21 results
Now showing1 - 10 of 21
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , RegularizedLeastSquares.jl: modality agnostic Julia package for solving regularized least squares problems(Infinite Science Publishing, 2024-03-10); ; ; ; Image reconstruction in Magnetic Particle Imaging (MPI) is an ill-posed linear inverse problem. A standard method for solving such a problem is the regularized least squares approach, which uses, a regularization function to reduce the impact of measurement noise in the reconstructed image by leveraging prior knowledge. Various optimization algorithms, including the Kazcmarz method or the Alternating Direction Method of Multipliers (ADMM), and regularization functions, such as l2 or Fused Lasso priors have been employed. Therefore, the creation and implementation of cutting-edge image reconstruction techniques necessitate a robust and adaptable optimization framework. In this work, we present the open-source Julia package RegularizedLeastSquares.jl, which provides a large selection of common optimization algorithms and allows flexible inclusion of regularization functions. These features enable the package to achieve state-of-the-art image reconstruction in MPI.Publicationtype: Journal Article12 39 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Exploiting the Fourier neural operator for parameter identification in MPIModel-based magnetic particle imaging (MPI) is a challenging task both due to the complicated underlying physical model and the high numerical effort required for the solution of the corresponding equations of motion. A third challenge for practical applications is the identification of model parameters that are consistent with the given experimental setting and produce accurate predictions of the MPI signals. In this work, we show how the parameter identification problem can be addressed using a learned physics simulator based on the Fourier neural operator. As an application, we show how model-based system matrices can be estimated from a small set of calibration measurements, which can also be interpreted as a model-based approach to system matrix recovery. We compared our approach to established compressed sensing and interpolation schemes and found that it outperformed both.Publicationtype: Journal Article8 67 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Extension of the Kaczmarz algorithm with a deep plug-and-play regularizer(Infinite Science Publishing, 2024); ; ; ; The Kaczmarz algorithm is widely used for image reconstruction in magnetic particle imaging (MPI) because it converges rapidly and provides good image quality even after a few iterations. It is often combined with Tikhonov regularization to cope with noisy measurements and the ill-posed nature of the imaging problem. In this abstract, we propose to combine the Kaczmarz method with a plug-and-play (PnP) denoiser for regularization, which can provide more specific prior knowledge than handcrafted priors. Using measurement data of a spiral phantom, we show that Kaczmarz-PnP yields excellent image quality, while speeding up the already fast convergence. Since the PnP denoiser is not coupled to the imaging operator, the Kaczmarz-PnP method is very generic and can be used for image reconstruction independently of the measurement sequence and MPI tracer type.Publicationtype: Journal Article47 116 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , A deep learning approach for automatic image reconstruction in MPIImage reconstruction in magnetic particle imaging is a challenging task because the optimal image quality can only be obtained by tuning the reconstruction parameters for each measurement individually. In particular, it requires a proper selection of the Tikhonov regularization parameter. In this work we propose a deep-learning-based post-processing technique, which removes the need for manual parameter optimization. The proposed neural network takes several images reconstructed with different parameters as input and combines them into a single high-quality image.Publicationtype: Journal ArticleScopus© Citations 3 75 195 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Exploiting the Fourier Neural Operator for faster magnetization model evaluations based on the Fokker-Planck equation(Infinite Science Publishing, 2023-03-19); ; ; ; Accurate modeling of the mean magnetic moment of an ensemble of magnetic particles in dynamic magnetic fields is a challenging task that requires sophisticated differential equation solvers. However, these methods are computationally costly and therefore not practical for long excitation sequences such as those of the Lissajous type. In this paper we propose to accelerate simulations by using a neural network mapping from the input parameter functions that are applied to the original particle simulator directly to the mean magnetic moment output function. The architecture of the neural network is based on the Fourier neural operator, which allows to train mappings between function spaces. Our results show that the particle simulation can be accelerated by a factor of about 200 while the relative error of the neural network simulator remains below 1.5 %.Publicationtype: Journal ArticleScopus© Citations 4 54 89 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , NFFT.jl: Generic and fast Julia implementation of the nonequidistant fast fourier transform*(Society for Industrial & Applied Mathematics (SIAM), 2023); ; The nonequidistant fast Fourier transform (NFFT) is an extension of the famous fast Fourier transform (FFT) that can be applied to nonequidistantly sampled data in time/space or frequency domain. It is an approximative algorithm that allows one to control the approximation error in such a way that machine precision is reached while keeping the algorithmic complexity in the same order as a regular FFT. The NFFT plays a major role in many signal processing applications and has been intensively studied from a theoretical and computational perspective. The fastest CPU implementations of the NFFT are implemented in the low-level programming languages C and C++ and require a compromise between code generalizability, code readability, and code efficiency. The programming language Julia promises new opportunities in optimizing these three conflicting goals. In this work we show that Julia indeed allows one to develop an NFFT implementation which is completely generic and dimension-agnostic and requires about two to three times less code than the other famous libraries NFFT3 and FINUFFT while still being one of the fastest NFFT implementations developed to date.Publicationtype: Journal Article7 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Compressed sensing and experiment design in magnetic field based imaging(2023); This work investigates the use of compressed sensing to reduce measurement times in magnetic resonance imaging (MRI) and magnetic particle imaging (MPI). To this end, a flexible yet performant image reconstruction framework is developed. Based on this, an efficient reconstruction scheme for non Cartesian MRI is proposed and a low-rank-based method for the estimation of MPI system matrices is developed. Finally, an experiment design framework is developed to obtain optimized sampling patterns for both MRI and MPI.Publicationtype: Doctoral Thesis41 122 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Warmstart Approach for Accelerating Deep Image Prior Reconstruction in Dynamic Tomography(2022-07); Deep image prior (DIP) has been successfully used in the field of tomography to obtain high-quality images from under-sampled and noisy measurements. The key advantage of DIP compared to conventional deep-learning based image reconstruction techniques is that it requires no training data and thus can be used in a flexible manner without incorporating domain specific knowledge. The downside of DIP is that it shifts the training step to reconstruction time where usually fast algorithms are required to reduced the latency between acquisition and display of the reconstructed image. In this work we tackle this problem for dynamic tomography scenarios in which a large number of temporally resolved images are taken over time. By initializing the DIP network using a previous frame of the time series, it is possible to significantly reduce the overall reconstruction time. To cope with abrupt changes in the captured time-series, we propose to use an adaptive restart method having the ability to switch between warm- and coldstart depending on the amount of inter-frame changes.Publicationtype: Conference Paper10 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Accelerated Kaczmarz for convergence speed-up in multi-contrast magnetic particle imaging(Infinite Science Publishing, 2022-03-21); ; ; Magnetic Particle Imaging (MPI) is a tracer based medical imaging modality with great potential due to its high sensitivity, high spatio-temporal resolution, and ability to quantify the tracer concentration. Image reconstruction in MPI is an ill-posed problem, which can be addressed by the use of regularization methods. The corresponding optimization problem is most commonly solved using the Kaczmarz algorithm. Reconstruction using the Kaczmarz method for single-contrast MPI is very efficient as it produces the desired images fast with a small number of iterations. For multi-contrast MPI, however, the regular Kaczmarz algorithm fails to obtain good quality images without channel leakage when using a small number of iterations. In this work, we propose to use an accelerated Kaczmarz method in order to reduce the reconstruction time needed to achieve a good separation of the channels and a good image quality in multi-contrast MPI.Publicationtype: Journal Article51 141 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Fat quantification in dual-layer detector spectral computed tomography : experimental development and first in-patient validation(Lippincott Williams & Wilkins, 2022); ; ; ; Objectives Fat quantification by dual-energy computed tomography (DECT) provides contrast-independent objective results, for example, on hepatic steatosis or muscle quality as parameters of prognostic relevance. To date, fat quantification has only been developed and used for source-based DECT techniques as fast kVp-switching CT or dual-source CT, which require a prospective selection of the dual-energy imaging mode. It was the purpose of this study to develop a material decomposition algorithm for fat quantification in phantoms and validate it in vivo for patient liver and skeletal muscle using a dual-layer detector-based spectral CT (dlsCT), which automatically generates spectral information with every scan. Materials and Methods For this feasibility study, phantoms were created with 0%, 5%, 10%, 25%, and 40% fat and 0, 4.9, and 7.0 mg/mL iodine, respectively. Phantom scans were performed with the IQon spectral CT (Philips, the Netherlands) at 120 kV and 140 kV and 3 T magnetic resonance (MR) (Philips, the Netherlands) chemical-shift relaxometry (MRR) and MR spectroscopy (MRS). Based on maps of the photoelectric effect and Compton scattering, 3-material decomposition was done for fat, iodine, and phantom material in the image space. After written consent, 10 patients (mean age, 55 ± 18 years; 6 men) in need of a CT staging were prospectively included. All patients received contrast-enhanced abdominal dlsCT scans at 120 kV and MR imaging scans for MRR. As reference tissue for the liver and the skeletal muscle, retrospectively available non-contrast-enhanced spectral CT data sets were used. Agreement between dlsCT and MR was evaluated for the phantoms, 3 hepatic and 2 muscular regions of interest per patient by intraclass correlation coefficients (ICCs) and Bland-Altman analyses. Results The ICC was excellent in the phantoms for both 120 kV and 140 kV (dlsCT vs MRR 0.98 [95% confidence interval (CI), 0.94-0.99]; dlsCT vs MRS 0.96 [95% CI, 0.87-0.99]) and in the skeletal muscle (0.96 [95% CI, 0.89-0.98]). For log-transformed liver fat values, the ICC was moderate (0.75 [95% CI, 0.48-0.88]). Bland-Altman analysis yielded a mean difference of -0.7% (95% CI, -4.5 to 3.1) for the liver and of 0.5% (95% CI, -4.3 to 5.3) for the skeletal muscle. Interobserver and intraobserver agreement were excellent (>0.9). Conclusions Fat quantification was developed for dlsCT and agreement with MR techniques demonstrated for patient liver and muscle. Hepatic steatosis and myosteatosis can be detected in dlsCT scans from clinical routine, which retrospectively provide spectral information independent of the imaging mode.Publicationtype: Journal ArticleScopus© Citations 12 55 85