Chest CT-based automated vertebral fracture assessment using artificial intelligence and morphologic features.

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Pulmonology

Background:

• Spinal degeneration and vertebral compression fractures are prevalent in the elderly, impacting mobility and quality of life.
• Chronic obstructive pulmonary disease (COPD) patients have a high risk of osteoporosis and associated vertebral fractures, necessitating careful assessment.

Purpose of the Study:

• To develop automated methods for segmenting and labeling vertebrae in chest CT images using deep learning (DL) and a freeze-and-grow (FG) algorithm.
• To detect vertebral deformities and fractures using computed vertebral height features and computational modeling, mirroring expert protocols.

Main Methods:

• A deep learning network computed vertebral body likelihood maps from chest CT.
• An iterative multi-parametric FG algorithm delineated and labeled individual vertebrae.
• Intensity autocorrelation separated fused vertebrae, followed by contour analysis for height computation and fracture assessment.

Main Results:

• The automated method achieved high accuracy in vertebral segmentation (Dice score .984) and labeling (100%).
• Fracture assessment demonstrated high overall accuracy (98.3%), sensitivity (94.8%), and specificity (98.5%) across 40,050 vertebrae.
• The method showed excellent generalizability to low-dose CT imaging.

Conclusions:

• The developed CT-based automated method for vertebral fracture assessment is accurate and efficient.
• It provides a feasible alternative to manual expert reading, particularly for large population studies.
• The method's generalizability to low-dose CT expands its applicability in studies aiming to reduce radiation exposure.