Refine
Has Fulltext
- yes (19)
Is part of the Bibliography
- yes (19)
Year of publication
Document Type
- Doctoral Thesis (19)
Language
- English (19) (remove)
Keywords
- Kernspintomografie (19) (remove)
Institute
- Physikalisches Institut (9)
- Graduate School of Life Sciences (8)
- Institut für diagnostische und interventionelle Radiologie (Institut für Röntgendiagnostik) (2)
- Deutsches Zentrum für Herzinsuffizienz (DZHI) (1)
- Graduate School of Science and Technology (1)
- Klinik und Poliklinik für Psychiatrie, Psychosomatik und Psychotherapie (1)
- Medizinische Klinik (bis 2004) (1)
Sonstige beteiligte Institutionen
Acceleration is a central aim of clinical and technical research in magnetic resonance imaging (MRI) today, with the potential to increase robustness, accessibility and patient comfort, reduce cost, and enable entirely new kinds of examinations. A key component in this endeavor is image reconstruction, as most modern approaches build on advanced signal and image processing. Here, deep learning (DL)-based methods have recently shown considerable potential, with numerous publications demonstrating benefits for MRI reconstruction. However, these methods often come at the cost of an increased risk for subtle yet critical errors. Therefore, the aim of this thesis is to advance DL-based MRI reconstruction, while ensuring high quality and fidelity with measured data. A network architecture specifically suited for this purpose is the variational network (VN). To investigate the benefits these can bring to non-Cartesian cardiac imaging, the first part presents an application of VNs, which were specifically adapted to the reconstruction of accelerated spiral acquisitions. The proposed method is compared to a segmented exam, a U-Net and a compressed sensing (CS) model using qualitative and quantitative measures. While the U-Net performed poorly, the VN as well as the CS reconstruction showed good output quality. In functional cardiac imaging, the proposed real-time method with VN reconstruction substantially accelerates examinations over the gold-standard, from over 10 to just 1 minute. Clinical parameters agreed on average.
Generally in MRI reconstruction, the assessment of image quality is complex, in particular for modern non-linear methods. Therefore, advanced techniques for precise evaluation of quality were subsequently demonstrated.
With two distinct methods, resolution and amplification or suppression of noise are quantified locally in each pixel of a reconstruction. Using these, local maps of resolution and noise in parallel imaging (GRAPPA), CS, U-Net and VN reconstructions were determined for MR images of the brain. In the tested images, GRAPPA delivers uniform and ideal resolution, but amplifies noise noticeably. The other methods adapt their behavior to image structure, where different levels of local blurring were observed at edges compared to homogeneous areas, and noise was suppressed except at edges. Overall, VNs were found to combine a number of advantageous properties, including a good trade-off between resolution and noise, fast reconstruction times, and high overall image quality and fidelity of the produced output. Therefore, this network architecture seems highly promising for MRI reconstruction.
In Magnetic Resonance Imaging (MRI), acquisition of dynamic data may be highly complex due to rapid changes occurred in the object to be imaged. For clinical diagnostic, dynamic MR images require both high spatial and temporal resolution. The speed in the acquisition is a crucial factor to capture optimally dynamics of the objects to obtain accurate diagnosis. In the 90’s, partially parallel MRI (pMRI) has been introduced to shorten scan times reducing the amount of acquired data. These approaches use multi-receiver coil arrays to acquire independently and simultaneously the data.
Reduction in the amount of acquired data results in images with aliasing artifacts. Dedicated methods as such Sensitivity Encoding (SENSE) and Generalized Autocalibrating Partially Parallel Acquisition (GRAPPA) were the basis of a series of algorithms in pMRI.
Nevertheless, pMRI methods require extra spatial or temporal information in order to optimally reconstruct the data. This information is typically obtained by an extra scan or embedded in the accelerated acquisition applying a variable density acquisition scheme.
In this work, we were able to reduce or totally eliminate the acquisition of the training data for kt-SENSE and kt-PCA algorithms obtaining accurate reconstructions with high temporal fidelity.
For dynamic data acquired in an interleaved fashion, the temporal average of accelerated data can generate an artifact-free image used to estimate the coil sensitivity maps avoiding the need of extra acquisitions. However, this temporal average contains errors from aliased components, which may lead to signal nulls along the spectra of reconstructions when methods like kt-SENSE are applied. The use of a GRAPPA filter applied to the temporal average reduces these errors and subsequently may reduce the null components in the reconstructed data. In this thesis the effect of using temporal averages from radial data was investigated. Non-periodic artifacts performed by undersampling radial data allow a more accurate estimation of the true temporal average and thereby avoiding undesirable temporal filtering in the reconstructed images. kt-SENSE exploits not only spatial coil sensitivity variations but also makes use of spatio-temporal correlations in order to separate the aliased signals. Spatio-temporal correlations in kt-SENSE are learnt using a training data set, which consists of several central k-space lines acquired in a separate scan. The scan of these extra lines results in longer acquisition times even for low resolution images. It was demonstrate that limited spatial resolution of training data set may lead to temporal filtering effects (or temporal blurring) in the reconstructed data.
In this thesis, the auto-calibration for kt-SENSE was proposed and its feasibility was tested in order to completely eliminate the acquisition of training data. The application of a prior TSENSE reconstruction produces the training data set for the kt-SENSE algorithm. These training data have full spatial resolution. Furthermore, it was demonstrated that the proposed auto-calibrating method reduces significantly temporal filtering in the reconstructed images compared to conventional kt-SENSE reconstructions employing low resolution training images. However, the performance of auto-calibrating kt-SENSE is affected by the Signal-to-Noise Ratio (SNR) of the first pass reconstructions that propagates to the final reconstructions.
Another dedicated method used in dynamic MRI applications is kt-PCA, that was first proposed for the reconstruction of MR cardiac data. In this thesis, kt-PCA was employed for the generation of spatially resolved M0, T1 and T2 maps from a single accelerated IRTrueFISP or IR-Snapshot FLASH measurement. In contrast to cardiac dynamic data, MR relaxometry experiments exhibit signal at all temporal frequencies, which makes their reconstruction more challenging. However, since relaxometry measurements can be represented by only few parameters, the use of few principal components (PC) in the kt-PCA algorithm can significantly simplify the reconstruction. Furthermore, it was found that due to high redundancy in relaxometry data, PCA can efficiently extract the required information from just a single line of training data.
It has been demonstrated in this thesis that auto-calibrating kt-SENSE is able to obtain high temporal fidelity dynamic cardiac reconstructions from moderate accelerated data avoiding the extra acquisition of training data. Additionally, kt-PCA has been proved to be a suitable method for the reconstruction of highly accelerated MR relaxometry data.
Furthermore, a single central training line is necessary to obtain accurate reconstructions. Both reconstruction methods are promising for the optimization of training data acquisition and seem to be feasible for several clinical applications.
This work considered the frequency-modulated balanced steady-state free precession (fm-bSSFP) sequence as a tool to provide banding free bSSFP MR images. The sequence was implemented and successfully applied to suppress bandings in various in vitro and in vivo examples. In combination with a radial trajectory it is a promising alternative for standard bSSFP applications. First, two specialized applications were shown to establish the benefits of the acquisition strategy in itself. In real time cardiac imaging, it was shown that the continuous shift in frequency causes a movement of the bandings across the FOV. Thus, no anatomical region is constantly impaired, and a suitable timeframe can be found to examine all important structures. Furthermore, a combination of images with different artifact positions, similar to phase-cycled acquisitions is possible. In this way, fast, banding-free imaging of the moving heart was realized. Second, acquisitions with long TR were shown. While standard bSSFP suffers from increasing incidence of bandings with higher TR values, the frequency-modulated approach provided banding free images, regardless of the TR.
A huge disadvantage of fm-bSSFP, in combination with the radial trajectory, is the decrease in signal intensity. In this work a specialized reconstruction method, the multifrequency reconstruction for frequency-modulated bSSFP (Muffm), was established, which successfully compensated that phenomena. The application of Muffm to several anatomical sites, such as inner ear, legs and cardiac acquisitions, proofed the advantageous SNR of the reconstruction.
Furthermore, fm-bSSFP was applied to the clinically highly relevant task of water-fat separation. Former approaches of a phase-sensitive separation procedure in combination with standard bSSFP showed promising results but failed in cases of high inhomogeneity or high field strengths where banding artifacts become a major issue. The novel approach of using the fm-bSSFP acquisition strategy with the separation approach provided robust, reliable images of high quality. Again, losses in signal intensity could be regained by Muffm, as both approaches are completely compatible.
Opposed to conventional banding suppression techniques, like frequency-scouts or phase-cycling, all reconstruction methods established in this work rely on a single radial acquisition, with scan times similar to standard bSSFP scans. No prolonged measurement times occur and patient time in the scanner is kept as short as possible, improving patient comfort, susceptibility to motion or physiological noise and cost of one scan.
All in all, the frequency-modulated acquisition in combination with specializes reconstruction methods, leads to a completely new quality of images with short acquisition times.
Nuclear spins in motion is an intrinsic component of any dynamic process when studied using magnetic resonance imaging (MRI). Moving spins define many functional characteristics of the human body such as diffusion, perfusion and blood flow. Quantitative MRI of moving spins
can provide valuable information about the human physiology or of a technical system. In particular, phase-contrast MRI, which is based on two images with and without a flow-encoding gradient, has emerged as an important diagnostic tool in medicine to quantify human blood flow. Unfortunately, however, its clinical usage is hampered by long acquisition times which only provide mean data averaged across multiple cardiac cycles and therefore preclude Monitoring the immediate physiological responses to stress or exercise. These limitations are expected to be overcome by real-time imaging which constitutes a primary aim of this thesis.
Short image acquisition times, as the core for real-time phase-contrast MRI, can be mainly realized through undersampling of the acquired data. Therefore the development focused on related technical aspects such as pulse sequence design, k-space encoding schemes and image
reconstruction. A radial encoding scheme was experimentally found to be robust to motion and
less sensitive to undersampling than Cartesian encoding. Radial encoding was combined with a FLASH acquisition technique for building an efficient real-time phase-contrast MRI sequence.
The sequence was further optimized through overlapping of gradients to achieve the shortest possible echo time. Regularized nonlinear inverse reconstruction (NLINV), a technique which jointly estimates the image content and its corresponding coil sensitivities, was used for image
reconstruction. NLINV was adapted specifically for phase-contrast MRI to produce both Magnitude images and phase-contrast maps. Real-time phase-contrast MRI therefore combined two highly undersampled (up to a factor of 30) radial gradient-echo acquisitions with and without a
flow-encoding gradient with modified NLINV reconstructions. The developed method achieved
real-time phase-contrast MRI at both high spatial (1.3 mm) and temporal resolution (40 ms).
Applications to healthy human subjects as well as preliminary studies of patients demonstrated
real-time phase-contrast MRI to offer improved patient compliance (e.g., free breathing) and immediate access to physiological variations of flow parameters (e.g., response to enhanced intrathoracic pressure). In most cases, quantitative blood flow was measured in the ascending aorta as an important blood vessel of the cardiovascular circulation system commonly studied
in the clinic. The performance of real-time phase-contrast MRI was validated in comparison to standard Cine phase-contrast MRI using studies of flow phantoms as well as under in vivo conditions. The evaluations confirmed good agreement for comparable results.
As a further extension to real-time phase-contrast MRI, this thesis implemented and explored a dual-echo phase-contrast MRI method which employs two sequential gradient echoes with and without flow encoding. The introduction of a flow-encoding gradient in between the two echoes
aids in the further reduction of acquisition time. Although this technique was efficient under in vitro conditions, in vivo studies showed the influence of additional motion-induced Phase contributions. Due to these additional temporal phase information, the approach showed Little promise for quantitative flow MRI.
As a further method three-dimensional real-time phase-contrast MRI was developed in this thesis to visualize and quantify multi-directional flow at about twice the measuring time of the standard real-time MRI method, i.e. at about 100 ms temporal resolution. This was achieved
through velocity mapping along all three physical gradient directions. Although the method is still too slow to adequately cover cardiovascular blood flow, the preliminary results were found to be promising for future applications in tissues and organ systems outside the heart. Finally, future developments are expected to benefit from the adaptation of model-based reconstruction
techniques to real-time phase-contrast MRI.
The focus of the work concerned the development of a series of MRI techniques that were specifically designed and optimized to obtain quantitative and spatially resolved information about characteristic parameters of the lung. Three image acquisition techniques were developed. Each of them allows to quantify a different parameter of relevant diagnostic interest for the lung, as further described below:
1) The blood volume fraction, which represents the amount of lung water in the intravascular compartment expressed as a fraction of the total lung water. This parameter is related to lung perfusion.
2) The magnetization relaxation time T\(_2\) und T*\(_2\)
, which represents the component of T\(_2\) associated with the diffusion of water molecules through the internal magnetic field gradients of the lung. Because the amplitude of these internal gradients is related to the alveolar size, T\(_2\) und T*\(_2\) can be used to obtain information about the microstructure of the lung.
3) The broadening of the NMR spectral line of the lung. This parameter depends on lung inflation and on the concentration of oxygen in the alveoli. For this reason, the spectral line broadening can be regarded as a fingerprint for lung inflation; furthermore, in combination with oxygen enhancement, it provides a measure for lung ventilation.
The noninvasive magnetic resonance imaging technique allows for the investigation of functional processes in the living plant. For this purpose during this work, different NMR imaging methods were further developed and applied.
For the localisation of the intrusion of water into the germinating rape seed with the simultaneous depiction of the lipid-rich tissue via a 3D rendering, in Chap. 5 the technique of interleaved chemical selective acquisition of water and lipid was used in the germinating seed. The utilization of high-resolution MR images of germinated seeds enabled the localization of a predetermined water gap in the lipid-rich aleurone layer, which resides directly under the seed coat. The for a long time in biology prevalent discussion, whether such a gap exists or the seed soaks up the water from all sides, rather like a sponge, could hereby, at least for the rapeseed seed, be answered clearly. Furthermore, the segmentation and 3D visualization of the vascular tissue in the rapeseed seeds was enabled by the high-resolution datasets, a multiply branched structure preconstructed in the seed could be shown. The water is directed by the vascular tissue and thus awakens the seed gradually to life. This re-awakening could as well be tracked by means of invasive imaging via an oxygen sensor. In the re-awakened seeds, the lipid degradation starts, other than expected, not in the lipid-rich cotyledons but in the residual endosperm remaining from seed development and in the aleurone layer which previously protected the embryo. Within this layer, the degradation could be verified in the high-resolution MR datasets.
The method presented in Chap. 6 provides a further characteristic trait for phenotyping of seeds and lipid containing plants in general. The visualization of the compounds of fatty acids in plant seeds and fruits could be achieved by the distinct utilization of chemical shift-selective imaging techniques. Via the application of a CSI sequence the fatty acid compounds in an olive were localized in a 2D slice. In conjunction with an individually adjusted CHESS presaturation module Haa85 the high-resolution 3D visualization of saturated and unsaturated fatty acid compounds in different seeds was achieved. The ratio maps calculated from these datasets allow to draw conclusions from the developmental stage or the type of seed. Furthermore, it could be shown that the storage condition of two soybean seeds with different storage time durations lead to no degradation of the fatty acid content.
Additional structural information from inside of dry seeds are now accessible via MRI. In this work the imaging of cereal seeds could be significantly improved by the application of the UTE sequence. The hitherto existing depictions of the lipid distribution, acquired with the spin echo sequence, were always sufficient for examinations of the lipid content, yet defects in the starchy endosperm or differences in the starch concentration within the seed remained constantly unseen with this technique. In a direct comparison of the datasets acquired with the previous imaging technique (spin echo) and with UTE imaging, the advantage of data acquisition with UTE could be shown. By investigating the potential seed compounds (starch, proteins, sugar) in pure form, the constituent parts contributing to the signal could be identified as bound water (residual moisture) and starch. The application of a bi-exponential fit on the datasets of the barley seed enabled the separate mapping of magnetization and of relaxation time of two components contributing to the NMR signal. The direct comparison with histological stainings verified the previous results, thus this technique can be used for the selective imaging of starch in dry seeds.
Conclusions on the translocation characteristics in plants can be drawn by the technique proposed in Chap. 8. The associated translocation velocities can now, even in the range of several um/h, be determined in the living plant. Based on calculated concentrations of an MR contrast agent, which was taken up by the plant, these translocation velocities were estimated both in longitudinal direction, thus along the vascular bundle, and in horizontal direction, thus out of the bundle. The latter velocity is located below the contrast agent's velocity value of free diffusion. By adjusting a dynamic contrast-enhancing imaging technique (DCE-Imaging, Tof91) the acquisition duration of a T1-map was significantly reduced. By means of these maps, local concentrations of the contrast agent in plant stems and the siliques of the rapeseed plant could be determined.
Numerous questions in plant science can only be answered by non-invasive techniques such as MRI. For this reason, besides the experimental results achieved in this work, further NMR methods were tested and provided for the investigation of plants.
As an example, the study on the imaging of magnetic exchange processes are mentioned, which provided the groundwork for a possible transfer of CEST experiments (Chemical Exchange Saturation Transfer) to the plant. The results are presented in the bachelor thesis of A. Jäger Jae17, which was performed under my supervision, they find great interest under biologists.
The development of new technologies, which extend the possibilities for the investigation of living organisms, is of great importance. For this reason, I have contributed to the development of the currently unpublished method RACETE (Refocused Acquisition of Chemical Exchange Transferred Excitations [Jak17, Reu17, Gut18a]). By rephasing the transferred magnetization the utilization of properties which have not been available in chemical "`exchange"' experiments is enabled. With this method a positive contrast is generated, thus a reference experiment is not mandatory. Furthermore, the image phase, which in classical experiments contains no information about the exchanged protons, can be used for the distinct identification of multiple substances which have been excited simultaneously.
This recently at the Department of Experimental Physics V developed method can be used in particular for the identification of lipids and for the localization of sugars and amino acids, thus it can serve the enhancement and improvement of non-invasive analytical methods.
Neuroimaging research has highlighted the relevance of well-balanced functional brain interactions as an essential basis for efficient emotion regulation. In contrast, abnormal coupling of fear-processing regions such as the amygdala, the anterior cingulate cortex (ACC) and the insula could be an important feature of anxiety disorders. Although activity alterations of these regions have been frequently reported in specific phobia, little is known about their functional interactions during phobogenic stimulus processing.
To explore these interrelationships in two subtypes of specific phobia – i.e., the blood-injection-injury subtype and the animal subtype – functional connectivity (FC) was analyzed in three fMRI studies. Two studies examined fear processing in a dental phobia group (DP), a snake phobia group (SP) and a healthy control group (HC) during visual phobogenic stimuli presentation while a third study investigated differences between auditory and visual stimuli presentation in DP and HC.
Due to a priori hypotheses of impaired interactions between the amygdala, the ACC and the insula, a first analysis was conducted to explore the FC within these three regions of interest. Based on emerging evidence of functionally diverse subregions, the ACC was further divided into a subgenual, pregenual and dorsal ACC and the insula was divided into a ventral-anterior, dorsal-anterior and posterior region. Additionally, an exploratory seed-to-voxel analysis using the amygdala, ACC and insula as seeds was conducted to scan for connectivity patterns across the whole brain.
The analyses revealed a negative connectivity of the ACC and the amygdala during phobogenic stimulus processing in controls. This connectivity was predominantly driven by the affective ACC subdivision. By contrast, SP was characterized by an increased mean FC between the examined regions. Interestingly, this phenomenon was specific for auditory, but not visual symptom provocation in DP. During visual stimulus presentation, however, DP exhibited further FC alterations of the ACC and the insula with pre- and orbitofrontal regions.
These findings mark the importance of balanced interactions between fear-processing regions in specific phobia, particularly of the inhibitory connectivity between the ACC and the amygdala. Theoretically, this is assumed to reflect top-down inhibition by the ACC during emotion regulation. The findings support the suggestion that SP particularly is characterized by excitatory, or missing inhibitory, (para-) limbic connectivity, reflecting an overshooting fear response based on evolutionary conserved autonomic bottom-up pathways. Some of these characteristics applied to DP as well but only under the auditory stimulation, pointing to stimulus dependency. DP was further marked by altered pre- and orbitofrontal coupling with the ACC and the insula which might represent disturbances of superordinate cognitive control on basal emotion processes. These observations strengthen the assumption that DP is predominantly based on evaluation-based fear responses.
In conclusion, the connectivity patterns found may depict an intermediate phenotype that possibly confers risks for inappropriate phobic fear responses. The findings presented could also be of clinical interest. Particularly the ACC – amygdala circuit may be used as a predictive biomarker for treatment response or as a promising target for neuroscience-focused augmentation strategies as neurofeedback or repetitive transcranial magnetic stimulation.
The subject of this work was to develop, implement, optimize and apply methods for quantitative MR imaging of tumors. In the context of functional and physiological characterization, this implied transferring techniques established in tumor model research to human subjects and assessing their feasibility for use in patients. In the context of the morphologic assessment and parameter imaging of tumors, novel concepts and techniques were developed, which facilitated the simultaneous quantification of multiple MR parameters, the generation of “synthetic” MR images with various contrasts, and the fast single-shot acquisition of purely T2-weighted images.
Morphological and Functional Ultrashort Echo Time (UTE) Magnetic Resonance Imaging of the Human Lung
(2019)
In this thesis, a 3D Ultrashort echo time (3D-UTE) sequence was introduced in the Self-gated Non-Contrast-Enhanced Functional Lung Imaging (SENCEFUL) framework. The sequence was developed and implemented on a 3 Tesla MR scanner. The 3D-UTE technique consisted of a nonselective RF pulse followed by a koosh ball quasi-random sampling order of the k-space. Measurements in free-breathing and without contrast agent were performed in healthy subjects and a patient with lung cancer.
A gating technique, using a combination of different coils with high signal correlation, was evaluated in-vivo and compared with a manual approach of coil selection. The gating signal offered an estimation of the breathing motion during measurement and was used as a reference to segment the acquired data into different breathing phases.
Gradient delays and trajectory errors were corrected during post-processing using the Gradient Impulse Response Function. Iterative SENSE was then applied to determine the fully sampled data.
In order to eliminate signal changes caused by motion, a 3D image registration was employed, and the results were compared to a 2D image registration method.
Ventilation was assessed in 3D and regionally quantified by monitoring the signal changes in the lung parenchyma. Finally, image quality and quantitative ventilation values were compared to the standard 2D-SENCEFUL technique.
3D-UTE, combined with an automatic gating technique and SENCEFUL MRI, offered ventilation maps with high spatial resolution and SNR. Compared to the 2D method, UTE-SENCEFUL greatly improved the clinical quality of the structural images and the visualization of the lung parenchyma.
Through‐plane motion, partial volume effects and ventilation artifacts were also reduced with a three-dimensional method for image registration.
UTE-SENCEFUL was also able to quantify regional ventilation and presented similar results to previous studies.
Myocardial infarction (MI) is a leading cause of death worldwide. Timely restoration of coronary blood flow to ischemic myocardium significantly reduces acute infarct mortality and attenuates ventricular remodeling. However, surviving MI patients frequently develop heart failure, which is associated with reduced quality of life, high mortality rate (10% annually), as well as high healthcare expenditures. The main processes involved in the evolution of heart failure post-MI are the great loss of contractile cardiomyocytes during ischemia-reperfusion and the subsequent complex structural and functional alterations, which are rooted in modifications at molecular and cellular levels in both the infarcted and non-infarcted myocardium. However, we still lack efficient treatments to prevent the development and progression of left ventricular remodeling. The improved survival rate of acute MI patients combined with the lack of effective therapy for post-MI remodeling contributes to the high prevalence of heart failure. Cardiac Magnetic Resonance Imaging (MRI) is an important tool for diagnosis and assessment of MI. With the advancement of this technology, the frontier of MRI has been extended to probing molecular and cellular events in vivo and non-invasively. In combination with assessment of morphology and function, the visualization of essential molecular and cellular markers in vivo could provide comprehensive, multifaceted views of the healing process in infarcted hearts, which might give new insight for the treatment of acute MI. In this thesis, molecular and cellular cardiac MRI methods were established to visualize and investigate inflammation and calcium flux in the healing process of acute MI in vivo, in a clinically relevant rat model.