X-ray energy spectrum estimation based on a virtual computed tomography system.

Area of Science:

• Medical Physics
• Radiological Imaging
• Artificial Intelligence in Healthcare

Background:

• Accurate X-ray energy spectrum information is crucial for quantitative computed tomography (CT) applications.
• Traditional methods for spectrum determination can be complex and require specialized equipment.
• Developing methods to estimate spectra directly from CT data is essential for clinical workflow integration.

Purpose of the Study:

• To present a novel method for estimating the X-ray energy spectrum in diagnostic CT from reconstructed CT images.
• To develop and validate artificial neural network (ANN) models for spectrum estimation with and without bow-tie filters.
• To assess the accuracy and limitations of the proposed method under varying conditions.

Main Methods:

• A virtual CT system was used to generate datasets of Gammex phantom CT images and corresponding X-ray energy spectra.
• Artificial neural network (ANN) models were trained to predict the energy spectrum from CT values.
• The method was validated using simulated data and experimental data from a Canon Medical System Activion16 scanner.

Main Results:

• Both ANN models (with and without bow-tie filters) achieved an average agreement of over 80% in spectrum estimation.
• Estimation accuracy improved with increasing tube voltage.
• Accurate prediction required a signal-to-noise ratio greater than 10 in the CT image, with noise being a limiting factor.
• The ANN model with a bow-tie filter successfully estimated spectra from experimental data, allowing for filter shape optimization.

Conclusions:

• The proposed ANN-based method enables accurate X-ray energy spectrum estimation directly from CT images of a Gammex phantom.
• The technique is robust, with performance improving at higher tube voltages and requiring adequate image quality.
• This method offers a practical approach for clinical applications like beam hardening reduction, CT dose management, and material decomposition, requiring no special setup beyond a Gammex phantom image.