Radiomics on 177Lu-DOTATATE Posttreatment Scans: Feasibility, Preprocessing Optimization, and Planar-SPECT Comparison.

Abstract

¹⁷⁷Lu-DOTATATE peptide receptor radionuclide therapy (PRRT) is an established therapeutic option for patients with neuroendocrine tumors (NETs). Although radiomics has been increasingly applied in NET research, it was developed almost exclusively using ⁶⁸Ga-DOTATOC PET imaging, whereas PRRT-related scintigraphic acquisitions remain largely unexplored. This study evaluates the feasibility of extracting hepatic radiomic features from posttreatment whole-body planar scintigraphic (WBS) and tomographic (SPECT) scans, enabling the optimization of preprocessing strategies and a direct comparison of radiomic profiles between imaging modalities. Methods: We conducted a retrospective analysis of data for 49 patients with well-differentiated, progressive metastatic gastroenteropancreatic NETs from 2 centers. All patients underwent posttreatment imaging within 2 d after each ¹⁷⁷Lu-DOTATATE PRRT cycle, for a total of 183 WBS and 108 SPECT scans. Preprocessing operations included image-intensity normalization, dividing each image by the mean counts of a region of interest positioned on the spleen (S-norm) and the third lumbar vertebra (V-norm). Radiomic features were extracted with no spatial resampling, using 64-bin discretization and relative intensity rescaling. Results: Texture extraction was successfully performed on all planar and SPECT images from both centers. Correlations between intensity-based features measured using S-norm and V-norm images showed weak yet statistically significant r values. Conversely, there was a robust correlation between histogram-based and second-order features measured on S-norm and V-norm images, reflected in high r values. V-norm intensity-based features exhibited significantly higher values and lower SDs when compared with S-norm intensity-based variables. When comparing WBS and SPECT scans, a strong correlation was observed between intensity-based variables, whereas histogram-based and second-order features showed no significant correlation. Conclusion: Radiomic analysis on ¹⁷⁷Lu-DOTATATE posttreatment scans was feasible, with proportional scaling to third-lumbar-vertebra activity resulting in the most effective intensity normalization approach, given the decreased variability compared with scaling to spleen activity. Correlations between features assessed using the 2 different scaling methods were weak for intensity-based features and strong for histogram-based and second-order features. Except for intensity-based features, radiomic features derived from WBS images were not comparable with those obtained from SPECT scans.