To evaluate the preoperative value of ML-driven classification based on conventional 18F-FDG PET metrics and clinical information for forecasting features of EC aggressiveness. A retrospective analysis was conducted on 123 patients with EC who received 18F-FDG PET scans for preoperative staging between 2009 and 2021. SUVmax, SUVmean, metabolic tumor volume (MTV), and total lesion glycolysis (TLG) were measured from the primary lesion. Patient age and body mass index (BMI) were recorded. Histological subtype, depth of myometrial invasion (MI), risk category, lymph node (LN) status, and p53 expression were obtained from pathology reports. The cohort was randomly partitioned into training (80%) and validation (20%) subsets. The training data served for feature selection (via Mann-Whitney U test and ROC analysis) and model development, while the validation set assessed predictive performance. In the validation cohort, the highest accuracies achieved were: 61% using TLG alone versus 87% with ML for deep MI; 71% using SUVmax alone versus 79% with ML for risk group classification; 72% using TLG alone versus 83% with ML for LN involvement; and 45% using SUVmax or SUVmean alone versus 73% with ML for p53 expression. Classification models based on ML integrating standard 18F-FDG PET parameters and clinical data effectively characterized the examined features of EC aggressiveness. This offers a non-invasive approach to aid preoperative risk stratification in EC patients.
