Dentin-Normalized CBCT Radiomics within a Periapical Lesion Differential Diagnosis System for Differentiating...

Area of Science:

• Oral and Maxillofacial Radiology
• Medical Imaging Informatics
• Computational Pathology

Background:

• Periapical granulomas and radicular cysts are difficult to distinguish using conventional methods.
• Histopathology is definitive but invasive; cone-beam computed tomography (CBCT) radiomics offers a noninvasive alternative.
• CBCT radiomics performance is hindered by image intensity variability.

Purpose of the Study:

• To evaluate a dentin-based intensity normalization method.
• To enhance the robustness and accuracy of CBCT radiomics for periapical lesion classification.

Main Methods:

• Retrospective analysis of CBCT scans from 140 patients with confirmed periapical lesions (100 granulomas, 40 cysts).
• Dentin-adjacent to the lesion was used as an internal reference for affine intensity normalization.
• Radiomic features were extracted before and after normalization; models were evaluated using various feature sets and classifiers (XGBoost, logistic regression).

Main Results:

• Dentin normalization significantly improved the area under the curve (AUC) for full radiomics models using XGBoost (0.68 to 0.78, P<.05).
• Normalization enhanced AUC in first-order/texture settings for both classifiers, indicating improved gray-value-dependent features.
• The highest performance (AUC 0.78) was achieved with dentin-normalized full radiomics using XGBoost.

Conclusions:

• Dentin-based intensity normalization improves the performance and stability of CBCT radiomics for differentiating periapical granulomas and radicular cysts.
• This approach offers a practical internal reference for stabilizing radiomic features, with first-order and texture features providing valuable complementary information.
• Further validation across diverse CBCT systems and protocols is needed for clinical implementation.