AI-enhanced X-ray spectrum reconstruction for medical imaging system

Area of Science:

• Medical Imaging Physics
• Computational Radiology
• Diagnostic Imaging Technology

Background:

• Accurate X-ray energy spectrum knowledge is essential for image quality assessment and patient dose calculation in medical imaging modalities like CT and mammography.
• X-ray spectra are influenced by X-ray tube parameters such as electron beam energy (tube voltage) and anode angle, making them system-specific.

Purpose of the Study:

• To develop an efficient and rapid method for determining X-ray energy spectra across various medical imaging systems.
• To predict X-ray spectra using a limited set of simulations by varying tube voltage and anode angle.

Main Methods:

• Simulated X-ray spectra using the Monte Carlo N Particle (MCNP) method for seven anode angles (12°-24°) and multiple tube voltages (20-150 kV).
• Trained 150 Radial Basis Function Neural Networks (RBFNNs) with tube voltage and anode angle as input parameters to predict X-ray spectra.

Main Results:

• Successfully trained RBFNNs to accurately predict X-ray spectra for a range of tube voltages and anode angles.
• The developed method provides point-by-point X-ray spectrum predictions.

Conclusions:

• The RBFNN approach offers an efficient method for rapid X-ray energy spectrum determination in medical imaging.
• This simulation-based methodology is generalizable to real-world X-ray imaging systems, aiding in dose optimization and image quality enhancement.