Abstract:
Computed Tomography (CT) is a commonly used diagnostic tool that uses X-Rays in order to detect and diagnose injuries, tumors and other diseases by providing an in-depth view inside the patient’s body. A CT scanner takes cross sectional images of your body in layers or “slices”. This technique helps diagnosticians to be able to observe the patient’s body in great detail. However, one of the major problems in CT imaging is that of artifacts caused by high-density regions in the vicinity of normal tissues, for example calcified regions and metal implants. These high-density regions in the CT images could cause false voxel values. The study aimed to develop an innovative machine learning-based technique called residual dense U-Net (RDU-Net), specifically for spectral photon-counting CT (SPCCT), to mitigate metal artefacts across all energy bins. The proposed model was quantitatively evaluated and compared with and without the metal artefact reduction algorithm (MAR) using the line profiles, histogram analysis, signal-to-noise ratio (SNR), root mean squared error (RMSE), and structural similarity index measure (SSIM). The results show significant improvements with the average SNR across the five energy bins increasing from 3.37 to 17.40 after the application of the MAR algorithm. The average RMSE decreased from 0.016 to 0.001, and the average SSIM increased by 34.9%. The study also evaluated material density images of hydroxyapatite (HA) and iodine (I), with and without the MAR, using the receiver operating characteristic (ROC) paradigm. The results showed improved accuracy in the material identification for HA (86% to 91%) and I (84% to 93%) after MAR. Overall, the evaluation of the model showed promising results and the potential to significantly decrease the metal artefacts in all the parameters used in the energy analysis while preserving the attenuation profile of SPCCT images.
