Jyh-Miin Lin has a total of 42 co-authors. Here are the top co-authors by number of shared publications:

  • Hao Chiao Yang

    Madou Sin-Lau Hospital,Department of Diagnostic Imaging,Tainan,Taiwan

    6 shared publications
  • Hsin-Chia Chen

    Madou Sin-Lau Hospital,Department of Diagnostic Imaging,Tainan,Taiwan

    6 shared publications
  • Yu-Chieh Chao

    Madou Sin-Lau Hospital,Department of Diagnostic Imaging,Tainan,Taiwan

    6 shared publications
  • Chih-Ching Chen

    Chung Yuan Christian University,Department of Finance,Taoyuan City,Taiwan

    3 shared publications
  • Hsiao-Wen Chung

    Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei, Taiwan

    3 shared publications
  • Tzu-Hsueh Tsai

    Kaohsiung Medical University Hospital, Graduate Institute of Clinical Medicine, Kaohsiung Medical University,Department of Medical Imaging,Kaohsiung,Taiwan

    3 shared publications

Latest Publications

  • Generalised Inverses for Parallel MRI

    by Hsin-Chia Chen, Hao Chiao Yang, Yu-Chieh Chao, Jyh-Miin Lin

    Parallel Magnetic Resonance Imaging (MRI) has emerged as a significant technique for accelerated scans. Techniques such as Sensitivity Encoding (SENSE) and generalised Autocalibrating Partially Parallel Acquisitions (GRAPPA) have enhanced image acquisition efficiency. However, these numerical methods were developed without a harmonised background. To overcome the gap between fundamentals and practical computations, we overview the fundamentals of generalised inverse and its application in parallel imaging. Efficient numerical methods are demonstrated to enhance the convergence rate while diminishing computational complexity. Furthermore, we check the generalised inverse of matrix products within the framework of coil sensitivity profiles and correlations to Fourier domain reconstruction. Simulation results indicate that the generalised inverse approach surpasses conventional techniques by facilitating quicker and more stable reconstructions. The Newton-Schulz algorithm achieves convergence to machine precision within 20 iterations, demonstrating a superlinear convergence rate. In vivo data also validate the improvement of the theory of generalised inverse. Our findings imply that generalised inverse-based techniques could represent a viable solution for parallel MRI reconstruction, offering potential for clinical application without necessitating significant hardware enhancements or extensive datasets. In conclusion, this paper demonstrates the fundamental basis and practical numerical methods for generalised inverse methods for parallel MRI, which is not restricted by singular-value decomposition. We discuss the correlation of generalised inverses to properties of SENSE-like and GRAPPA-like methods.

  • Krylov Subspace Methods for MRI Coil Sensitivity Profiles

    by Hsin-Chia Chen, Tzu-Hsueh Tsai, Hao Chiao Yang, Yu-Chieh Chao, Jyh-Miin Lin

    Magnetic Resonance Imaging (MRI) is a cornerstone of modern medicine, offering invaluable insights into the human body. However, traditional MRI scans have often been hampered by lengthy acquisition times, limiting their use in certain clinical situations. Over the past two decades, parallel MRI has emerged as a game-changer, offering significant reductions in scan times. This technology hinges on the critical concept of coil sensitivity profiles, which act as a map for accurately interpreting the complex signals acquired during an MRI scan. The quest for optimal performance in parallel MRI requires precise and efficient coil sensitivity profiles. Inaccurate profiles can distort images and introduce artefacts, compromising the diagnostic value of the scan. Additionally, inefficiently generated profiles can significantly increase the computational burden, further extending scan times. Researchers have explored various techniques to extract reliable coil sensitivity profiles. One promising approach involves inexact inverse iterations, which iteratively refine an initial estimate of the profile until it accurately reflects the underlying data. While effective, this method suffers from a major drawback: the computational expense associated with the inner loop of the iterative process. This translates to longer scan times, potentially negating the benefits of parallel MRI in time-sensitive clinical settings. To overcome this limitation, our research team has investigated the implementation of Krylov subspace methods. These powerful mathematical techniques offer a novel approach to tackling the computational challenge. By exploiting the inherent structure of the problem, Krylov subspace methods effectively reduce the dimensionality of the problem, essentially transforming it into a simpler and less complex form. This significantly reduces the computational workload required to achieve accurate coil sensitivity profiles. To validate the efficacy of our proposed method, we conducted a comprehensive evaluation, meticulously comparing the image quality and phase response obtained using various techniques. The results show that the Krylov subspace method not only generated highly accurate and robust coil sensitivity profiles but also achieved this with significantly reduced computation times compared to the previous inexact inverse iterations. This translates to a remarkable improvement in scan speed, paving the way for wider adoption of parallel MRI in clinical practice. The potential impact of this research extends beyond simply reducing scan times. Faster MRI scans can lead to improved patient comfort, particularly for individuals who experience anxiety or claustrophobia during the procedure. Additionally, the ability to acquire high-quality images in a short time opens doors for new applications of MRI, such as dynamic imaging of rapidly changing processes within the body. Results show that the proposed methods improve our previous inexact inverse iterations, leading to smoother coil sensitivity profiles with fewer structural artefacts. The computation speed of the new methods is 5 × faster than the previous inexact inverse iterations. In conclusion, our research demonstrates the significant potential of Krylov subspace methods in improving the efficiency of clinical parallel MRI. This advancement paves the way for faster, more comfortable, and ultimately, more informative MRI scans, contributing to improved patient care and advancing the field of medical imaging.


  • A recursive deconvolutional kernel for ECG

    by Hsin-Chia Chen, Hao Chiao Yang, Yu-Chieh Chao, Jyh-Miin Lin

    Convolution is a fundamental technique in biomedical signal processing, but deconvolution requires a more dedicated processing technique than the convolution. Previous iterative Krylov subspace methods or division in the Fourier domain might be inappropriate for high-precision applications. To the end of stable and efficient deconvolution, pseudoinverse reconstruction could be an ideal deconvolution technique. However, pseudoinverse reconstruction through singular-value decomposition (SVD) requires excessive memory to save the singular vectors. In this study, we proposed a recursive deconvolution kernel using the pseudoinverse without SVD. We applied the deconvolution kernel to the PhysioNet electrocardiography (ECG) data with a Gaussian convolution kernel.


  • On the generalized inverse for MRI reconstruction

    by Tzu-Hsueh Tsai, Hsin-Chia Chen, Hao Chiao Yang, Yu-Chieh Chao, Jyh-Miin Lin, Chih-Ching Chen, Hing-Chiu Chang, Chin-Kuo Chang, Wei-Hsuan Yu, Feng-Nan Hwang, et al.

    Recent studies have suggested that the boundary between data-driven deep-learning non-Cartesian magnetic resonance imaging (MRI) reconstruction methods and conventional optimization-based, iterative reconstruction methods is becoming blurred. For instance, the unrolled iterative reconstruction method can be regarded as a trainable neural network. Another example is that the Moore-Penrose pseudoinverse plays a central role in finding the predefined solution to many imaging processes. However, the application of pseudoinverse in MRI reconstruction was obstructed in clinical imaging, mostly due to the excessive storage required for singular vectors. Since the spatial encoding of MRI is fully determined by the known k-space trajectory, the generalized inverse can be ”iteratively learning in a data-free fashion”, which leads to surprising but realizable properties. To compare our method with other conventional methods, numerical simulations were performed using in vivo MRI. The proposed method leads to nearly equivalent image quality with a much shorter run-time (only 0.68%) than the conjugate gradient (CG) method. We discuss the potential impact of the generalized inverse as a feasible reconstruction method for non-Cartesian MRI.


  • On the generalized eigenvalue problem for robust coil sensitivities of MRI

    by Hsin-Chia Chen, Tzu-Hsueh Tsai, Hao Chiao Yang, Yu-Chieh Chao, Chih-Ching Chen, Hing-Chiu Chang, Hsiao-Wen Chung, Chin-Kuo Chang, Wei-Hsuan Yu, Feng-Nan Hwang, et al.

    Although parallel imaging is a routine technique in clinical magnetic resonance imaging (MRI), estimating coil sensitivities remains a constant challenge to accelerated MRI. Recent coil estimation methods are computationally expensive, requiring external libraries to compute the singular-value decomposition (SVD) of a large matrix. Moreover, phase variations can appear in the final complex-valued coil sensitivity profiles. For instance, phase singularity is frequently observed in contrast-enhanced MRI. In this study, a robust generalized eigenvalue problem (GEP) approach is proposed to address the phase variations (also known as phase flipping or phase singularity) in complex-valued coil sensitivity of parallel MRI. Based on the inexact inverse iteration method in the outer loop and the conjugate gradient method in the inner loop, the low-cost inexact inverse iteration improves the homogeneity of the coil sensitivity without phase variations of the existing method. We performed repeated simulation studies. Results of 2D and 3D computational simulations are demonstrated, and the potential use of the method is also discussed.


  • A parallel hierarchical tensor product method for n-dimensional interpolation in the Fourier domain

    by Hsin-Chia Chen, Hao Chiao Yang, Yu-Chieh Chao, Julian Nicholls, Jyh-Miin Lin, Chih-Ching Chen, Wei-Hsuan Yu, Feng-Nan Hwang

    Parallel multi-dimensional interpolation in the complex Fourier domain (also known as the non-uniform fast Fourier transform) encounters major challenges, due to computation issues such as increasing computation complexity and space complexity. For instance, the contemporary graphics processing unit (GPU) is limited by the relatively small memory size and the increasing size of the interpolator. This issue makes multidimensional Fourier domain interpolation problematic, while finding an optimized configuration remains an unsolved challenge in industrial applications, e.g. magnetic resonance imaging (MRI) or computerized tomography. To enhance the performance of multi-dimensional interpolation on GPU, a new parallel hierarchical tensor products tree approach is proposed. The method combines the composite 1D interpolators under the limitation imposed by the memory size of the device. The resultant run-time performance on the GPU varies with different configurations. The best-tuned method is 2.52-4.98× faster than the compressed sparse row (CSR) on the discrete GPU and 4.16-9.59× faster than CSR on the integrated GPU. The hierarchical tensor product interpolation is used to compute the multi-dimensional nonuniform fast Fourier transform. An acceleration of 30× was achieved in 3D MRI reconstruction.

  • Open Access

    Improved Image Quality for Static BLADE Magnetic Resonance Imaging Using the Total-Variation Regularized Least Absolute Deviation Solver

    by Hsin-Chia Chen, Haw-Chiao Yang, Chih-Ching Chen, Seb Harrevelt, Yu-Chieh Chao, Jyh-Miin Lin, Wei-Hsuan Yu, Hing-Chiu Chang, Chin-Kuo Chang, Feng-Nan Hwang

    In order to improve the image quality of BLADE magnetic resonance imaging (MRI) using the index tensor solvers and to evaluate MRI image quality in a clinical setting, we implemented BLADE MRI reconstructions using two tensor solvers (the least-squares solver and the L1 total-variation regularized least absolute deviation (L1TV-LAD) solver) on a graphics processing unit (GPU). The BLADE raw data were prospectively acquired and presented in random order before being assessed by two independent radiologists. Evaluation scores were examined for consistency and then by repeated measures analysis of variance (ANOVA) to identify the superior algorithm. The simulation showed the structural similarity index (SSIM) of various tensor solvers ranged between 0.995 and 0.999. Inter-reader reliability was high (Intraclass correlation coefficient (ICC) = 0.845, 95% confidence interval: 0.817, 0.87). The image score of L1TV-LAD was significantly higher than that of vendor-provided image and the least-squares method. The image score of the least-squares method was significantly lower than that of the vendor-provided image. No significance was identified in L1TV-LAD with a regularization strength of λ= 0.4–1.0. The L1TV-LAD with a regularization strength of λ= 0.4–0.7 was found consistently better than least-squares and vendor-provided reconstruction in BLADE MRI with a SENSitivity Encoding (SENSE) factor of 2. This warrants further development of the integrated computing system with the scanner.


  • Open Access

    Python Non-Uniform Fast Fourier Transform (PyNUFFT): An Accelerated Non-Cartesian MRI Package on a Heterogeneous Platform (CPU/GPU)

    by Jyh-Miin Lin

    A Python non-uniform fast Fourier transform (PyNUFFT) package has been developed to accelerate multidimensional non-Cartesian image reconstruction on heterogeneous platforms. Since scientific computing with Python encompasses a mature and integrated environment, the time efficiency of the NUFFT algorithm has been a major obstacle to real-time non-Cartesian image reconstruction with Python. The current PyNUFFT software enables multi-dimensional NUFFT accelerated on a heterogeneous platform, which yields an efficient solution to many non-Cartesian imaging problems. The PyNUFFT also provides several solvers, including the conjugate gradient method, ℓ1 total variation regularized ordinary least square (L1TV-OLS), and ℓ1 total variation regularized least absolute deviation (L1TV-LAD). Metaprogramming libraries have been employed to accelerate PyNUFFT. The PyNUFFT package has been tested on multi-core central processing units (CPUs) and graphic processing units (GPUs), with acceleration factors of 6.3–9.5× on a 32-thread CPU platform and 5.4–13× on a GPU.


  • Open Access

    Python Non-Uniform Fast Fourier Transform (PyNUFFT): multi-dimensional non-Cartesian image reconstruction package for heterogeneous platforms and applications to MRI

    by Jyh-Miin Lin

    This paper reports the development of a Python Non-Uniform Fast Fourier Transform (PyNUFFT) package, which accelerates non-Cartesian image reconstruction on heterogeneous platforms. Scientific computing with Python encompasses a mature and integrated environment. The NUFFT algorithm has been extensively used for non-Cartesian image reconstruction but previously there was no native Python NUFFT library. The current PyNUFFT software enables multi-dimensional NUFFT on heterogeneous platforms. The PyNUFFT also provides several solvers, including the conjugate gradient method, $\ell$1 total-variation regularized ordinary least square (L1TV-OLS) and $\ell$1 total-variation regularized least absolute deviation (L1TV-LAD). Metaprogramming libraries were employed to accelerate PyNUFFT. The PyNUFFT package has been tested on multi-core CPU and GPU, with acceleration factors of 6.3 - 9.5$\times$ on a 32 thread CPU platform and 5.4 - 13$\times$ on the GPU.

  • Effects of RF pulse profile and intra-voxel phase dispersion on MR fingerprinting with balanced SSFP readout

    by Su-Chin Chiu, Te-Ming Lin, Jyh-Miin Lin, Hsiao-Wen Chung, Cheng-Wen Ko, Martin Büchert, Michael Bock