Results
A total of 200 radiological examinations was evaluated in this work. Table 1 represents the baseline characteristics of the CM-I group. The mean age was 29.92 ± 15.03 years, and the study population consisted of 55 females and 45 males. The primary presenting symptoms were headache (62%), neck pain (30%), sensorial disturbance (18%), and paraparesis (15%). Syringomyelia and retro-odontoid were detected in 34% and 8% of patients, respectively.
The measured parameters were identified to ML algorithms, and all of the morphometric measures were significantly different between the groups, except for the distance from dens axis to posterior margin of FM (F10). Table 2 represents the statistics regarding the measured parameters.
For the first parameter (F1) XGBoost, SGB, Bagged CART, Random Forest, and Logistic Regression values were found to be 100% as expected. For the other parameters (F2-11), the most accurate values were 0.88 for F6 and F10 (Random Forest model), 0.87 for F5 (Random Forest model), and 0.86 for F2 (Random Forest model). The accuracy of the Logistic Regression model for F2-11 was calculated as 0.89, whereas it was 100% for other models. Detailed data are presented in Table 3.
The results indicate that the Random Forest model has produced the best 1.0 (14 of 14) ratio of accuracy regarding 14 different combinations of morphometric features. This ratio is found to be 0.50, 0.29, 0.21, and 0.07 in the Bagged CART, Stochastic Gradient Boosting, XGBoost, and Logistic Regression models, respectively.