, demonstrating excellent predictive capability.
• Significant patient stratification was achieved (p<.001) with a hazard ratio of 3⨯10^-5.
• The MLR approach significantly outperformed single feature predictors in benchmark comparisons.

Conclusions:

• Comprehensive, lesion-level, longitudinal ⁶⁸Ga-SSTR-PET analysis combined with MLR modeling provides superior prediction of PRRT outcomes in NET patients.
• This integrated approach effectively captures disease heterogeneity and outperforms predictions based on non-comprehensive, patient-level, or single time-point features.