anduin

anduin performs automated segmentation and quantitative analysis of vertebrae in CT NIFTI images to compute trabecular and integral volumetric bone mineral density (vBMD), CT-based areal BMD (aBMD), bone mineral content (BMC), and vertebral centroids for spinal bone assessment.

Key Features:

• Automated Segmentation: Utilizes convolutional neural networks (CNNs) to segment vertebrae from CT NIFTI images.
• Volumetric BMD Calculation: Calculates trabecular vBMD and integral vBMD for segmented vertebral regions.
• CT-based aBMD and Calibration: Derives CT-based areal BMD (aBMD) using asynchronous calibration.
• Bone Mineral Content (BMC) Extraction: Extracts vertebral bone mineral content (BMC) from segmented vertebral bodies.
• Data Export: Exports vertebral segmentations as NIFTI files and vertebral centroid coordinates as JSON.
• Diagnostic Thresholds: Applies integral vBMD cutoffs for osteoporosis (<160 mg/cm³) and low bone mass (160 ≤ BMD < 190 mg/cm³) with sensitivity and specificity comparable to quantitative CT thresholds.

Scientific Applications:

• Osteoporosis Screening: Enables opportunistic screening for osteoporosis from routine clinical CT scans using automated vertebral measures.
• Vertebral Fracture Prediction: Provides CT-based measures, particularly trabecular and integral vBMD, with demonstrated superior predictive performance for vertebral fractures compared to DXA-derived aBMD.
• Bone Health Research: Supports studies of the relationship between vertebral bone density measures and fracture risk by supplying automated quantitative metrics.

Methodology:

anduin employs a deep learning framework using convolutional neural networks (CNNs) for vertebra segmentation, reduces vertebral bodies to essential components, and extracts BMC, trabecular vBMD, integral vBMD, and CT-based aBMD through asynchronous calibration.