Data-Driven Iterative Image Reconstruction For Motion Correction In Head Computed Tomography

Computed tomography (CT) has become the diagnostic modality of choice for head trauma due to its accuracy, reliability, safety, and its availability. CT scanning of the head is typically used to detect infarction, calcifications, tumors, bone trauma, and hemorrhage. Motion correction (MC) is of...

Full description

Saved in:
Bibliographic Details
Main Author: Ashfaq, Afshan
Contributors: Luster, Markus (Prof. Dr.) (Thesis advisor)
Format: Doctoral Thesis
Language:English
Published: Philipps-Universität Marburg 2022
Subjects:
Online Access:PDF Full Text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Computed tomography (CT) has become the diagnostic modality of choice for head trauma due to its accuracy, reliability, safety, and its availability. CT scanning of the head is typically used to detect infarction, calcifications, tumors, bone trauma, and hemorrhage. Motion correction (MC) is of general interest in CT imaging. Motion correction techniques have two categories. The first group requires motion information and consists of motion acquisition and motion compensation processes. The motion acquisition derives the motion information from reference images, surrogate signals, or the data themselves. Motion compensation compensates the motion during a reconstruction process. The second group is based on image-processing techniques and corrects the motion without prior knowledge of the motion. Based on what prior information is available, one can choose the appropriate approach to perform the motion correction. The study aimed to validate the data-driven 3D iterative motion estimation (ME) and motion compensation algorithm on phantom as well as clinical studies with head movements during computed tomography (CT) scan and to optimize the data-driven 3D iterative algorithm for robust application. The Hoffman 3D brain phantom provides a quantitative and qualitative study of the three-dimensional effects of scatter and attenuation similar to the human brain. We also performed MC for head movement during the CT part of the scan of the brain PET/CT and examined its significance for final image reconstruction. A series of PET/CT scans of Hoffman's brain phantom filled with fluorodeoxyglucose (18F-FDG) were acquired using mCT Siemens Biograph PET/CT scanner. The phantom was acquired with a variety of movements during the CT part of the acquisition, to simulate patient movements, but the phantom remained stationary during the PET scan. Each motion-corrupted CT scan was reconstructed using fully automated 3D iterative data-driven image reconstruction with motion compensation (MC) to remove motion artifacts and afterward an attenuation map was generated from this MC CT. The PET raw data was reconstructed offline using the JSrecon algorithm with an attenuation map from motion-corrected CT and compared with the PET scan reconstructed with an attenuation map from motion-corrupted CT. The data-driven motion compensation approach was also implemented on patients presenting head movement during the CT part of the brain PET/CT scan. All reconstructed images were independently assessed for qualitative analysis and the scenium analysis was performed for quantitative analysis. The reconstructed PET images of 10 primary brain regions using both nMC-PET and MC-PET were analyzed and the results showed that the SUVmean of all brain regions in nMC-PET was significantly higher than those in MC-PET. The 3D-standard surface projection (3D-SSP) Z score was evaluated on both nMC-PET and MC-PET. The 3D-SSP method compares the data from the individual to a database of healthy controls by defining a large number of points on a spatially normalized brain surface. With the little head motion, motion correction had only a slight impact on the Z score image in qualitative terms. The data-driven iterative motion compensation approach for head CT significantly increases the quantitative and qualitative accuracy of the PET/CT brain image affected by patient movement. The method could be applied to both stand-alone helical CT scans and the CT component of hybrid imaging systems such as PET/CT and SPECT/CT.
DOI:10.17192/z2023.0026