[Diagnostic value of random forest algorithm based on condylar CBCT features for the staging of patients with...

To evaluate the diagnostic value of a random forest (RF) algorithm based on CBCT features of the condyle for staging idiopathic condylar resorption(ICR).

A total of 120 patients with ICR admitted to Nanjing Stomatological Hospital from January 2016 to December 2024 were selected and divided into the stationary phase group (n=70) and the progressive phase group (n=50). Healthy individuals who underwent physical examinations during the same period (n=40) were selected as the control group. The CBCT features of condylar length, inner 1/3 width, middle 1/3 width, outer 1/3 width, condylar height, mandibular ramus height, condylar surface angle, condylar axis angle, condylar volume, condylar surface area and condylar cross-sectional area were measured.

There were no significant differences in gender, age, height and disease duration among each group(P＞0.05), while there were significant differences in CBCT characteristic indicators (P＜0.05). Compared with the stationary group, the condylar length, condylar height, mandibular elevation height, condylar volume, condylar surface area and condylar cross-sectional area in the progressive group were significantly reduced(P＜0.05). Taking the stationary or progressive stage of ICR patients as the dependent variable, the variables with significant differences in univariate analysis were included in the random forest (RF) algorithm, the order of importance from high to low was condylar volume, condylar surface area, condylar length, mandibular elevation height, condylar cross-sectional area, and condylar height. The above-mentioned CBCT features were subjected to stepwise RF,when the number of features was 5, the error rate of the out-of-pocket data was the lowest. The order of importance from high to low was condylar volume, condylar surface area, condylar length, mandibular elevation height and condylar area. Using the top 5 variables from the RF importance ranking for analysis, condylar volume, condylar surface area, condylar length and condylar cross-sectional area were identified as influencing factors for ICR staging (P＜0.05). The receiver operating characteristic (ROC) curve analysis demonstrated that condylar volume, condylar surface area, condylar length, condylar cross-sectional area and their combination could all be used to diagnose ICR staging(P＜0.05), with area under the curve(AUC)＞0.70, sensitivity＞58.5%, and specificity≥64%. Compared with a single indicator, the AUC, sensitivity and specificity of the combined detection were significantly increased.

RF based on the CBCT features of the condyle of ICR patients has a high diagnostic value for disease staging. Condylar volume, condylar surface area, condylar length and condylar cross-sectional area are the key imaging indicators for differentiating the stationary phase from the progressive phase, among which condylar volume is of the highest importance. The combined detection of multiple features can significantly improve diagnostic efficiency, manifested as higher AUC, sensitivity and specificity.