Development of a digital energy modulation framework in projectional radiography

Area of Science:

• Medical Imaging
• Machine Learning
• Digital Health

Background:

• Projectional x-ray imaging is a cornerstone of medical diagnostics.
• Enhancing image quality and diagnostic information from x-ray imaging remains a critical area of research.
• Digital energy modulation offers a novel framework to translate between different x-ray energy domains.

Purpose of the Study:

• To evaluate the feasibility of integrating machine learning (ML) into a digital energy modulation framework for projectional x-ray imaging.
• To assess the capability of ML models to perform energy translation using digitally reconstructed radiographs (DRRs) derived from dual-energy CT datasets.

Main Methods:

• Generated 3500 DRRs per energy domain from dual-energy CT data at 15° increments (0°-90°).
• Employed a supervised deep-learning approach to train ML models for energy translation between polyenergetic domains and from polyenergetic to monoenergetic images.
• Assessed model performance using peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), mean squared error (MSE), and mean absolute percentage error (MAPE), with cross-validation and projection-specific dataset splits.

Main Results:

• Cross-validation models achieved PSNR: 29.1 ± 2.0, SSIM: 0.947 ± 0.017, MSE: 169.1 ± 68.3, MAPE: 8.2% ± 1.8%.
• Projection-specific models (anterior-posterior and lateral views) for polyenergetic domain translation yielded PSNR: 27.4 ± 0.5, SSIM: 0.909 ± 0.003, MSE: 195.9 ± 39.7, MAPE: 10.4% ± 2.1%.

Conclusions:

• The study confirms the feasibility of using ML for energy translation within a digital energy modulation framework for projectional x-ray imaging.
• The findings suggest that this ML-driven approach has the potential to enhance projectional x-ray imaging.
• Further research is warranted to refine ML models and explore clinical applications of this technology.