Development of a Radiomic Model to Detect the Retromolar Canal on Panoramic Radiographs

The retromolar canal (RMC) is an extension of the mandibular canal located in the distal region of the mandibular third molar. Accurately detecting the RMC using conventional two-dimensional images is challenging, potentially leading to anesthetic failure and sensory disorders. This study aims to explore the clinical application of a radiomic model based on panoramic radiographs in detecting the RMC.

A retrospective collection of cone beam computed tomography (CBCT) and panoramic radiographs was conducted on 800 patients, covering 1555 hemimandibles. CBCT images served as the gold standard for confirming the presence of RMC. A dataset comprising 846 retromolar regions was established for model training and testing, with an 8:2 ratio. On the panoramic radiographs, the retromolar regions were delineated as the regions of interest, and radiomic features were extracted and selected. Support vector machine (SVM), logistic regression (LR), k-nearest neighbors (KNN), and multilayer perceptron (MLP) were employed to construct detection models for the RMC. The performance of these algorithms was assessed using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA), and the area under the receiver operating characteristics curve (AUC) values were compared with those of a dentist and a radiologist.

The RMC was identified in 423 (27.2%) out of 1555 hemimandibles on CBCT images. The four algorithms, particularly SVM and MLP, demonstrated outstanding classification abilities in detecting the RMC, with AUC values ranging from 0.831 to 0.895 in the training set and from 0.719 to 0.808 in the testing set. These results significantly surpassed those of the dentist and radiologist (p < 0.05).

Radiomics based on panoramic radiographs exhibit a high detection capability for the RMC, emphasizing its considerable clinical application value.