Automated Volumetric Assessment of Hounsfield Units Using a Deep-Reasoning and Learning Model: Correlations with DXA...

Area of Science:

• Radiology
• Artificial Intelligence
• Bone Health

Background:

• Accurate spinal bone density assessment is crucial for bone health evaluation and preoperative planning.
• Conventional Hounsfield unit (HU) measurements are limited by single-slice analysis, impacting precision in patients with vertebral deformities.

Purpose of the Study:

• To develop and validate a deep-reasoning artificial intelligence (DR-AI) model for automated volumetric spinal bone density assessment.
• To evaluate the correlation of DR-AI volumetric HU measurements with dual-energy X-ray absorptiometry (DXA) derived bone mineral density (BMD) and T/Z-scores.

Main Methods:

• A cross-sectional study analyzed lumbar CT scans from 84 patients with recent DXA assessments.
• A fully-automated DR model was used to compute volumetric HUs from the soft-tissue window of lumbar vertebrae.
• Spearman correlation coefficients assessed relationships between volumetric HUs and BMD, T-, and Z-scores.

Main Results:

• Significant positive correlations (p < 0.0001) were observed between volumetric HUs and BMD, T-, and Z-scores for L1-L4 vertebrae.
• The strongest correlation between HU and BMD was found at L2 (ρ = 0.75).

Conclusions:

• The DR model provides a reliable, efficient, and precise automated method for volumetric spinal bone density assessment.
• This AI-driven approach offers a promising alternative to conventional DXA measurements for evaluating bone health.