RT Journal Article SR Electronic T1 A machine learning approach for the analysis of radiomic features of pretreatment 18F-FDG PET/CT to predict prognosis of patients with endometrial cancer JF Journal of Nuclear Medicine JO J Nucl Med FD Society of Nuclear Medicine SP 130 OP 130 VO 62 IS supplement 1 A1 Masatoyo Nakajo A1 Megumi Jinguji A1 Atsushi Tani A1 Takashi Yoshiura YR 2021 UL http://jnm.snmjournals.org/content/62/supplement_1/130.abstract AB 130Objectives: To examine the prognostic significance of pretreatment 2-deoxy-2-[18F]fluoro-D-glucose (18F-FDG) PET-based radiomic features using a machine learning approach in patients with endometrial cancers. Methods: This retrospective study included 53 patients with endometrial cancers who underwent 18F-FDG-PET/CT before treatment. Four clinical (age, histological type, stage, and treatment procedures) and 40 18F-FDG PET-based radiomic features were used to predict disease progression employing machine learning approaches. A ranking-based feature selection method with the Gini coefficient was used to reduce feature dimensions. Clinical and radiomic features were ranked based on the Gini coefficient score derived by evaluating their associations with the disease progression. Nine feature subsets were selected to identify the optimal feature selection size (range, 5-44) in steps of 5. The machine learning algorithms (random forest, neural network, k-nearest neighbors (kNN), naive Bayes, logistic regression, and support vector machine) were compared using the areas under the receiver operating characteristic curve (AUC) and validated by the random sampling method. Progression-free survival (PFS) and overall survival (OS) were assessed by Cox regression analysis. Results: Twenty-six patients showed eventual progression (17 patients expired and 9 patients were alive). Twenty-seven patients were alive without progression during clinical follow-up. The five best predictors of disease progression were coarseness, gray-level nonuniformity for zone (GLZLM_GLNU), gray-level run length nonuniformity (GLRLM_RLNU), stage, and metabolic tumor volume (MTV). The patients with progression showed significantly lower coarseness (3.1 ± 3.0 [×10−3] vs. 12.2 ± 1.2 [×10−3], p < 0.001), higher GLZLM_GLNU (127.2 ± 178.6 vs. 25.1 ± 23.5, p < 0.001), higher GLRLM_RLNU (11897.0 ± 27559.9 vs. 1249.2 ± 1301.8, p < 0.001), higher stage (p = 0.001), and larger MTV (481.4 ± 1091.0 vs. 41.2 ± 47.0, p < 0.001) than those without progression. The kNN model obtained the best performance classifiers for predicting the disease progression when ten features subset was used (AUC =0.881, accuracy =0.830, F1 score =0.828, precision =0.818, and recall =0.830). Coarseness which was the first ranked radiomics was selected for survival analyses, and the 5-year PFS (64.9% vs. 22.0%, p < 0.001) and 5-year OS (77.5% vs. 41.8%, p < 0.001) rates were significantly higher in the high- (> 4.5×10−3) than low-coarseness groups (≤ 4.5×10−3). Only coarseness remained the significant and independent factor both for PFS (hazard ratios (HR), 0.68; 95% CI, 0.57-0.88; p=0.005) and OS (HR, 0.55; 95% CI, 0.39-0.77; p<0.001) at multivariate Cox regression analysis. Conclusions: 18F-FDG PET-based radiomic analysis using a machine learning approach may be useful for predicting the tumor progression and prognosis in patients with endometrial cancers.