Evaluation of FET PET radiomics feature repeatability in patients with brain tumors

Ziel/Aim Amino acid PET using the tracer O-(2-[F-18]fluoroethyl)-L-tyrosine (FET) has become an important diagnostic tool in neurooncology. Consequently, the potential of FET PET radiomics in patients with brain tumors has been investigated. Feature-based radiomics builds models based on predefined features from routine medical imaging data supporting clinical decision making. However, conclusiveness of the developed models strongly depends on generalizability of the investigated features. Hence, repeatability of feature-based FET PET radiomics was evaluated.

Methodik/Methods A test-retest analysis based on equivalent, but statistically independent subsamples of PET images, was performed in a cohort of 51 glioma patients. A total of 1,302 radiomics features were calculated for tumor volumes and spheres of different sizes positioned in the tumor and in healthy brain tissue. Feature repeatability was assessed by the intraclass correlation coefficient (ICC). Finally, a hierarchical cluster analysis for the differentiation of patients based on the isocitrate dehydrogenase (IDH) genotype, an important prognostic biomarker, was performed.

Ergebnisse/Results For tumor volumes, 73% of first order features and 71% of features extracted from the grey level co-occurrence matrix showed high repeatability (ICC CI95% > 0.9). In the largest spheres (volume, 33 ml), 67% of calculated features showed a high repeatability (ICC CI95% > 0.9), which significantly decreased in small spheres (volume, < 4ml). The IDH genotype had no effect on feature repeatability. Based on a subset of 297 repeatable features, two clusters differentiating glioma patients according to the IDH genotype were determined.

Schlussfolgerungen/Conclusions The developed method allows investigating PET radiomics feature repeatability based on clinical PET scans. A large number of robust FET PET features could be identified, potentially useful towards further standardization of radiomics analysis in neurooncology.

Publication History

Article published online:
08 April 2021