Cervical Spondylosis Diagnosis Based on Convolutional Neural Network with X-ray Images

Abstract

The increase in Cervical Spondylosis cases and the expansion of the affected demographic to younger patients have escalated the demand for X-ray screening. Challenges include variability in imaging technology, differences in equipment specifications, and the diverse experience levels of clinicians, which collectively hinder diagnostic accuracy. In response, a deep learning approach utilizing a ResNet-34 convolutional neural network has been developed. This model, trained on a comprehensive dataset of 1235 cervical spine X-ray images representing a wide range of projection angles, aims to mitigate these issues by providing a robust tool for diagnosis. Validation of the model was performed on an independent set of 136 X-ray images, also varied in projection angles, to ensure its efficacy across diverse clinical scenarios. The model achieved a classification accuracy of 89.7%, significantly outperforming the traditional manual diagnostic approach, which has an accuracy of 68.3%. This advancement demonstrates the viability of deep learning models to not only complement but enhance the diagnostic capabilities of clinicians in identifying Cervical Spondylosis, offering a promising avenue for improving diagnostic accuracy and efficiency in clinical settings.

Keywords: X-ray classification; cervical spondylosis; deep learning; multi-label.

Figure 1

F1 score of ResNet-34 model from validation results. The abscissa is the F1-score of each validation from training the model, and the ordinate is the number of times the model is trained. We saved the final CNN from the validation with the best performance within 100 epochs.

Figure 2

CNN model flow chart. We categorize the input data into a test set and a training-validation set. The data in the training-validation set undergo manual labeling, and the results are fed into the subsequent step. Subsequently, preprocessing is applied to the training-validation set data. Following preprocessing, the training set data are employed to train the ResNet-34 model, while the validation set data are utilized for selecting the final model. The ultimately saved model is then evaluated for performance using the test dataset.

Figure 3

Confusion matrix for classification of cervical vertebrae X-ray projection position on test dataset.

Figure 4

Confusion matrix for classification of CS on test dataset (P: positive, N: negative).

Figure 5

Confusion matrix for classification of CS from different positions. (a) is from anteroposterior positions; (b) is from lateral positions; (c) is from left oblique positions; (d) is from right oblique positions.

Figure 6

Cervical vertebra X-ray projection position classification and CS classification results. Figure (a–c) represent that the cervical spine position is correctly classified but CS is incorrectly classified. Figure (d) represents that the cervical spine position is classified incorrectly but CS is correctly classified. Figure (e–h) represent images with correct cervical vertebra position classification and CS classification. (“N” indicates CS is negative; “P” indicates CS is positive; the red line indicates the physiological curvature of the cervical spine; the red dotted arrow indicates the hyoid bone blocking the cervical vertebrae. the yellow arrow indicates the intervertebral space; the yellow dotted arrow indicates osteophyte formation; the blue dotted arrow indicates ligament calcification; and the blue arrow indicates the intervertebral foramen).