AI-based Quantitative Volumetric [68Ga]Ga-DOTATOC PET/CT Image Analysis in Predicting Response to Peptide Receptor Radionuclide Therapy and Disease Outcome.

Aim: To evaluate the predictive value of [68Ga]Ga-DOTATOC PET/CT for treatment response and disease outcomes, using a prototype convolutional neural network (pCNN) for automated detection and semi-quantitative analysis in patients with metastatic neuroendocrine tumours (NETs) receiving peptide receptor radionuclide therapy (PRRT).
Materials and Methods: This retrospective study included patients with histologically proven well-differentiated metastatic NET treated with [177Lu]Lu-DOTATATE who underwent baseline (bPET) and post-treatment (pPET) [68Ga]Ga-DOTATOC PET/CT after the four cycles of [177Lu]Lu-DOTATATE treatment. Lesion detection and semi-quantitative PET parameters were extracted using a pCNN. Standardized uptake values (SUVs), as well as whole-body somatostatin-expressed total tumour volume (SSTR-TV) and total lesion SSTR expression (SSTR-TLA) were analyzed. PERCIST criteria were used to assess the response to treatment. Progression-free survival (PFS) and overall survival (OS) were defined from the first cycle of PRRT.
Results: From the 19 patients with metastatic NET (6 females and 13 males), 10 (52.6%) demonstrated partial response, 3 (15.8%) showed stable disease, 3 (15.8%) exhibited completed response, and 3 (15.8%) experienced disease progression. Among semi-quantitative bPET parameters, the SUVmax-to-liver background ratio of the most avid lesion was significantly higher in responders. Analysis of pCNN-based volumetric bPET parameters revealed significantly higher SSTR-TLA (p=0.003) and total SSTR-TV (p=0.005) in responders. The median PFS in patients was 24 months. From the bPET-derived parameters, SULpeak-to-spleen background ratio and total SUVmean of the most avid lesion were significant predictors of PFS (p=0.009 and p=0.021, respectively). Among the volumetric PET parameters, ∆total SSTR-TV and ∆total SSTR-TLA ratio were identified as significant predictors of PFS (p=0.035 and p=0.038, respectively). The median OS in patients was 39 months. Total SUVmean significantly predicted OS (p<0.001), as also did ∆total SSTR-TV from the volumetric parameters (p=0.020).
Conclusion: This study provides preliminary evidence supporting the use CNN-based automated volumetric whole-body assessment of [68Ga]Ga-DOTA-peptide PET/CT images in metastatic NET patients undergoing PRRT. The findings highlight the feasibility and prognostic value of this approach, warranting validation in larger patient cohorts.