Detection of Cervical Foraminal Stenosis from Oblique Radiograph Using Convolutional Neural Network Algorithm

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Cervical foraminal stenosis diagnosis often relies on MRI, which can be costly and time-consuming.
• Oblique radiographs are a more accessible imaging modality, but their diagnostic accuracy for foraminal stenosis can be limited.
• Developing automated diagnostic tools can improve efficiency and accuracy in clinical settings.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) algorithm for diagnosing cervical foraminal stenosis using oblique radiographs.
• To compare the diagnostic accuracy of the CNN algorithm against human expert interpretation.

Main Methods:

• A dataset of 997 patients with cervical MRI and oblique radiographs was used.
• Oblique radiographs were labeled for foraminal stenosis based on MRI ground truth.
• A CNN model (DenseNet161) was developed using data augmentation, preprocessing, and transfer learning.
• Gradient-weighted class activation mapping (Grad-CAM) was used for model visualization.

Main Results:

• The CNN model achieved an area under the curve (AUC) of 0.889.
• The model demonstrated high performance with an F1 score of 88.5%, accuracy of 84.6%, precision of 88.1%, and recall of 88.5%.
• The CNN's accuracy significantly surpassed that of two orthopedic surgeons (64.0% and 58.0%).
• Grad-CAM analysis indicated the CNN focused on foraminal and disc space regions.

Conclusions:

• A CNN algorithm effectively detects cervical neural foraminal stenosis from oblique radiographs.
• The developed CNN shows promising results with high AUC, F1 score, and accuracy.
• Cervical oblique radiography, enhanced by this CNN model, could serve as an effective screening tool for neural foraminal stenosis.