A Novel Machine Learning-Based Semi-Automated Phantom-Less QCT Model for Osteoporosis Screening on 100 kVp...

Area of Science:

• Radiology
• Medical Imaging
• Osteoporosis Research

Background:

• Conventional quantitative CT (QCT) models for osteoporosis screening are calibrated at 120 kVp.
• Ultra-low-dose (ULD) chest CT scans at lower tube voltages (e.g., 100 kVp) may lead to inaccurate bone mineral density (BMD) measurements with these conventional models.
• Opportunistic screening for osteoporosis using chest CT is a growing area of interest.

Purpose of the Study:

• To develop and validate a machine learning-assisted phantom-less QCT (PL-QCT) model for accurate BMD quantification at 100 kVp.
• To assess the diagnostic performance of the developed 100 kVp PL-QCT model.
• To compare the accuracy of the 100 kVp PL-QCT model against a conventional 120 kVp QCT model.

Main Methods:

• Development of a 100 kVp PL-QCT model using machine learning, calibrated against European Spine Phantom (ESP) reference values.
• Internal validation using 508 patients and external validation on an independent CT system with ESP scans and 197 patients under a 100 kVp protocol.
• Diagnostic performance evaluation by comparing QCT results with dual-energy X-ray absorptiometry (DXA) in 178 individuals.

Main Results:

• The 100 kVp PL-QCT model demonstrated significantly lower BMD error compared to the 120 kVp model during internal validation (2.39±7.12 vs 16.68±8.26 mg/cm³, p<0.0001).
• External validation confirmed substantially lower error for the 100 kVp model (-0.11±4.39 vs 13.78±5.10 mg/cm³).
• The model achieved an 88.8% agreement with DXA for osteoporosis diagnosis.

Conclusions:

• A machine learning-assisted 100 kVp PL-QCT model enables accurate BMD quantification from ULD chest CT.
• This novel model outperforms conventional 120 kVp QCT models in accuracy.
• The findings support the use of this model for reliable, cross-scanner opportunistic osteoporosis screening.