Abstract
3167
Introduction: Peptide receptor radionuclide therapy (PRRT) is a well-established option for progressive, metastatic neuroendocrine tumor (NET). However, almost 30% of patients do not respond to this approach and no well-established criteria are suitable to predict response to PRRT. Radiomics aims at identifying features contained in biomedical images, analyzing them in various scenarios such as the patients' outcome prediction. Therefore, we aimed to develop a radiomics predictive model of response to PRRT in progressive, metastatic gastroenteropancreatic (GEP) NET analyzing [68Ga]DOTATOC PET/CT images pre-PRRT
Methods: We retrospectively analyzed 324 SSTR-2-positive lesions from 38 GEP-NET patients (9 G1, 27 G2 and 2 G3) who underwent restaging [68Ga]DOTATOC PET/CT before the start of complete PRRT with [177Lu]DOTATOC. Clinical, laboratory and radiological follow-up data were collected for at least 6 months after the last cycle. We used LifeX software to extract 65 features from PET data for each lesion based on a manually placed standardized-size region of interest (0.443 cm3) in the most active part of each lesion. Grading (G1-G2-G3), number of PRRT cycles, PRRT cumulative activity, pre- and post-PRRT chromogranin A (CgA) values were also considered as additional clinical features. [68Ga]DOTATOC PET/CT follow-up with the same scanner for each patient determined the disease status (progression vs response in terms of stability/reduction/disappearance) for each lesion. All features (PET and clinical) were correlated to the follow-up data also in a per-site analysis (liver, lymph nodes, and bone), and for features significantly associated with response, the delta-radiomics for each lesion was assessed on follow-up [68Ga]DOTATOC PET/CT performed until 9 months post-PRRT. A statistical system based on the point-biserial correlation and the logistic regression analysis was used for the reduction and selection of the features; the Discriminant Analysis was used, instead, to obtain the predictive model using the k-fold strategy to split data into training and validation sets.
Results: From the reduction and selection process, HISTO_Skewness and HISTO_Kurtosis resulted able to predict the disease status at follow-up with an area under the receiver operator characteristic curve (AUC ROC), sensitivity, and specificity of 0.745, 80.6%, 67.2% and 0.722, 61.2%, 75.9%, respectively. Also, the combination of 3 features (2 from PET: HISTO_Skewness; HISTO_Kurtosis and one clinical: Grading) reached an AUC ROC, sensitivity, and specificity of 0.744, 66.4%, and 70.3%, respectively. Differently, the SUVmax was not significant (p = 0.49) to predict the response to PRRT in terms of progression vs objective benefit or response (AUC ROC 0.523, sensitivity 36.7%, specificity 63.3%) as shown in figure 1. At Δradiomics analysis for the two significant features, in responsive/stable lesions we observed a mean percentage reduction for ΔHISTO_Skewness (-3.3 ± 664.3%) and a mean percentage increase for ΔHISTO_Kurtosis (16 ± 71.4%); for progressive lesions we observed a mean percentage increase for ΔHISTO_Skewness (112.5 ± 348.3%; p = 0.209) and ΔHISTO_Kurtosis (5.8 ± 52.3%; p = 0.255), less evident than responsive/stable lesions.
Conclusions: The presented preliminary “theragnomics” model proved to be superior to conventional quantitative parameters to predict the response of GEP-NET lesions in patients treated with complete [177Lu]DOTATOC PRRT, regardless of the lesions' site.