Radiomics-Based Differential Diagnosis of Radicular Cysts and Apical Granulomas on CBCT Images Using RadC-CNN Architecture

Background/Objectives: This study aims to evaluate the diagnostic performance of radiomic features derived from cone-beam computed tomography (CBCT) images in differentiating radicular cysts (RC) from periapical granulomas (PG). The study also compares the performance of traditional machine learning (ML) algorithms with a novel deep learning (DL) model, Radiomics Cyst Convolutional Neural Network (RadC-CNN). Methods: CBCT images of 98 patients (55 RC, 43 PG), confirmed by histopathological diagnosis, were retrospectively analyzed. Lesions were semi-automatically segmented in 3D Slicer, and 48 radiomic features were extracted. Features with high inter-observer agreement (Intraclass Correlation Coefficient ICC ≥ 0.80) were included in the analysis. Statistical tests and classification models (Decision Tree, K-Nearest Neighbors, Support Vector Machine) were used, and performance was compared to that of the proposed RadC-CNN architecture. Results: Among the 34 features with sufficient reliability, 18 showed statistically significant differences between RC and PG (p < 0.05). Shape, first-order, and texture-based features, including the Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRLM), Gray Level Size Zone Matrix (GLSZM), and Neighboring Gray Tone Difference Matrix (NGTDM), were extracted. The RadC-CNN model demonstrated superior classification performance with an accuracy of 90%, sensitivity of 90%, and precision of 91.3%, outperforming all traditional ML algorithms. Conclusions: CBCT-based radiomic analysis, particularly when combined with DL techniques like RadC-CNN, offers a promising non-invasive approach to distinguish RC from PG.