Can Typical Cervical Vertebrae Be Distinguished From One Another By Using Machine Learning Algorithms? Radioanatomic New Markers

Abstract

Objective: The aim of this study is to distinguish the typical cervical vertebrae that cannot be separated from one another with the naked eye by using machine algorithms (ML) with measurements made on computerized tomography (CT) images and to show the differences of these vertebrae. 
 Method: This study was conducted by examining the 536 typical cervical vertebrae CT images of 134 (between the ages of 20 and 55) individuals. Measurements of cervical vertebrae were made on coronal, axial and sagittal section. 6 different combinations (Group 1: C3 – C4, Group 2: C3 – C5, Group 3: C3 – C6, Group 4: C4 – C5, Group 5: C4 – C6, Group 6: C5 – C6) were formed with parameters of each vertebrae and they were analyzed in ML algorithms. Accuracy (Acc), Matthews correlation coefficient (Mcc), Specificity (Spe), Sensitivity (Sen) values were obtained as a result of the analysis. 
 Results: As a result of this study, the highest success was obtained with Linear Discriminant Analysis (LDA) and Logistic Regression (LR) algorithms. The highest Acc rate was found as 0.94 with LDA and LR algorithm in Groups 3 and Group 4, the highest Spe value was found as 0.95 with LDA and LR algorithm in Group 5, the highest Mcc value was found as 0.90 with LDA and LR algorithm in Group 5 and the highest Sen value was found as 0.94 with LDA and LR algorithm in Groups 3 and 5. 
 Conclusion: As a conclusion, it was found that typical cervical vertebrae can be clearly distinguished from one another by using ML algorithms.

Ähnliche Arbeiten