京都大学 / 新制・課程博士 / 博士(医学) / 甲第23073号 / 医博第4700号 / 新制||医||1049(附属図書館) / 京都大学大学院医学研究科医学専攻 / (主査)教授 溝脇 尚志, 教授 黒田 知宏, 教授 増永 慎一郎 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DGAM

Sparse signals are abundant among both natural and man-made signals. Sparsity implies that the signal essentially resides in a small dimensional subspace. The sparsity of the signal can be exploited to improve its recovery from limited and noisy observations. Traditional estimation algorithms generally lack the ability to take advantage of signal sparsity. This dissertation considers several problems in the areas of biomedical engineering and geosciences with the aim of enhancing the recovery of information by exploiting the underlying sparsity in the problem. The objective is to overcome the fundamental bottlenecks, both in terms of estimation accuracies and required computational resources. In the first part of dissertation, we present a high precision technique for the monitoring of human respiratory movements by exploiting the sparsity of wireless ultra-wideband signals. The proposed technique provides a novel methodology of overcoming the Nyquist sampling constraint and enables robust performance in the presence of noise and interferences. We also present a comprehensive framework for the important problem of extracting the fetal electrocardiogram (ECG) signals from abdominal ECG recordings of pregnant women. The multiple measurement vectors approach utilized for this purpose provides an efficient mechanism of exploiting the common structure of ECG signals, when represented in sparse transform domains, and allows leveraging information from multiple ECG electrodes under a joint estimation formulation. In the second part of dissertation, we adopt sparse signal processing principles for improved information recovery in large-scale subsurface reservoir characterization problems. We propose multiple new algorithms for sparse representation of the subsurface geological structures, incorporation of useful prior information in the estimation process, and for reducing computational complexities of the problem. The techniques presented here enable significantly enhanced imaging of the subsurface earth and result in substantial savings in terms of convergence time, leading to optimized placement of oil wells. This dissertation demonstrates through detailed experimental analysis that the sparse estimation approach not only enables enhanced information recovery in variety of application areas, but also greatly helps in reducing the computational complexities associated with the problems.

In multiuser wireless networks, the base stations (BSs) rely on the channel state information (CSI) of the users to in order to perform user scheduling and downlink transmission. While the downlink channels can be easily estimated at all user terminals via a single broadcast, several key challenges are faced during uplink (feedback) transmission. Firstly, the noisy and fading feedback channels are usually unknown at the base station, and therefore, channel training is usually required from all users. Secondly, the amount of air-time required for feedback transmission grows linearly with the number of users. This domination of the network resources by feedback information leads to increased scheduling delay and outdated CSI at the BS. In this thesis, we tackle the above challenges and propose feedback reduction algorithms based on the theory of compressive sensing (CS). The proposed algorithms encompass both single and dual hop wireless networks, and; i) permit the BS to obtain CSI with acceptable recovery guarantees under substantially reduced feedback overhead, ii) are agnostic to the statistics of the feedback channels, and iii) utilize the apriori statistics of the additive noise to identify strong users. Numerical results show that the proposed algorithms are able to reduce the feedback overhead, improve detection at the BS, and achieve a sum-rate close to that obtained by noiseless dedicated feedback algorithms.

No description available.

No description available.

No description available.

No description available.

Traditional Magnetic Resonance Imaging (MRI) reconstruction methods, which may be highly time-consuming and sensitive to noise, heavily depend on solving nonlinear optimization problems. By contrast, deep learning (DL)-based reconstruction methods do not need any explicit analytical data model and are robust to noise due to its large data-based training, which both make DL a versatile tool for fast and high-fidelity MR image reconstruction. While DL can be performed completely independently of traditional methods, it can, in fact, benefit from incorporating these established methods to achieve better results. To test this hypothesis, we proposed a hybrid DL-based MR image reconstruction method, which combines two state-of-the-art deep learning networks, U-Net and Generative Adversarial Network with Cycle loss (CycleGAN), with a traditional data reconstruction method: Projection Onto Convex Sets (POCS). Experiments were then performed to evaluate the method by comparing it to several existing state-of-the-art methods. Our results demonstrate that the proposed method outperformed the current state-of-the-art in terms of higher peak signal-to-noise ratio (PSNR) and higher Structural Similarity Index (SSIM). / Electrical and Computer Engineering

Phosphorus magnetic resonance spectroscopy and spectroscopic imaging (31P-MRS/MRSI) non-invasively provide very important information regarding energy metabolism as they can detect high energy metabolites and membrane phospholipids in vivo. They have repeatedly proven their utility in the study of healthy and disease conditions, as many disorders are related to imbalances in bioenergetic processes. However, they are not often used in a clinic setting as there are technical challenges that lead to very long acquisition times. To address this issue, the present work focused on the implementation of two fast phosphorus magnetic resonance spectroscopic imaging (31P-MRSI) pulse sequences. The first one, "fidEPSI" uses a flyback echo planar readout trajectory calculated in real time to achieve an acceleration factor up to x10. The second, "fidepsiCS" further accelerates the acquisition by combining the flyback EPSI readout with a compressed sensing (CS) sampling scheme. For this latter approach two different data reconstruction processes were compared. Both sequences were tested in phantoms as well as in skeletal muscle and brain tissues of healthy volunteers. The results showed feasibility of the flyback Echo Planar Spectroscopic Imaging (EPSI) to acquire good quality data in a fraction of the time when compared to traditional phase encoded MRSI. Furthermore, the compressed sensing approach was used in an exercise-recovery paradigm to evaluate skeletal muscle high energy phosphate dynamics, achieving a temporal resolution of 9 seconds. Additionally, the comparison of CS reconstruction algorithms suggested that a low-rank approach is more suitable for 31P-MRSI data, compared to traditional thresholding, due to the fact that it exploits the sparsity of the NMR signal as the least number of spectral peaks rather than the fewest amount of non-zero values. Overall, this thesis presents new accelerated methods for the acquisition of 31P-MRSI, and its use in the evaluation of energy metabolism. / Thesis / Doctor of Philosophy (PhD)

In most applications, the structure components have to be tested under different loading conditions before being placed in operation. A reliable and low cost measuring technique is desirable. However, most currently employed measuring approaches can only provide the structural response at several discrete locations. The accuracy of the measurements varies with the location and orientation of the sensors. Practically, it is not possible to place sensors at all the critical locations for different excitations. Therefore, an approach that derives the full field response using a limited set of measured data is desirable. In contrast to experimental full field measurement techniques, the expansion approach involves analytically expanding the limited measurements to all the degrees of freedom of the structure. Among all the analytical methods, the modal expansion method is computationally efficient and thus more suitable for real time expansion of measured data. In this method, the full-field response is approximated by the linear combination of mode shapes. In previous studies, the modal expansion method is limited by errors from mode aliasing, inaccuracy of the calculated mode shapes and the noise in measurements. In order to overcome these limitations, the modal expansion method is enhanced by mode selection and error compensation in this study. First, the key parameters used in modal expansion method were analyzed using a cantilever beam model and a method for optimal placement of sensors was developed. A mode selection method and error compensation method based on operation modal analysis and adaptive compressed sensing techniques were then developed to reduce the effects of mode aliasing, mode shape inaccuracy and measurement noise. The developed approach was further tested virtually using a numerical model of rotor 67. The numerical model was created using a two-way coupled fluid structure interaction technique. By developing these methods, the enhanced modal expansion approach can provide full field response for structures under different load conditions. Compared to the traditional modal expansion method, it can expand the data with high noise and under general dynamic loading. / Doctor of Philosophy / Accurate knowledge of the strain and stress at critical locations of a given structure is crucial when assessing its integrity. However, currently employed measuring approaches can only provide the structural response at several discrete locations. Practically, it is not possible to place sensors at all the critical locations for different excitations. Therefore, an approach that derives the full field response using a limited set of measured data is desirable. Compared to experimental full field measurement techniques, the expansion approach is focused on analytically expanding the limited measurements to all the degrees of freedom of the structure. Among all the analytical methods, the modal expansion method is computationally efficient and thus more suitable for real-time expansion of measured data. The current modal expansion method is limited by errors from mode aliasing, inaccuracy of the mode shapes, and the noise in measurements. Therefore, an enhanced method is proposed to overcome these shortcomings of the modal expansion. The following objectives are accomplished in this study: 1) Develop a method for optimal placement of sensors for modal expansion; 2) Eliminate the mode aliasing effects by determining the significance of participated modes using operational modal analysis techniques; 3) Compensate for the noise in measurements and computational model by implementing the compressed sensing approach. After accomplishing these goals, the developed approach is able to provide full field response for structures under different load conditions. Compared to the traditional modal expansion method, it can expand the data under dynamic loading; it also shows promise in reducing the effects of noise and errors. The developed approach is numerically tested using fluid-structure interaction model of rotor 67 fan blade.