Evaluation of differential diagnosis of odontogenic lesions in cone beam computed tomography images using radiomics-based machine learning
DENTOMAXILLOFACIAL RADIOLOGY, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1093/dmfr/twag062
- Dergi Adı: DENTOMAXILLOFACIAL RADIOLOGY
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, EMBASE, MEDLINE, Biomedical Reference Collection: Corporate Edition (EBSCO)
- Hatay Mustafa Kemal Üniversitesi Adresli: Evet
Özet
Objective: The aim of this study was to evaluate the performance of machine learning (ML) algorithms based on radiomic features derived from cone beam computed tomography (CBCT) images in the differential diagnosis of radicular cysts (RC), dentigerous cysts (DC), odontogenic keratocysts (OKC), and odontomas (OD) using a noninvasive approach. Methods: CBCT images of a total of 240 patients-comprising 60 RC, 60 DC, 60 OKC, and 60 OD, all with histopathologically confirmed diagnoses-were retrospectively reviewed. Images were semi-automatically segmented using 3D Slicer software. Patients were divided into training (80%, n = 192) and test (20%, n = 48) cohorts. A total of 18 independent features were selected through intraclass correlation coefficient (ICC) filtering, Spearman correlation analysis, and LASSO-CV regularization. Ten different ML algorithms were trained and compared on the independent test set. Results: Kruskal-Wallis testing identified 107 radiomic features that exhibited statistically significant differences among the 4 groups (P < .001). The mean ICC across all features was 0.895. On the independent test set, Random Forest achieved the highest classification performance with 87.5% accuracy, a macro AUC of 0.962, and a macro F1 score of 0.871. Class-specific AUC values obtained with the one-vs-rest strategy were 0.940 for RC, 1.000 for DC, 0.910 for OKC, and 0.998 for OD. Conclusion: CBCT-based radiomic analysis offers high diagnostic accuracy in the preoperative noninvasive assessment of odontogenic lesions. Integration of this approach into routine clinical practice may contribute significantly to treatment planning prior to histopathological confirmation.