title: Digital and Physical Phantoms for Image-guided Interventions
creator: Bauer, Dominik Fabian
subject: ddc-530
subject: 530 Physics
subject: ddc-600
subject: 600 Technology (Applied sciences)
description: With recent advances in Deep Learning, the need for labeled training data in medical imaging has increased. However, due to privacy laws and because annotation by medical experts is time-consuming, labeled training data are scarce.  In this work, digital and physical phantom data is introduced to overcome this data shortage. The phantoms are used to develop image processing algorithms and to validate interventional workflows with a focus on liver interventions. Digital phantom data is used to develop image registration algorithms and a deep learning computed tomography (CT) reconstruction for the mitigation of metal artifacts. Furthermore, physical phantoms are manufactured for the validation of robotic needle guidance systems and to optimize interventional imaging protocols.  In the first part of this thesis, a synthesis framework for multimodal abdominal image data is presented. The generated CT, cone beam CT (CBCT), and magnetic resonance imaging (MRI) dataset is inherently registered and was used to optimize registration algorithms. Compared to real patient data, the synthetic data showed good agreement regarding the image voxel intensity distribution and the noise characteristics.  In the second part, an end-to-end deep learning CT reconstruction technique called iCTU-Net is developed for metal artifact reduction. The network was trained with simulated metal artifact data obtained from a data generation system that was developed in this work. The iCTU-Net was the only investigated method that was able to eliminate metal artifacts. For projection data exhibiting severe artifacts, the iCTU-Net achieved reconstructions with SSIM = 0.970±0.009 and PSNR = 40.7±1.6. The best reference method, an image based post-processing network, only achieved SSIM = 0.944±0.024 and PSNR = 39.8±1.9.  The third and fourth part focus on the manufacturing of physical phantoms for the validation of interventional workflows. An abdominal phantom incorporating a liver and six liver lesions with varying visibility in CT and MRI was manufactured to validate a standardized oligometastatic disease (OMD) diagnosis workflow. The workflow includes multimodal image acquisition, image segmentation and registration, and robotically assisted liver biopsy. Using similar materials and a similar manufacturing process, a pelvis phantom with a prostate and four prostate lesions was manufactured. The pelvis phantom was used to perform an MRI-guided prostate biopsy using an MRI-compatible robotic system for needle guidance.  The presented work enables the creation of phantom data for the development and validation of a plethora of medical imaging applications. Algorithms for multimodal image registration and CT image reconstruction were successfully developed and physical abdomen and pelvis phantoms for image-guided interventions were manufactured. While the focus of this work was on liver interventions, the presented frameworks can readily be adapted to other body regions.
date: 2022
type: Dissertation
type: info:eu-repo/semantics/doctoralThesis
type: NonPeerReviewed
format: application/pdf
identifier: https://archiv.ub.uni-heidelberg.de/volltextserverhttps://archiv.ub.uni-heidelberg.de/volltextserver/31807/1/Bauer_Dissertation.pdf
identifier: DOI:10.11588/heidok.00031807
identifier: urn:nbn:de:bsz:16-heidok-318076
identifier:   Bauer, Dominik Fabian  (2022) Digital and Physical Phantoms for Image-guided Interventions.  [Dissertation]     
relation: https://archiv.ub.uni-heidelberg.de/volltextserver/31807/
rights: info:eu-repo/semantics/openAccess
rights: http://archiv.ub.uni-heidelberg.de/volltextserver/help/license_urhg.html
language: eng