Application of PET imaging delta radiomics for predicting progression-free survival in rare high-grade glioma

Abstract

Abstract This study assesses the feasibility of using radiomics changes over time to predict progression-free survival in rare diseases. High-grade glioma patients (n = 53) underwent L-3,4-dihydroxy-6-[18F]-fluoro-phenylalanine (18F-FDOPA) positron emission tomography (PET) scans at the time of temozolomide chemotherapy discontinuation. Among these patients, 18 had previously undergone an 18F-FDOPA PET during treatment. Radiomics features from static/dynamic parametric images, and conventional features, were extracted. After excluding highly correlated features, various feature selection methods and time-to-event survival algorithms were employed to generate 16 model combinations. Delta radiomics features, as relative and absolute changes, were either computed using informative features derived from the entire cohort or directly selected from the subset of 18 patients, and performances evaluated with the cross-validation technique. Delta-absolute radiomics achieved the highest performance when the pipeline was applied to the 18-patient subset (combination of support vector machine (SVM) and recursive feature elimination (RFE): C-index = 0.783 [0.744–0.818]). This result was comparable to using top-rank features from all 53 patients (SVM + RFE: C-index = 0.730 [0.700–0.760], p = 0.0034) and significantly outperformed delta-absolute conventional features (C-index = 0.584 [0.548–0.620], p < 0.001) and single-time-point radiomics features (C-index = 0.546 [0.512–0.580], p < 0.001). This highlights the considerable potential of delta radiomics to outperform delta conventional features and single-time-point radiomics features, in rare cancer cohorts.