Estimation of X-ray Energy Spectrum of Cone-Beam Computed Tomography Scanner Using Percentage Depth Dose Measurements and Machine Learning Approach

This study presents, for the first time, a method to indirectly estimate the cone-beam computed tomography (CBCT) x-ray spectrum in the diagnostic energy range from the percentage depth dose (PDD) using machine learning (ML) algorithms. Assuming that the measured PDD is a weighted mean of monochromatic PDDs (mPDDs) resulting from monochromatic x-ray energies, mPDDs from the diagnostic energy range of 10 to 140 keV are simulated at 1 keV intervals by Monte Carlo (MC) calculation. Then, x-ray spectrum prediction models are constructed using two different ML approaches, namely the artificial neural network (ANN) based on a generative model and a maximum a posterior (MAP) model. Both models account for more than 80% of the x-ray photons obtained by full MC simulations in commercial CBCT systems. The present method is expected to be applied into a beam hardening reduction in CBCT reconstruction, CBCT dose calculation, and a material decomposition which require exact information on the x-ray energy spectrum