Background/Objectives: The present study investigates the influence of advanced windowing techniques and the combination of different classification methods on the accuracy of dens axis fracture detection in computed tomography (CT) images. The aim was to evaluate and compare the diagnostic performance of two different computational models-a pure deep learning (DL) approach and a combined approach of DL segmentation, windowing, and radiomics. Methods: In this retrospective study, CT datasets of the upper cervical spine of 366 patients were included. All datasets were further divided into training, validation, and test sets. Model 1 (M1) relied on a pure DL method using a Convolutional Neural Network (CNN) and a Feedforward Neural Network (FNN), without prior manual segmentation. Model 2 (M2) incorporated a fully automatic U-Net-based segmentation followed by radiomics feature extraction and classification using a Machine Learning (ML) Classifier. The performance of both models was measured by classification accuracy, with a particular focus on the impact of CT windowing parameters and the chosen ML classification strategies. Results: M1 achieved a maximum classification accuracy of 93.7%, while M2 accomplished a classification accuracy of up to 95.7% by using ROI-based windowing and advanced feature extraction. Conclusions: Integrating advanced windowing techniques, U-Net segmentation, and radiomics improves the detection of dens axis fractures in CT imaging. This approach could enhance diagnostic accuracy and warrants further exploration and clinical integration.
Keywords: CT imaging; deep learning; dens axis fracture; machine learning; radiomics.