Towards Improved Identification of Vertebral Fractures in Routine Computed Tomography (CT) Scans: Development and...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Bone Health and Osteoporosis

Background:

• Vertebral fractures (VFs) are common osteoporosis indicators, often missed in computed tomography (CT) scans.
• Undetected VFs contribute to significant patient morbidity and mortality.
• Current identification methods in routine CT scans have limitations.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) algorithm for detecting VFs in abdominal/chest CT scans.
• To assess the algorithm's performance on an independent external validation dataset.
• To explore the potential of ML in improving VF identification and osteoporosis management.

Main Methods:

• Acquired two independent datasets of routine abdominal/chest CT scans (n=1011 training, n=2000 validation) from patients aged 50+.
• Trained four ML models using cross-validation and created an ensemble model.
• Validated the ensemble model against Genant semiquantitative (SQ) grading for reference standard VF readings.

Main Results:

• The ML ensemble model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.88.
• High performance metrics included accuracy (0.92), kappa (0.72), sensitivity (0.81), and specificity (0.95).
• The algorithm demonstrated strong performance in identifying moderate to severe VFs on the external validation set.

Conclusions:

• A machine learning algorithm can effectively detect vertebral fractures in CT scans.
• The developed algorithm shows significant potential to assist healthcare professionals in early VF identification.
• Improved VF detection can lead to timely intervention and prevention of subsequent fragility fractures.