Detector nonlinearity in measuring the bone mineral density based on neural networks

Area of Science:

• Medical Imaging
• Biophysics
• Radiology

Background:

• Dual-layer flat panel detectors (DFD) enable dual-energy imaging with a single X-ray exposure.
• Traditional bone mineral density (BMD) measurement methods often involve complex scanning procedures and nonlinearity correction schemes.

Purpose of the Study:

• To perform noise sensitivity analyses on a single-shot DFD approach for BMD measurement.
• To propose and evaluate a method for calculating BMD using features from a nonlinear response detector without nonlinearity correction.

Main Methods:

• Evaluated condition numbers and mean squared error (MSE) for BMD estimation functions.
• Employed a fully connected neural network (FCNN) model for BMD surface description.
• Conducted experiments to model DFD nonlinear response and performed simulations with nonlinear intensities.

Main Results:

• BMD measurements using nonlinear intensities without correction outperformed methods with correction.
• The FCNN model-based approach significantly surpassed conventional multiple regression models.
• The proposed method demonstrates robustness against additive and multiplicative errors.

Conclusions:

• A simple, robust bone densitometry system can be implemented by bypassing complex detector nonlinearity issues.
• This approach offers a highly accurate and cost-effective alternative to traditional dual-energy X-ray absorptiometry (DXA) systems.
• The single-shot DFD method provides a promising advancement in osteoporosis diagnosis and management.