Using an Ensemble of Segmentation Methods to Detect Vertebral Bodies on Radiographs.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in radiology
• Osteoporosis screening

Background:

• Vertebral compression fractures are common indicators of osteoporosis but are frequently underdiagnosed.
• Radiologists often underreport vertebral compression fractures, leading to missed diagnoses.
• Automated tools are needed for opportunistic screening of osteoporosis via vertebral body fracture detection.

Purpose of the Study:

• To develop an ensemble of vertebral body (VB) segmentation models for lateral radiographs.
• To create a component for an automated, opportunistic screening tool for osteoporosis.
• To detect the location of thoracic and lumbar VBs and identify fractured vertebrae on lateral radiographs.

Main Methods:

• Two segmentation models, U-Net and Mask R-CNN (Region-based Convolutional Neural Network), were trained on the Osteoporotic Fractures in Men Study (MrOS) dataset.
• An ensemble model was created by combining the predictions of the U-Net and Mask R-CNN models.
• Performance was evaluated using precision, recall, F1 score, Intersection over Union (IoU), and Dice coefficient on internal and external datasets.

Main Results:

• The ensemble model achieved an F1 score of 88.34% for detecting all VBs and 97.14% for detecting severely fractured vertebrae.
• The ensemble model demonstrated strong performance on an external dataset, achieving an F1 score of 87.72% for detecting all VBs.
• The ensemble model showed superior performance compared to individual U-Net and Mask R-CNN models in VB detection.

Conclusions:

• An ensemble model combining U-Net and Mask R-CNN significantly improves the detection of vertebral bodies and fractures on lateral radiographs.
• The developed ensemble model shows excellent generalizability to external datasets, indicating its potential for real-world clinical application.
• This model serves as a crucial component for an automated opportunistic screening tool to detect vertebral fractures and aid in osteoporosis diagnosis.