Opportunistic osteoporosis screening in multi-detector CT images using deep convolutional neural networks

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Primary osteoporosis diagnosis often relies on bone mineral density (BMD) measurements.
• Current methods may not be fully automated or opportunistic.
• Deep learning offers potential for automated analysis of medical images.

Purpose of the Study:

• To develop and evaluate a fully automatic deep learning method for vertebral body segmentation and BMD calculation in CT images.
• To assess the accuracy of deep learning in identifying osteoporosis, osteopenia, and normal bone density.
• To explore the application of deep convolutional neural networks (DCNNs) for opportunistic screening.

Main Methods:

• A retrospective study used 1449 patients' spinal or abdominal CT scans from three vendors.
• U-Net was employed for automated vertebral body segmentation.
• DenseNet-121 was used for BMD calculation, with quantitative computed tomography (QCT) as the standard.

Main Results:

• Automated segmentation of lumbar vertebrae (L1-L4) showed high correlation with manual segmentation (minimum Dice coefficient > 0.78).
• Automated BMD calculations highly correlated (r > 0.98) and agreed with QCT-derived values across different CT vendors.
• The deep learning model demonstrated robust performance on diverse datasets.

Conclusions:

• A deep learning-based approach enables fully automatic identification of osteoporosis, osteopenia, and normal bone density from CT images.
• The method provides accurate vertebral segmentation and BMD estimation, comparable to established methods.
• This technique facilitates opportunistic osteoporosis screening during routine CT examinations.