Application of deep learning algorithm to detect and visualize vertebral fractures on plain frontal radiographs

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Vertebral fractures (VFs) increase the risk of subsequent fractures, necessitating effective secondary prevention strategies.
• Plain abdominal frontal radiographs (PARs) are frequently obtained and offer an opportunity for opportunistic VF identification.
• Current VF diagnosis from PARs is often missed, highlighting a gap in secondary fracture prevention.

Purpose of the Study:

• To evaluate the feasibility of using a deep convolutional neural network (DCNN) for opportunistic screening, detection, and localization of VFs on PARs.
• To assess the performance of a DCNN model in identifying VFs from a PARs database.

Main Methods:

• A DCNN was pretrained on ImageNet and retrained using 1306 PAR images.
• Model performance was evaluated using accuracy, sensitivity, specificity, and AUC.
• Gradient-weighted class activation mapping (Grad-CAM) was employed for model interpretability.

Main Results:

• Only 46.6% of VFs were identified in original PAR reports.
• The DCNN model achieved 73.59% accuracy, 73.81% sensitivity, and 73.02% specificity.
• The area under the receiver operating characteristic curve (AUC) was 0.72 for VF identification.

Conclusions:

• DCNN-integrated computer-aided solutions can accurately identify VFs opportunistically on PARs.
• This approach has the potential to enhance clinician efficiency and cost-effectiveness in managing fragile fracture treatment pathways.