A Phantom Study on Sn100kVp Ultra-Low Dose Chest CT: Machine Learning-Assisted Automated Phantom-Less QCT Bone Mineral Density Measurement
Abstract
Purpose
Due to the fact that the CT value varies non-linearly with the change of tube voltage, traditional 120kVp QCT models are incompatible to CT scans with different tube voltage. In this work, we propose and validate a PL-QCT BMD calibration model applicable to CT scans with tube voltage of 100kVp, a commonly used tube voltage on ultra-low dose chest CT scanning.
Methods
Using the European Spine Phantom (ESP) as the subject, this study performed 16 repeated scans using a Sn100 kVp chest CT protocol to acquire corresponding CT image data. The ESP was manufactured as 3 seperated vertebrae: L1 (Ref: 49.0 mg/cm3), L2 (Ref: 98.0 mg/cm3), L3 (Ref: 195.0 mg/cm3). 508 patients (269 females, 239 males) who underwent chest CT screening with the same protocol were retrospectively enrolled to obtain their muscle and fat CT data. A 100 kVp PL-QCT BMD calibration model was trained using ML techniques. With ESP's manufacture BMD reference values as the gold standard, the accuracy of the 100 kVp PL-QCT model and the traditional 120 kVp PL-QCT model in measuring BMD under Sn100 kVp CT scanning was compared and analyzed.
Results
Mean CTDIvol was 1.14 ± 0.27 mGy. The absolute values of BMD prediction error of the 100 kVp PL-QCT model compared to the 120 kVp model (Overall: 4.790 ± 5.956 vs. 16.692 ± 8.255 mg/cm3; L1: 2.839 ± 2.186 vs. 3.727 ± 4.131 mg/cm3, W=-2.7256, P<0.05; L2: 4.538 ± 5.750 vs. 16.027 ± 7.023 mg/cm3, W=-8.7677, P<0.0001; L3: 6.992 ± 9.933 vs. 30.321 ± 13.610 mg/cm3, W=-8.7677, P<0.05).
Conclusion
We developed a phantom-less QCT model accurately predicts BMD from Sn100kVp chest CT, demonstrating significantly superior accuracy compared to traditional 120kVp QCT models. It is advised to match the corresponding PL-QCT BMD analysis model to the actual scanning tube voltage of the CT equipment.