A learning-based material decomposition pipeline for multi-energy x-ray imaging.

Area of Science:

• Medical Imaging
• Machine Learning Applications
• Materials Science

Background:

• Multi-energy X-ray imaging enables material characterization through decomposition.
• Accurate material decomposition is hindered by non-ideal imaging system effects and model limitations.
• Explicit modeling of imaging systems for decomposition is challenging.

Purpose of the Study:

• To explore the feasibility of using machine learning (ML) for material decomposition tasks in X-ray imaging.
• To develop and validate a learning-based pipeline for material decomposition.
• To compare the performance of various ML algorithms against traditional methods and deep learning solutions.

Main Methods:

• A learning-based pipeline incorporating feature extraction (e.g., neighboring information) was proposed.
• Hold-out validation with continuous interleaved sampling was used for model evaluation and selection.
• Algorithms evaluated include Artificial Neural Network (ANN), Random Tree, REPTree, and Random Forest in simulation and experimentation.

Main Results:

• ML algorithms successfully trained material decomposition models in both simulated and experimental studies.
• Incorporating neighboring information significantly improved the performance of all tested ML algorithms.
• ANN and Random Forest demonstrated substantial performance improvements over state-of-the-art methods, particularly in noisy conditions.

Conclusions:

• The proposed pipeline effectively builds generic material decomposition models applicable to diverse scenarios.
• Machine learning methods, when combined with appropriate features and algorithms, significantly enhance material decomposition performance.
• The study confirms the feasibility and promise of ML for material decomposition, supporting future clinical translation.