Radiomic Analysis of Mandibular Bone Changes in Patients with Chronic Myeloid Leukemia on Panoramic Radiographs

To investigate whether panoramic radiograph-based radiomic features identify mandibular bone alterations associated with chronic myeloid leukemia (CML) and distinguish patients from healthy controls. This retrospective case-control study included 19 patients with CML and 57 healthy controls. Radiomic features were extracted from the mandibular cortical bone, trabecular bone, and ramus. Features were selected using LASSO regression and correlation filtering (r ≥ 0.90). Logistic regression models were evaluated by ROC analysis, and multivariable regression identified independent predictors. Three features were retained: first-order minimum, first-order 10th percentile, and GLRLM Run Length Non-Uniformity. The three-feature model achieved an AUC of 0.868 (95% CI, 0.755-0.981). Adding sex did not improve performance (AUC = 0.871; p = 0.762). The first-order minimum model achieved an AUC of 0.822, with no improvement after adding sex (AUC = 0.825; p = 0.594). First-order minimum was the only independent predictor (OR = 1.120; 95% CI, 1.041-1.224; p = 0.006). Cortical bone performed best in the single-feature model, whereas the ramus performed best in the three-feature model. Panoramic radiograph-based radiomics identified mandibular bone alterations associated with CML. A three-feature radiomics signature showed good diagnostic performance and may serve as a non-invasive imaging biomarker.