Magnetic Resonance Imaging (MRI) is increasingly applied at high magnetic field strengths of 3T and above. This trend aids in achieving superior image quality, accelerated scan times, and enhanced diagnostic capabilities. However, the increased magnetic field strengths also amplify the acoustic and electromagnetic field interactions during MRI. This presents considerable challenges for comfortable and reliable imaging.
Firstly, the acoustic noise at high magnetic field strength MRI is greatly exacerbated. The sound may reach up to 130 dB, greatly exceeding the safe prolonged noise exposure limit at 90 dB. This causes anxiety in patients, elevate claustrophobic reactions, and are even associated with increased scan refusal rates. The sound pressure levels at modern scanners generally exceed the public safety threshold. Thus, if not adequately accounted for, it bears the risk of temporary or permanent hearing damage.
Secondly, achieving a homogeneous distribution of the transmitted radiofrequency (RF) field, which is used to create the magnetization signal, becomes increasingly challenging at high magnetic fields. The degree of constructive and destructive interference of the RF field components becomes exacerbated as the RF field wavelength shortens and starts to approximate the body dimensions in tissue. This results in spatial signal variation that cannot be universally predetermined, which can hinder diagnostic assessment in clinical imaging.
In this thesis, multiple challenges of high-field MRI were tackled by developing advanced devices and MR scanning methodologies. Firstly, a solution to the acoustic noise problem was proposed based on predicting the scanner sound and superimposing its anti-noise to reduce the sound pressure. The results indicate robust noise reduction of up to 13 dB with a proof-or-principle system across various scanning methods.
To tackle the transmitted field B1+ inhomogeneity, an adaptive shimming device was proposed, based on an array of electrically-coupled pads with high permittivity. Based on electromagnetic field simulations, the device prototype was constructed. Up to 11% modulation was measured in phantoms at 15 mm depth, with an ability to tune the location of shimming hotspot.
A novel scanning methodology to quantify B1+ magnitude in the heart was developed, using Bloch-Siegert shift-based preparations. The sequence has shown highly-robust 2D and 3D mapping in healthy subjects, and achieved superior noise-resilience, when compared to more established mapping methods.
The proposed novel B1+ mapping sequence was then adapted to enable conductivity mapping in the myocardium. The method provided the first complex B1+ field-based conductivity reconstruction in healthy subjects, with robust results in agreement with previous literature.
Finally, another MRI scanning methodology was proposed for quantitative cardiac imaging robust to field inhomogeneities.T2* - a common biomarker used in MRI - was reconstructed using phase-cycled preparations. Compared to conventional sequence, higher noise resilience was achieved in healthy subjects, while enabling a wider range of sequence tailoring options.@en

Subjective assessment of craniosynostosis severity often leads to inconsistent treatment outcomes. This study introduces and evaluates innovative quantitative methods, including distance-based, spectral, shape descriptor, and deep learning view-based approaches, to objectively assess the severity of sagittal craniosynostosis. The effectiveness of these quantitative scores is determined by their correlation with expert clinical ratings, with the goal of improving the consistency and accuracy of severity evaluations. Furthermore, we applied the methods developed with synthetic data to real clinical datasets, focusing on pre- and post-surgery severity assessments. This study showcases how our quantitative approach effectively guides and optimizes surgical treatments, underscoring its potential utility in clinical practice.

Dementia, characterized by a significant decline in cognitive abilities, encompasses various neurodegenerative disorders. This includes Alzheimer's disease (AD), frontotemporal dementia (FTD) and their clinical manifestation named primary progressive aphasia (PPA), mainly involving language difficulties. In general these diseases exhibit some disease specific clinical characteristics. However, considerable heterogeneity exists in both atrophy patterns and symptom expression within each disease. Additionally, the overlap in disease patterns between AD and FTD complicates diagnosis. In this study, an unsupervised clustering algorithm is used to identify distinct atrophy patterns within AD and FTD, aiming to explore heterogeneity as well as disease overlap. Through the clustering of grey matter volumes, we identified two primary clusters potentially reflective of disease stage. These clusters were externally validated. Additionally, the two primary clusters were subdivided into four distinct clusters, each representing an unique atrophy pattern. These four clusters exhibited differences in cognitive performance in the five cognitive domains: visuospatial functioning, language, memory/learning, processing speed, and attention/executive functioning. The identified clusters provide inside in the atrophy patterns in AD and FTD, as well as their relation to clinical diagnosis and cognitive performance.

The periconceptional period, encompassing the embryonic phase, is a critical window where a majority of reproductive failures, pregnancy complications, and adverse pregnancy outcomes arise. The Carnegie staging system comprises 23 stages which are based on embryonic morphological development. This allows for the assessment of normal and abnormal embryonic development during this critical period. In-utero Carnegie staging using three-dimensional (3D) ultrasound scans visualized with virtual reality offers valuable insights but is currently a time-consuming manual process. To address this, we propose a deep learning approach for Carnegie staging in 3D ultrasound scans.

We used a dataset comprising 1413 3D ultrasound scans from the Rotterdam Periconceptional Cohort, annotated with Carnegie stages spanning from stages 13 to 23, including fetal subjects. Various training strategies were explored. We compared a metric regression approach, which considers the ordered nature of the Carnegie stages by treating the Carnegie stages as a continuous variable, with a multi-class classification approach, treating stages as independent categories. Additionally, we evaluated the influence of using a loss function accommodating the categorical nature of the Carnegie stages in the metric regression approach and examined the impact of incorporating embryonic size in the model input. Ultimately, a regression approach using the Mean Squared Error (MSE) loss function emerged as the optimal choice.

This model achieved a classification accuracy of 0.59 and a Root Mean Squared Error (RMSE) of 0.62 on the test set. This performance is comparable to an intermediate human rater, which achieved an accuracy of 0.63 and a RMSE of 0.65. Our findings represent a significant step towards the development of an automated Carnegie staging method, offering the potential for a more comprehensive evaluation of the critical embryonic phase in the clinic.

Ischemic stroke, a leading cause of death and disability worldwide, occurs when a blood vessel is occluded by a thrombus. Current therapies for ischemic stroke, include Intravenous thrombolysis (IVT) and Endovascular thrombectomy (EVT). EVT relies on a Thrombolysis in Cerebral Infarction (TICI) score for assessing treatment effectiveness. This score, based on the visual evaluation of medical images by physicians, suffers from inter- and intra-observer variability, as it is influenced by the individual rater’s judgment. Digital Subtraction Angiography (DSA) imaging is commonly utilized both before therapy to identify the occluded vessel and after therapy to evaluate the treatment outcome. Accurate vessel correspondence before and after treatment is crucial for a reliable assessment. To enhance current evaluation methods and address this challenging task, we propose two automated approaches for determining vessel correspondence in pre- and post-EVT DSA imaging. The proposed methods
utilize graphical representations of the cerebral vascular network and distinct matching procedures. We refer to the methods as registration-based vessel matching (RB-VM) and graph based vessel matching (GB-VM). The methods were evaluated using manually annotated data with the RB-VM and GB-VM methods achieving a recall of 82.7% (78.2; 85.7) and 51.3% (47.4; 54.4) respectively. This work marks a significant step towards automatic stroke therapy assessment and showcases the potential benefits of graph based algorithms for this task, paving the way for more reliable and objective treatment assessments.

Introduction: Deep brain stimulation (DBS) is an important therapeutic option for various neurological diseases. For certain indications the optimal target cannot be identified on structural magnetic resonance imaging (MRI), but can be visualized with tractography. A recent improvement for clinical practice is the probabilistic tractography algorithm, providing superior results to the deterministic counterpart. However, it is suggested that clinical users are unfamiliar with the complex technology of this new approach, which might lead to missed potential of tractography in DBS-care. The overall aims of this thesis were threefold: to create an overview of the national landscape on the application of tractography, to narrow the gap between medicine and advanced technology in the application of tractography in DBS surgery planning, and to report potentialities and pitfalls of the implementation of tractography in a general hospital.

Methods: In the first part, a survey was conducted among Dutch DBS clinicians on the deployment of tractography. The survey consisted of 25 questions about the provision of DBS care, the use of tractography, the expert’s opinion and the respondent’s demographic characteristics. A comprehensive literature study is conducted in the second part. In the third part, an economic evaluation is performed by analyzing the costs and benefits of tractography. Guidelines are suggested based on literature and expert experience.

Results: Tractography is considered valuable for essential tremor (p<0.001) and not valuable for epilepsy (p=0.002) and chronic cluster headache (p=0.016). The majority uses deterministic approaches like DTI. Probabilistic users consider tractography more valuable than deterministic users (p=0.036). There is a heterogeneity in used image acquisition parameters and a lack of knowledge exist on technical background of tractography. The key elements of the technical principles of tractography are summarized, including diffusion weighted MRI, the concept of estimating the diffusion tensors, image acquisition parameters and tract reconstruction with the deterministic and probabilistic approach and quantitative measures.

Conclusion: Tractography is used by DBS clinicians with limited knowledge on the technical aspects. Generally the inferior deterministic approach is used without an (inter)nationally standardized protocol for image acquisition and tract reconstruction. A comprehensive explanation of the technical concepts and suggested practical guidelines should contribute to better application of this promising technology in DBS-care.

Background: Histopathological growth patterns (HGP) are a biomarker for predicting survival and systemic treatment effectiveness in colorectal liver metastasis (CRLM). Currently, HGP assessment in CRLM requires the resection specimen. Predicting the HGP from preoperative medical imaging could allow more personalised care and better outcomes. Methods: 252 patients underwent CRLM resection between 2004 and 2018 without receiving any systemic treatment. Patients were characterised as having either pure desmoplastic growth (dHGP) or any other type of growth pattern combination (non-dHGP) (21% dHGP; 79% non-dHGP). These categories were chosen because pure desmoplastic growth is predictive of better overall survival. regions of interest were automatically extracted using a UNet based segmentation model. These ROIs were passed to a radiomics model and a deep learning model to classify between dHGP/non-dHGP and predict the fraction of dHGP. Results: The best-performing classification method was the radiomic approach achieving an AUC of 0.67 (95% CI: 0.58-0.78), whereas the best-performance deep learning model achieved an average AUC value of 0.59 (95% CI: 0.53-0.65). Additionally, regression predicting the fraction of dHGP failed, with the predicted values showing no significant correlation with the actual value. Conclusions: Radiomics can be used to assess HGP, however further improvements in predictive performance are needed before these methods can be applied.

Magnetic resonance fingerprinting is an MRI-based technique that allows for fast, simultaneous quantitative mapping of multiple tissue parameters. Multi-component MRF (MC-MRF) additionally allows for the mapping of multiple tissue components per voxel. Currently, most MC-MRF implementations ignore any nonlinear effects on signals resulting from multi-component systems, such as the effects introduced by magnetisation transfer (MT). Here, we investigate the effects of free pool to free pool magnetisation transfer on the accuracy of MC-MRF, with a focus on the application of this technique for myelin water fraction (MWF) imaging. Assessment of the different MC-MRF techniques is done through application of these algorithms to two-component numerical phantoms, where the signals from two interacting components were simulated using the EPG-X framework. Results show that MT has a negative effect on the accuracy of the acquired parameter estimates for all estimated parameters, resulting in biases and incorrect parameter estimates. Several adjusted methods for MC-MRF including magnetisation transfer were proposed and tested. Although theoretically improvements were expected, the used sequences showed to be inadequate for accurate MT estimates. More research into these techniques is still required to improve their performance and accuracy. A technique left mainly unexplored here is sequence optimisation to minimise the effects of magnetisation transfer on the resulting signals. A quick exploration, however, showed that this might be a viable approach for future research as well.

Introduction: Magnetic Resonance Imaging is a commonly used technique for the initial diagnosis of gliomas. T1, T2, T2-FLAIR, and post-contrast T1 with gadolinium-based contrast agents (GBCAs) can show tumor characteristics. However, using this contrast agent poses a risk to patients with kidney failures, has environmental impact, and increases cost. To address these issues, we aimed to evaluate the potential of deep learning in generating synthetic post-contrast T1 images without using contrast agents.

Method: The project investigates the potential of using deep learning (DL) to generate synthetic post-contrast T1 images based on T1, T2, and T2-FLAIR provided by the Erasmus Glioma Database. Exploring different model architectures, loss functions, and input sizes to discover the optimal approach.

Results: Results show that individual loss functions, input size, and model complexity slightly impact the accuracy of synthetic post-contrast T1 images. Combining loss functions, however, was the most promising approach for image generation. Models trained with ℒm could generate low detail enhancement. Resulting in 0.0478±0.0076, 0.0139±0.0036, and 0.879±0.024 for MAE, MSE, and SSIM, respectively.

Conclusion: The study's findings indicate that DL is promising for generating synthetic post-contrast T1 images without using GBCAs. However, further research is required to generate realistic synthetic post-contrast T1 images. The study, however, provides a basis for future work and highlights the importance of reducing the use of GBCAs in clinical practice.

Magnetic Resonance Imaging (MRI) is a flexible medical imaging technique that facilitates measurement of a wide range of contrasts particularly in soft tissue (e.g. brain and heart). Conventionally, qualitative images are acquired in which certain physical tissue properties are emphasized such as the transverse and longitudinal relaxation times. Such images are frequently referred to as "weighted", i.e. T₁-weighted. Quantitative MRI (qMRI) aims at measuring the underlying tissue parameters governing the contrast instead of yielding mere weighted images. These quantitative parameter estimations were proven to be more reproducible than conventional MR images and more sensitive to certain disease processes, enabling enhanced longitudinal comparisons within subjects as well as comparisons between subjects.

MR Fingerprinting (MRF) is an example of such a quantitative technique. MRF uses a combination of transient state acquisitions with varying flip angle patterns, severe undersampling and advanced signal models to allow for fast qMRI acquisitions and accurate estimation of a wide range of parameters.
While most qMRI methods assume a single tissue type per voxel, this is almost never a valid assumption. This assumption especially breaks down at tissue boundaries or when tissues consist of multiple, mixed compartments, such as water contained between myelin sheets in the brain, often called myelin water surrounded by extra-cellular water.

The goal of this thesis is to develop enhanced methodology for quantitative MRI by extending traditional signal and image post-processing methods. Specifically, the focus is on MR Fingerprinting in combination with multi-component estimations, in which different compartments are included in a mixed estimation model. This is done to obtain more information from the acquired data and to improve quantification, therefore possibly obtain new clinical insights. Important steps towards clinical use are to enhance estimation accuracy and precision compared to previous methods and reduce the scan time.


In this thesis the Sparsity Promoting Iterative Joint NNLS (SPIJN) algorithm is proposed for obtaining multi-component estimations from MRF data. This enabled sub-voxel, fractional estimation of signal components in a region of interest, without making a priori assumptions about tissues expected to be present. The main novelty of this method is to combine a non-negativity with a joint-sparsity constraint that limits the total number of tissues identified in a region of interest. As a result it became possible to obtain magnetization fraction maps of the white matter, gray matter, CSF and a component with shorter relaxation times related to myelin water.
The repeatability of the proposed method is studied in 5 subjects that were scanned 8 times with one week in between the scans each time. Comparison of the obtained white matter, gray matter and CSF maps with segmentations from conventional methods shows high repeatability of the estimated relaxation times and more fine structures in the CSF magnetization fraction maps.
Additionally, the proposed SPIJN algorithm was applied to data from a more conventional qMRI sequence, i.e. a multi-echo spin-echo sequence, to obtain estimations of the so-called myelin water fraction in the brain. The resulting images show significantly improved noise robustness compared to the standard multi-component analysis method, improving the usability.

MRF scans can be acquired in a relatively short acquisition time of less than 30 seconds per slice, but this will still result in 15 minutes of total scan-time when full brain coverage is needed. A further reduction in acquisition time is desirable for clinical usage, in which every minute counts. Therefore, improved reconstruction methods for MRF data are proposed, especially tailored to multi-component estimations. In in vivo scans we showed the improved image quality enabled by the proposed methods.

In another study, We applied the SPIJN algorithm to MRF brain scans from MS patients. In the results that we obtained we observe that white matter changes are reflected in a component with prolonged transverse relaxation times which is less pronounced in data of healthy controls. We hypothesize that the observed component reflects an increase in extra-cellular water and allows for early characterization of white matter damage.

In a related project, an adaptation on the SPIJN algorithm was introduced that is more sensitive to small local changes. The adjusted algorithm is applied to imaging data of MS patients and it is shown that it can help to identify small cerebral lesions.

MRF sequences can be chosen rather freely, to further reduce the scan time and reduce the estimation error these sequences can be optimized. A method is proposed in which parameter maps of the brain are used as reference upon which the MRF flip-angle series is optimized, taking into account the used undersampling trajectories. As a result undersampling errors, a major source of estimation errors, are effectively minimized.

Finally, we investigated an adjusted simulation method of MRF sequences that is able to accurately model the effects of through-plane motion, which is a major source of errors in MRF scans. Such a model may support the development of new retrospective correction methods for this type of motion as it enables proper simulation of its effects.

In summary, this thesis proposes new methods for multi-component reconstruction and analysis, sequence optimization and studying the effects of motion in MRF and further investigates the possibilities of multi-component MRF.@en

Many pulmonary diseases, such as cystic fibrosis (CF), chronic obstructive pulmonary disease (COPD) or asthma are disrupting the lung perfusion/ventilation ratio. Therefore, monitoring it gives a good overview on the evolution of the disease. There are several imaging modalities to assess the ventilation and perfusion, but they either use ionizing radiation, are expensive, uncomfortable or pose the risk of allergic reactions. Fourier decomposition MRI is a radiation and contrast free technique to image the lungs that is also used while the patient can breathe freely. It is based on the separation of the ventilation and perfusion signals, according to their frequency. Built on top of this is the Phase-Resolved Functional Lung MRI (PREFUL) which allows for the reconstruction of the breathing and cardiac cycle based on the phases of the acquired signal. The present study analysed different methods to extract the phase, to obtain the ventilation and perfusion maps and to threshold and detect the defects. It was found that the methods based on sine fitting and Morlet wavelet are the most promising, with the latter having the advantages of a variable window size and frequency. The most promising threshold for perfusion is the 75th percentile * 0.6 and for ventilation the 90th percentile * 0.4. Despite not correlating very well with the established methods, PREFUL analysis could still be useful in comparing different patients and tracking the progress of lung diseases.

We present application and developments of qMRI. We reconstructed data from the Prediva Ouderen Extentie (POE) database and calculated quantitative maps. The POE database consists of 29 7 Tesla (T) whole-brain MRI scans, including both male and female participants covering the age group between 80 and 91 years old. We used this data to extend the currently existing Amsterdam Ultra-high field adult lifespan database (AHEAD), covering the entire adult life span (18-80 years). Data was acquired at a submillimeter resolution using a the MP2RAGEME sequence, resulting in complete anatomical alignment of quantitative $R_1$-maps, $R_2^*$-maps, $T_1$-maps, $T_2^*$-maps, and quantitative susceptibility mapping (QSM).

As qMRI acquisitions and reconstructions are slow, we also investigated the potential of deep learning to speed up this process using the same dataset. Conventional MRI acquisitions are already time consuming, scanning the multiple images necessary for qMRI may take even longer. Computing a quantitative susceptibility map requires several processing steps involving phase unwrapping, background phase removal, and solving an ill-posed inverse problem, which all together is computationally expensive. COSMOS is currently seen as gold standard for solving the dipole inversion, although sampling from different orientations makes it a lengthy procedure and not feasible for clinical practice. Iterative methods were developed to make high quality QSM maps out of a single orientation measurement, however they suffer from artifacts and can be computationally expensive. Deep learning may improve the QSM pipeline in terms of time, convenience and quality. Neural networks were developed to learn processing steps like phase unwrapping and dipole inversion. Especially for the dipole inversion step several methods are proposed to calculate a high quality QSM map out of single orientation measurements which could speed up the QSM pipeline. We propose a method which can be useful in accelerated end-to-end QSM calculation and highlight advances and possible points for improvement. We will conclude with a discussion of the challenges that need to be overcome to establish a fast, easy and high quality QSM pipeline.

MRI has been used worldwide as an imaging modality for decades due to its ability to distinguish between soft tissues. However, MRIs cannot be used in all situations due to their extensive hard- and software. For this reason, research in the field of low-field MRI has been conducted. Low-field MRI systems do not use superconducting magnets but restive, or in this research, permanent magnets. This reduces the costs of these scanners, but these same changes result in a poorer signal-to-noise ratio (SNR) than high-field scanners. Based on this hardware, research in single-sided scanners has emerged, and various research groups have been able to reconstruct images with these scanners. They are based on conventional MRI hardware including a radiofrequency (RF) coil, permanent magnets and gradients coils for encoding. In collaboration with the Technical University (TU) Delft, a handheld scanner has been constructed in the Leiden University Medical Center (LUMC). This scanner consists of an RF coil and a permanent magnet and does not have a gradient coil. The inhomogeneity of the 50 can be used for spatially encoding the signal based on translations. The SNR in images is improved with model-based image reconstruction. Although, in theory, this set-up should work, the image results so far have been somewhat disappointing. Therefore, in this thesis, we have investigated a different approach, using a neural network. In this research, the correlation between the received signal of a measurement and a simulation of the same shape is found to be greater than the correlation between other simulations and the received signal. Due to this proven correlation, an image reconstruction deep learning algorithm is constructed based on the AUTOMAP model. Simulated signals are reconstructed into images with promising results; however, the algorithm cannot reconstruct mea- sured data. Therefore, suggestions for future improvement include improving the simulated data set and adding more real measurements that would allow for training on a measured data set.

Low-field MRI is susceptible to temperature changes in magnet temperature, resulting in time-dependent inhomogeneities. This causes time-dependent phase shifts in the measured signal, leading to image distortions. In this project, a method is proposed to interleave short magnetic field (B0) mapping sequences in the target image acquisition for Turbo Spin Echo (TSE) and Magnetic Resonance Fingerprinting (MRF). Simulations show that the proposed correction method works in 2D. Experiment results show that interleaved B0 mapping can be done at relatively low spatial and temporal resolution, shortening the B0 acquisition time. Also, the distortion correction works for both TSE and MRF, though its advantages over static inhomogeneity correction are limited to acquisitions with a small bandwidth per voxel and large or very inhomogeneous frequency drifts.

Purpose. Distal radius fractures are common fractures of the wrist. These fractures are often displaced and need reduction, after adequate reduction, the patients will have follow-up X-rays to check if the fracture stays stable. This is important because surgery might be required if the fracture becomes unstable. This can lead to delayed surgery which can worsen the treatment outcome.
Therefore, it would be valuable to predict which distal radius fractures are likely to become unstable. Machine learning could help predict the stability of distal radius fractures based on CT. In addition, because there is only a small dataset available, this research also studies the effect of different machine learning methods.
Method. Two different methods were evaluated for the stability prediction of distal radius fractures: traditional machine learning (radiomics method) and a residual network with and without transfer learning (deep learning method). For the radiomics method, a python package called WORC was used, which automatically extracts radiomic features and optimizes machine learning models. For the deep learning method, the backbone of a residual network called Med3D with added layers for classification was used.
Results. The radiomics method combined with augmentation, gave the best results (AUC: 0.64± 0.01). Also, it was found that using an augmented data set in the radiomics method resulted in improved performance. This gives a slight indication that the radiomic method can learn to predict DRF when there would be more data available.
Conclusion. The radiomic method is the most promising method for predicting the stability of distal radius fractures, despite the small difference in performance compared to random guessing and the deep learning method. However, for further research, it is highly recommended to acquire a larger dataset.

Glioma progression is monitored by routine MR scanning, enabling tumor growth evaluation with respect to earlier time-points. This growth may present both as a mass effect and as an extension of abnormalities into previously healthy tissue. To accurately quantify tumor growth and tumor-induced deformations, longitudinal intrasubject image registration is often used. However, such registration in cases with large deformations and tissue change is highly challenging. Longitudinal image registration may benefit from groupwise strategies in which multiple images are concurrently aligned. This avoids introducing bias towards an a priori-selected reference image. However, existing learning-based methods for image registration mostly concern pair-wise approaches. Moreover, the few proposed learning-based methods for groupwise registration are designed for the analysis of images without pathologies and are prone to fail to register glioma images.
To bridge this gap, we present a learning-based method for the non-linear registration of longitudinal glioma images. We adapt an existing learning-based groupwise method to handle tumor infiltration by means of cost-function masking. The proposed method is able to register glioma images despite the presence of non-correspondences across the time-points by focusing on the normal-appearing tissue similarity. We train the framework both in one resolution and with a multi-stage strategy exploring multiple resolutions.
We evaluate on a dataset from the Glioma Longitudinal AnalySiS consortium and compare it to conventional groupwise registration methods. We achieve comparable Dice coefficients, with higher SSIM and more detailed registrations. These evaluation metrics are further improved when trained as a multi-stage method. The proposed framework preserves the diffeomorphic conditions and the geometric centrality of the deformation fields, while significantly reducing the runtime to under a minute. The proposed methods may serve as an alternative to conventional toolboxes to provide further insight into glioma growth.

Predicting functional outcome after intra-arterial treatment (IAT) in acute ischemic stroke (AIS) patients is an important aspect of treatment decision making and prognostics. Standard methods for functional outcome prediction after stroke combine baseline clinical (and radiological) parameters. In this study, we investigated to what extent baseline CTA images can be used for the prediction of functional outcome and how this relates to standard scoring methods. Furthermore, it was investigated whether combining baseline CTA images with clinical parameters improved the predictive accuracy compared to outcome prediction based on clinical parameters. We proposed two network architectures, a convolutional neural network (CNN) for the processing of image data and a multilayer perceptron for the processing of clinical (and radiological) parameters. Various training strategies were applied for the fusion of image and clinical data. The CNN processing CTA images achieved an average cross-validated area under the curve (AUC) score of 0.67, which was lower than for models processing clinical (and radiological) parameters. The best performing model combining CTA images and clinical parameters was trained end-to-end and applied weight initialization of the pre-trained CNN (AUC = 0.78). The DeLong test showed that the combined model performed significantly better than the model processing clinical parameters (AUC = 0.75). However, the difference is small and might not be clinically relevant. Compared to scoring methods processing clinical and radiological parameters the combined model achieved similar performance.

Hypertrophic cardiomyopathy (HCM) is known as a frequent, genetic cardiovascular disease, often caused by mutations of sarcomere protein genes. HCM is primarily characterized by the presence of an increased left ventricular wall thickness, i.e. left ventricular hypertrophy (LVH). However, the disease appears to be asymptomatic in some patients, which makes it a diagnostic challenge. Mutation carriers of HCM who have not yet developed LVH are called genotype-positive left ventricular hypertrophy-negative (G+/LVH-) patients. The primary aim of this study was to investigate whether a radiomics model is able to distinguish between G+/LVH- patients and healthy controls, based on cardiac magnetic resonance (CMR) images. In total three datasets are analysed. A development dataset was used to develop different radiomics models and to evaluate the performance of the models. The models were validated on both the prospective validation dataset and external validation dataset. G+/LVH- patients had to be known to carry a class 4 (likely pathogenic) or class 5 (pathogenic) gene mutation for HCM and a maximum left ventricular wall thickness of <13mm. Endocardial and epicardial borders were manually and automatically segmented on long-axis view (2-chamber (2CH), 3-chamber (3CH), and 4-chamber (4CH)) and short-axis (SA) view in both end-diastolic (ED) and end-systolic (ES) phase. From these segmentation 555 features including shape, intensity and texture were extracted. Evaluation of radiomics models was performed through a 100x stratified random-split cross-validation in development dataset. Next, the models were validated on prospective validation dataset and external validation dataset. The radiomics model with best performance developed on development dataset had a mean area under the receiver operating characteristic curve (AUC) of 0.89. A similar performance in prospective validation was found (mean AUC of 0.89), while a lower performance was found in external validation dataset (mean AUC of 0.63). In addition, the radiomics models performed with automatic segmentation showed all a decrease in performance; mean AUC of 0.75, 0.77 and 0.50 in development dataset, prospective validation dataset and external validation dataset, respectively. Our radiomics models using CMR images can non-invasively distinguish between +/LVH- patients and healthy controls on both development dataset and prospective validation dataset. However, it was not able to distinguish on external validation dataset.

Magnetic Resonance Fingerprinting (MRF) is a relatively new approach for simultaneously estimating multiple quantitative maps in one acquisition. Sequence optimisation for MRF can be a powerful tool in increasing the accuracy an precision of the quantitative results. Multi-component analysis in the MRF framework can distinguish multiple different tissues in one voxel such as myelin water and white matter which play an important role in monitoring progressive diseases such as multiple sclerosis. Using the estimation theoretic Cramér-Rao bound, optimisations of the acquisition sequences can be performed, that increase the precision of the resulting tissue maps. The effect of this optimisation has been confirmed using numerical simulations. Speed-ups in MRF are generated using significant undersampling of the k-space information. This results in spatially coherent undersampling artefacts, that generally is the dominating error source for regular $T_1$ and $T_2$ mapping. The undersampling artefacts can be predicted using a mathematical model leveraging on techniques from perturbation theory. Numerical simulations suggested that optimisations of the acquisition parameters are effective in reducing the undersampling error. This was confirmed using in vivo scans. The optimisations resulting from these two different models are easily implemented in future clinical practice.