Dual-Energy CT Material Decomposition and Evaluation for Breast Imaging

Sammanfattning

Dual-energy computed tomography (CT) is a CT technique in which imaging is performed at two different energy levels. The use of two different energy levels helps streamline and improve the accuracy of reconstructing the irradiated object. Dual-energy CT enables the characterization of materials within an object based on their attenuation coefficients and physical properties. In CT, noise is a major factor in image quality. Different intensities and types of noise are underlying issues that affect the reconstructed image [1]. This thesis aimed to evaluate material decomposition methods for dual-energy breast CT and a denoising method, and to examine how the denoising method influences their quantitative performance. Conversion of breast data into a readable file for the simulation-based framework X-ray Computed Imaging Simulation Toolkit (XCIST), where CT simulation was performed. The simulated data were reconstructed into Feldkamp-Davis-Kress algorithm data (FDK), and material decomposition was performed based on the FDK:s and a separate breast dataset. The material decomposition method used in this project was both non- iterative and iterative. The methods were also evaluated with a denoising step that reduces the quantum noise introduced during the simulation. The results show that the iterative and non- iterative methods achieve comparable performance, as both yield narrowing Contrasts-to-Noise Ratios (CNR), Root Mean Square Errors (RMSE) and noise score. By adding the Denoising method, CNR and RMSE values increased. This indicates that suppressing quantum noise improves the material decomposition performance. Iterative methods are as accurate as simpler material- decomposition methods. The results guide future dual-energy breast CT workflows and can serve as a benchmark for machine learning methods.