External validation of a convolutional neural network algorithm for opportunistically detecting vertebral fractures...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Vertebral fractures (VFs) are a common skeletal fragility, often underdiagnosed.
• Early identification of VFs is crucial for timely intervention and management.
• Computed Tomography (CT) scans are widely used for medical imaging, including the detection of VFs.

Purpose of the Study:

• To evaluate the diagnostic performance of a Convolutional Neural Network (CNN) model for automatic detection of vertebral fractures (VFs).
• To assess the CNN model's efficacy in an external validation cohort using CT scans.
• To determine the model's accuracy, sensitivity, and specificity in identifying VFs.

Main Methods:

• Retrospective selection of CT scans (chest, abdomen, thoracolumbar spine) from 4,810 Chinese patients (≥50 years).
• Assessment of VFs using the semiquantitative Genant classification by blinded readers.
• Evaluation of the CNN model's performance using AUROC, accuracy, kappa, sensitivity, and specificity.

Main Results:

• The CNN model achieved an AUROC of 0.94 for identifying scans with moderate to severe VFs.
• The algorithm demonstrated high accuracy (93%), sensitivity (94%), and specificity (93%).
• Performance metrics indicate robust detection capabilities of the CNN model.

Conclusions:

• The developed CNN algorithm shows excellent performance in detecting VFs in CT scans.
• The AI tool has the potential to support radiologists in the early identification of VFs.
• External validation confirms the algorithm's utility in routine clinical practice for Chinese patients.