Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for ra... more Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for radiation therapy treatment planning. The delineation procedure is typically performed manually which exposes two major issues: cost and reproducibility. Delineation is a time-consuming process that is subject to interand intra-observer variability. While methods have been proposed to predict GTV contours, typical approaches ignore variability and therefore fail to utilize the valuable confidence information offered by multiple contours. Materials and methods : In this work we propose an automatic GTV contouring method for softtissue sarcomas from X-ray computed tomography (CT) images, using deep learning by integrating interand intra-observer variability in the learned model. Sixty-eight patients with soft tissue and bone sarcomas were considered in this evaluation, all underwent pre-operative CT imaging used to perform GTV delineation. Four radiation oncologists and radiologists performe...
To quantify live-cell adhesion, a photonic crystal biosensor surface with an extracellular matrix... more To quantify live-cell adhesion, a photonic crystal biosensor surface with an extracellular matrix coating is monitored within a PCEM instrument to dynamically image changes in attached cell mass density during live stem cell attachment, spreading, and drug response.
Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for ra... more Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for radiation therapy treatment planning. The delineation procedure is typically performed manually which exposes two major issues: cost and reproducibility. Delineation is a time-consuming process that is subject to inter-and intra-observer variability. While methods have been proposed to predict GTV contours, typical approaches ignore variability and therefore fail to utilize the valuable confidence information offered by multiple contours. Materials and methods: In this work we propose an automatic GTV contouring method for softtissue sarcomas from X-ray computed tomography (CT) images, using deep learning by integrating inter-and intra-observer variability in the learned model. Sixty-eight patients with soft tissue and bone sarcomas were considered in this evaluation, all underwent pre-operative CT imaging used to perform GTV delineation. Four radiation oncologists and radiologists performed three contouring trials each for all patients. We quantify variability by defining confidence levels based on the frequency of inclusion of a given voxel into the GTV and use a deep convolutional neural network to learn GTV confidence maps. Results: Results were compared to confidence maps from the four readers as well as groundtruth consensus contours established jointly by all readers. The resulting continuous Dice score between predicted and true confidence maps was 87% and the Hausdorff distance was 14 mm. Conclusion: Results demonstrate the ability of the proposed method to predict accurate contours while utilizing variability and as such it can be used to improve clinical workflow.
Protein functions are largely affected by their conformations. This is exemplified in proteins co... more Protein functions are largely affected by their conformations. This is exemplified in proteins containing modular domains. However, the evolutionary dynamics that define and adapt the conformation of such modular proteins remain elusive. Here we show that cis-interactions between the C-terminal phosphotyrosines and SH2 domain within the protein tyrosine phosphatase Shp2 can be tuned by an adaptor protein, Grb2. The competitiveness of two phosphotyrosines, namely pY542 and pY580, for cis-interaction with the same SH2 domain is governed by an antagonistic combination of contextual amino acid sequence and position of the phosphotyrosines. Specifically, pY580 with the combination of a favourable position and an adverse sequence has an overall advantage over pY542. Swapping the sequences of pY542 and pY580 results in one dominant form of cis-interaction and subsequently inhibits the trans-regulation by Grb2. Thus, the antagonistic combination of sequence and position may serve as a basic design principle for proteins with tunable conformations.
The work presented in this dissertation addresses the optimization and application of a newly dev... more The work presented in this dissertation addresses the optimization and application of a newly developed imaging modality, named photonic crystal enhanced microscopy (PCEM) for label-free detection of the surface attached biological samples. Photonic crystal, supporting guided-mode resonances and providing local optical modes, can enhance the light-matter interaction on its surface and thus used for label-free detection and biosensing. One dimensional photonic crystal surfaces are designed and utilized as a iv ACKNOWLEDGMENTS First and foremost, I would like to thank my research advisor, Professor Brian Cunningham, for giving me the opportunity to work on this interesting research topic, and also for his great support and guidance throughout the past years. His wisdom, boundless energy and enthusiasm for research play an essential role for my research and will carry me on in my future work. His patience, devotion and continual encouragement to his students have been truly inspirational to me and helped me achieve my professional goals. I truly appreciate him for making this long journey of PhD study fulfilling and enjoyable as well as making my dream come true.
The ability to track individual cells in space over time is crucial to analyzing heterogeneous ce... more The ability to track individual cells in space over time is crucial to analyzing heterogeneous cell populations. Recently, microlaser particles have emerged as unique optical probes for massively multiplexed single-cell tagging. However, the microlaser far-field emission is inherently direction-dependent, which causes strong intensity fluctuations when the orientation of the particle varies randomly inside cells. Here, we demonstrate a general solution based on the incorporation of nanoscale light scatterers into microlasers. Two schemes are developed by introducing either boundary defects or a scattering layer into microdisk lasers. The resulting laser output is omnidirectional, with the minimum-to-maximum ratio of the angle-dependent intensity improving from 0.007 (−24 dB) to > 0.23 (−6 dB). After transfer into live cells in vitro, the omnidirectional laser particles within moving cells could be tracked continuously with high signal-to-noise ratios for 2 h, while conventional m...
2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2009
ABSTRACT Magnetic susceptibility differences exist near the interface of air/tissue in the ventra... more ABSTRACT Magnetic susceptibility differences exist near the interface of air/tissue in the ventral brain in fMRI (functional magnetic resonance imaging). These susceptibility differences will not only cause field inhomogeneity and its gradients which will result in image artifacts, but also cause shifts in the echo time for gradients echo acquisitions. The echo time shifts are caused by shifts in the effective k-space trajectory due to the gradients of the field inhomogeneity. Previous work has shown and validated methods for estimating the echo time shift based on the effective k-space trajectory. In this work, we demonstrate that accurate estimation of the R2* decay map (R2*=1/T2*) not only need to account for the field map and its gradients, but also needs to include the echo time shift. These changes in T2* sensitivity are directly related to changes in sensitivity in BOLD (blood oxygenation level dependent) fMRI studies.
Proper subcellular localization of focal adhesion kinase (FAK) is crucial for many cellular proce... more Proper subcellular localization of focal adhesion kinase (FAK) is crucial for many cellular processes. It remains, however, unclear how FAK activity is regulated at subcellular compartments. To visualize the FAK activity at different membrane microdomains, we develop a fluorescence resonance energy transfer (FRET)-based FAK biosensor, and target it into or outside of detergent-resistant membrane (DRM) regions at the plasma membrane. Here we show that, on cell adhesion to extracellular matrix proteins or stimulation by platelet-derived growth factor (PDGF), the FRET responses of DRM-targeting FAK biosensor are stronger than that at non-DRM regions, suggesting that FAK activation can occur at DRM microdomains. Further experiments reveal that the PDGF-induced FAK activation is mediated and maintained by Src activity, whereas FAK activation on cell adhesion is independent of, and in fact essential for the Src activation. Therefore, FAK is activated at membrane microdomains with distinct...
INTRODUCTION Many advanced MRI image acquisition and reconstruction methods see limited applicati... more INTRODUCTION Many advanced MRI image acquisition and reconstruction methods see limited application due to high computational cost in MRI. For instance, iterative reconstruction algorithms (e.g. non-Cartesian k-space trajectory, or magnetic field inhomogeneity compensation) can improve image quality but suffer from low reconstruction speed. General-purpose computing on graphics processing units (GPU) have demonstrated significant performance speedups and cost reductions in science and engineering applications. In fact, GPU can offer significant speedup due to MRI parallelized-data structure, e.g. multi-shots, multi-coil, multi-slice, multi-time-point, etc. We propose an implementation of iterative MR image reconstruction with magnetic field inhomogeneity compensation on multi-GPUs. The MR image model is based on non-Cartesian trajectory (i.e. spiral) in k-space, and can compensate for both geometric distortion and some signal loss induced by susceptibility gradients. THEORY Field inhomogeneity near air/tissue interfaces at the ventral brain (i.e. orbitofrontal cortex) leads to susceptibility artifacts, including geometric distortions and signal loss [1-2]. We have previously reported an imaging model including the physics of magnetic field inhomogeneity map and its 1 st derivate, which can compensate for both types of susceptibility artifacts [3]. Imaging model: The MR samples d acquired in k-space are described in Eq.(1). ρ(r) is the image estimate at spatial location r; k(t m) is the non-Cartesian trajectory in k-space; ω(r)=ω(x,y,z) denotes the magnetic field inhomogeneity, as shown in Eq(2), including off resonance constant ω n (in Hz), within-plane (G x,n , G y,n) and through-plane (G z,n) susceptibility gradients (in Hz/cm). The model is discretized using a 3D rectangle basis function φ n (x,y,z). Eq(3) shows Φ n (k(t m)), the Fourier Transform (after discretization) of the basis function at location k m , combined with the field inhomogeneity map and its gradients with Z-shimming gradients. Iterative CG solver: For image reconstruction, the gridding method can only compensate for the geometric distortion, but not signal loss (which is induced by the susceptibility gradients). Alternatively, our iterative algorithm can model the physics of field inhomogeneity and its gradients. In this work, we use a conjugate gradient iterative algorithm for reconstruction including correction for magnetic field inhomogeneity as described earlier. Therefore signal loss as well as signal distortion can be efficiently reduced and the resulting image quality is improved. We expected to gain a speedup by an order of magnitude for multi-GPU implementation. METHODS The reconstruction is implemented on a NVIDIA Tesla S1070 which containing 4 GT200 GPUs. Tesla S1070 yields 4 TFLOPS of peak performance and 408 GB/s memory access bandwidth, and 240 computing cores per processor. The reconstruction code is written using the NVIDIA CUDA 3.0, which supports the single-program, multiple-data (SPMD) parallel execution. The CPU used for comparison is a dual-core 2.4 GHz AMD Opterons with 8 GB of memory on operating system Fedora 10. Here, we extend our previous work [4-5] to field-compensated model described in theory. In this work, we use up to 4 GPUs to parallelize the computation of multiple slices of 3 different matrix sizes (64x64, 128x128, 256x256). Note that in our implementation, the data (including k-space trajectory, field map and its gradients) were stored on the GPU memory for computation of the whole conjugate gradient algorithm, therefore reducing the time-consuming data transfer and memory allocation. For multi-GPU support, a scheduler is implemented to start a single slice at a time on each GPU, while keeping a balanced load for all GPUs. RESULTS and DISCUSSION The performance comparison is shown in Table 1-2. For multi-GPUs, the speedup is 373x for 1 GPUs and 560x for 4 GPUs (with 64x64x4 matrix size). We also test different matrix sizes on 4 GPUs. For matrix size higher than 64x64 in each slice, tiling technique is implemented when using constant memory to speedup the memory operation. Therefore, the speedup reduces from matrix size 64x64 to 128x128 in each slice by considering the time for tiling. The speedup reaches 1102x (three orders of magnitude) for matrix size of 256x256x4. In summary, the computation time is significantly reduced by at least two orders of magnitude for multi-GPUs implementation of our iterative MR image reconstruction algorithm, and the speedup increases along with the number of working GPUs. The resulting images looks almost identical (Fig.1) between CPU and GPU, while the reconstruction is done with (a,c) or without (b,d) present of field map and its gradients. The differences (e,f) between GPU and CPU are negligible (range of 10-4) compared with resulting image intensity (around 1) in (a-d). CONCLUSION We proposed a fast multi-GPU-based implementation of our MR imaging model reconstruction algorithm with susceptibility artifacts compensation. The speedup is up to two orders of magnitudes and increases along with number of GPUs. The significant speed improvement reduces computation times down to a clinically acceptable scale.
We demonstrate a label-free biosensor imaging approach that utilizes a photonic crystal surface t... more We demonstrate a label-free biosensor imaging approach that utilizes a photonic crystal surface to detect surface attachment of individual dielectric and metal nanoparticles.
Adhesion is a critical cellular process that contributes to migration, apoptosis, differentiation... more Adhesion is a critical cellular process that contributes to migration, apoptosis, differentiation, and division. It is followed by the redistribution of cellular materials at the cell membrane or at the cell-surface interface for cells interacting with surfaces, such as basement membranes. Dynamic and quantitative tracking of changes in cell adhesion mass redistribution is challenging because cells are rapidly moving, inhomogeneous, and nonequilibrium objects, whose physical and mechanical properties are difficult to measure or predict. Here, we report a novel biosensor based microscopy approach termed Photonic Crystal Enhanced Microscopy (PCEM) that enables the movement of cellular materials at the plasma membrane of individual live cells to be dynamically monitored and quantitatively imaged. PCEM utilizes a photonic crystal biosensor surface, which can be coated with arbitrary extracellular matrix materials to facilitate cellular interactions, within a modified brightfield microsc...
Non-receptor protein kinases FAK and Src play crucial roles in regulating cellular adhesions, gro... more Non-receptor protein kinases FAK and Src play crucial roles in regulating cellular adhesions, growth, migration and differentiation. However, it remains unclear how the activity of FAK and Src is regulated during the differentiation process from mesenchymal stem cells (MSCs) to bone cells. In this study, we used genetically encoded FAK and Src biosensors based on fluorescence resonance energy transfer (FRET) to monitor the FAK and Src activity in live cells during the differentiation process. The results revealed that the FAK activity increased after the induction of differentiation, which peaked around 20-27 days after induction. Meanwhile, the Src activity decreased continuously for 27 days after induction. Therefore, the results showed significant and differential changes of FAK and Src activity upon induction. This opposite trend between FAK and Src activation suggests novel and un-coupled Src/FAK functions during the osteoblastic differentiation process. These results should provide important information for the biochemical signals during the differentiation process of stem cells toward bone cells, which will advance our understanding of bone repair and tissue engineering.
FAK and Src signaling play important roles in cell differentiation, survival and migration. Howev... more FAK and Src signaling play important roles in cell differentiation, survival and migration. However, it remains unclear how FAK and Src activities are regulated at the initial stage of stem cell differentiation. We utilized fluorescence resonance energy transfer (FRET)-based FAK and Src biosensors to visualize these kinase activities at the plasma membrane of human mesenchymal stem cells (HMSCs) under the stimulation of osteogenic, myoblastic, or neural induction reagents. Our results indicate that the membrane FAK and Src activities are distinctively regulated by these differentiation induction reagents. FAK and Src activities were both up-regulated with positive feedback upon osteogenic induction, while myoblastic induction only activated Src, but not FAK. Neural induction, however, transiently activated FAK and subsequently Src, which triggered a negative feedback to partially inhibit FAK activity. These results unravel distinct regulation mechanisms of FAK and Src activities during HMSC fate decision, which should advance our understanding of stem cell differentiation in tissue engineering.
Conference proceedings : ... Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual Conference, 2014
We demonstrate a novel microscopy-based biosensing approach that utilizes a photonic crystal (PC)... more We demonstrate a novel microscopy-based biosensing approach that utilizes a photonic crystal (PC) surface to detect protein-protein binding with the functionalized nanoparticles as tags. This imaging approach utilizes the measurement of localized shifts in the resonant wavelength and resonant reflection magnitude from the PC biosensor in the presence of individual nanoparticles. Moreover, it substantially increases the sensitivity of the imaging approach through tunable localized surface plasmon resonant frequency of the nanoparticle matching with the resonance of the PC biosensor. Experimental demonstrations of photonic crystal enhanced microscopy (PCEM) imaging with single nanoparticle resolution are supported by Finite-Difference Time-Domain (FDTD) computer simulations. The ability to detect the surface adsorption of individual nanoparticles as tags offers a route to single molecule biosensing with photonic crystal biosensor in the future.
Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for ra... more Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for radiation therapy treatment planning. The delineation procedure is typically performed manually which exposes two major issues: cost and reproducibility. Delineation is a time-consuming process that is subject to interand intra-observer variability. While methods have been proposed to predict GTV contours, typical approaches ignore variability and therefore fail to utilize the valuable confidence information offered by multiple contours. Materials and methods : In this work we propose an automatic GTV contouring method for softtissue sarcomas from X-ray computed tomography (CT) images, using deep learning by integrating interand intra-observer variability in the learned model. Sixty-eight patients with soft tissue and bone sarcomas were considered in this evaluation, all underwent pre-operative CT imaging used to perform GTV delineation. Four radiation oncologists and radiologists performe...
To quantify live-cell adhesion, a photonic crystal biosensor surface with an extracellular matrix... more To quantify live-cell adhesion, a photonic crystal biosensor surface with an extracellular matrix coating is monitored within a PCEM instrument to dynamically image changes in attached cell mass density during live stem cell attachment, spreading, and drug response.
Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for ra... more Background and purpose: The delineation of the gross tumor volume (GTV) is a critical step for radiation therapy treatment planning. The delineation procedure is typically performed manually which exposes two major issues: cost and reproducibility. Delineation is a time-consuming process that is subject to inter-and intra-observer variability. While methods have been proposed to predict GTV contours, typical approaches ignore variability and therefore fail to utilize the valuable confidence information offered by multiple contours. Materials and methods: In this work we propose an automatic GTV contouring method for softtissue sarcomas from X-ray computed tomography (CT) images, using deep learning by integrating inter-and intra-observer variability in the learned model. Sixty-eight patients with soft tissue and bone sarcomas were considered in this evaluation, all underwent pre-operative CT imaging used to perform GTV delineation. Four radiation oncologists and radiologists performed three contouring trials each for all patients. We quantify variability by defining confidence levels based on the frequency of inclusion of a given voxel into the GTV and use a deep convolutional neural network to learn GTV confidence maps. Results: Results were compared to confidence maps from the four readers as well as groundtruth consensus contours established jointly by all readers. The resulting continuous Dice score between predicted and true confidence maps was 87% and the Hausdorff distance was 14 mm. Conclusion: Results demonstrate the ability of the proposed method to predict accurate contours while utilizing variability and as such it can be used to improve clinical workflow.
Protein functions are largely affected by their conformations. This is exemplified in proteins co... more Protein functions are largely affected by their conformations. This is exemplified in proteins containing modular domains. However, the evolutionary dynamics that define and adapt the conformation of such modular proteins remain elusive. Here we show that cis-interactions between the C-terminal phosphotyrosines and SH2 domain within the protein tyrosine phosphatase Shp2 can be tuned by an adaptor protein, Grb2. The competitiveness of two phosphotyrosines, namely pY542 and pY580, for cis-interaction with the same SH2 domain is governed by an antagonistic combination of contextual amino acid sequence and position of the phosphotyrosines. Specifically, pY580 with the combination of a favourable position and an adverse sequence has an overall advantage over pY542. Swapping the sequences of pY542 and pY580 results in one dominant form of cis-interaction and subsequently inhibits the trans-regulation by Grb2. Thus, the antagonistic combination of sequence and position may serve as a basic design principle for proteins with tunable conformations.
The work presented in this dissertation addresses the optimization and application of a newly dev... more The work presented in this dissertation addresses the optimization and application of a newly developed imaging modality, named photonic crystal enhanced microscopy (PCEM) for label-free detection of the surface attached biological samples. Photonic crystal, supporting guided-mode resonances and providing local optical modes, can enhance the light-matter interaction on its surface and thus used for label-free detection and biosensing. One dimensional photonic crystal surfaces are designed and utilized as a iv ACKNOWLEDGMENTS First and foremost, I would like to thank my research advisor, Professor Brian Cunningham, for giving me the opportunity to work on this interesting research topic, and also for his great support and guidance throughout the past years. His wisdom, boundless energy and enthusiasm for research play an essential role for my research and will carry me on in my future work. His patience, devotion and continual encouragement to his students have been truly inspirational to me and helped me achieve my professional goals. I truly appreciate him for making this long journey of PhD study fulfilling and enjoyable as well as making my dream come true.
The ability to track individual cells in space over time is crucial to analyzing heterogeneous ce... more The ability to track individual cells in space over time is crucial to analyzing heterogeneous cell populations. Recently, microlaser particles have emerged as unique optical probes for massively multiplexed single-cell tagging. However, the microlaser far-field emission is inherently direction-dependent, which causes strong intensity fluctuations when the orientation of the particle varies randomly inside cells. Here, we demonstrate a general solution based on the incorporation of nanoscale light scatterers into microlasers. Two schemes are developed by introducing either boundary defects or a scattering layer into microdisk lasers. The resulting laser output is omnidirectional, with the minimum-to-maximum ratio of the angle-dependent intensity improving from 0.007 (−24 dB) to > 0.23 (−6 dB). After transfer into live cells in vitro, the omnidirectional laser particles within moving cells could be tracked continuously with high signal-to-noise ratios for 2 h, while conventional m...
2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2009
ABSTRACT Magnetic susceptibility differences exist near the interface of air/tissue in the ventra... more ABSTRACT Magnetic susceptibility differences exist near the interface of air/tissue in the ventral brain in fMRI (functional magnetic resonance imaging). These susceptibility differences will not only cause field inhomogeneity and its gradients which will result in image artifacts, but also cause shifts in the echo time for gradients echo acquisitions. The echo time shifts are caused by shifts in the effective k-space trajectory due to the gradients of the field inhomogeneity. Previous work has shown and validated methods for estimating the echo time shift based on the effective k-space trajectory. In this work, we demonstrate that accurate estimation of the R2* decay map (R2*=1/T2*) not only need to account for the field map and its gradients, but also needs to include the echo time shift. These changes in T2* sensitivity are directly related to changes in sensitivity in BOLD (blood oxygenation level dependent) fMRI studies.
Proper subcellular localization of focal adhesion kinase (FAK) is crucial for many cellular proce... more Proper subcellular localization of focal adhesion kinase (FAK) is crucial for many cellular processes. It remains, however, unclear how FAK activity is regulated at subcellular compartments. To visualize the FAK activity at different membrane microdomains, we develop a fluorescence resonance energy transfer (FRET)-based FAK biosensor, and target it into or outside of detergent-resistant membrane (DRM) regions at the plasma membrane. Here we show that, on cell adhesion to extracellular matrix proteins or stimulation by platelet-derived growth factor (PDGF), the FRET responses of DRM-targeting FAK biosensor are stronger than that at non-DRM regions, suggesting that FAK activation can occur at DRM microdomains. Further experiments reveal that the PDGF-induced FAK activation is mediated and maintained by Src activity, whereas FAK activation on cell adhesion is independent of, and in fact essential for the Src activation. Therefore, FAK is activated at membrane microdomains with distinct...
INTRODUCTION Many advanced MRI image acquisition and reconstruction methods see limited applicati... more INTRODUCTION Many advanced MRI image acquisition and reconstruction methods see limited application due to high computational cost in MRI. For instance, iterative reconstruction algorithms (e.g. non-Cartesian k-space trajectory, or magnetic field inhomogeneity compensation) can improve image quality but suffer from low reconstruction speed. General-purpose computing on graphics processing units (GPU) have demonstrated significant performance speedups and cost reductions in science and engineering applications. In fact, GPU can offer significant speedup due to MRI parallelized-data structure, e.g. multi-shots, multi-coil, multi-slice, multi-time-point, etc. We propose an implementation of iterative MR image reconstruction with magnetic field inhomogeneity compensation on multi-GPUs. The MR image model is based on non-Cartesian trajectory (i.e. spiral) in k-space, and can compensate for both geometric distortion and some signal loss induced by susceptibility gradients. THEORY Field inhomogeneity near air/tissue interfaces at the ventral brain (i.e. orbitofrontal cortex) leads to susceptibility artifacts, including geometric distortions and signal loss [1-2]. We have previously reported an imaging model including the physics of magnetic field inhomogeneity map and its 1 st derivate, which can compensate for both types of susceptibility artifacts [3]. Imaging model: The MR samples d acquired in k-space are described in Eq.(1). ρ(r) is the image estimate at spatial location r; k(t m) is the non-Cartesian trajectory in k-space; ω(r)=ω(x,y,z) denotes the magnetic field inhomogeneity, as shown in Eq(2), including off resonance constant ω n (in Hz), within-plane (G x,n , G y,n) and through-plane (G z,n) susceptibility gradients (in Hz/cm). The model is discretized using a 3D rectangle basis function φ n (x,y,z). Eq(3) shows Φ n (k(t m)), the Fourier Transform (after discretization) of the basis function at location k m , combined with the field inhomogeneity map and its gradients with Z-shimming gradients. Iterative CG solver: For image reconstruction, the gridding method can only compensate for the geometric distortion, but not signal loss (which is induced by the susceptibility gradients). Alternatively, our iterative algorithm can model the physics of field inhomogeneity and its gradients. In this work, we use a conjugate gradient iterative algorithm for reconstruction including correction for magnetic field inhomogeneity as described earlier. Therefore signal loss as well as signal distortion can be efficiently reduced and the resulting image quality is improved. We expected to gain a speedup by an order of magnitude for multi-GPU implementation. METHODS The reconstruction is implemented on a NVIDIA Tesla S1070 which containing 4 GT200 GPUs. Tesla S1070 yields 4 TFLOPS of peak performance and 408 GB/s memory access bandwidth, and 240 computing cores per processor. The reconstruction code is written using the NVIDIA CUDA 3.0, which supports the single-program, multiple-data (SPMD) parallel execution. The CPU used for comparison is a dual-core 2.4 GHz AMD Opterons with 8 GB of memory on operating system Fedora 10. Here, we extend our previous work [4-5] to field-compensated model described in theory. In this work, we use up to 4 GPUs to parallelize the computation of multiple slices of 3 different matrix sizes (64x64, 128x128, 256x256). Note that in our implementation, the data (including k-space trajectory, field map and its gradients) were stored on the GPU memory for computation of the whole conjugate gradient algorithm, therefore reducing the time-consuming data transfer and memory allocation. For multi-GPU support, a scheduler is implemented to start a single slice at a time on each GPU, while keeping a balanced load for all GPUs. RESULTS and DISCUSSION The performance comparison is shown in Table 1-2. For multi-GPUs, the speedup is 373x for 1 GPUs and 560x for 4 GPUs (with 64x64x4 matrix size). We also test different matrix sizes on 4 GPUs. For matrix size higher than 64x64 in each slice, tiling technique is implemented when using constant memory to speedup the memory operation. Therefore, the speedup reduces from matrix size 64x64 to 128x128 in each slice by considering the time for tiling. The speedup reaches 1102x (three orders of magnitude) for matrix size of 256x256x4. In summary, the computation time is significantly reduced by at least two orders of magnitude for multi-GPUs implementation of our iterative MR image reconstruction algorithm, and the speedup increases along with the number of working GPUs. The resulting images looks almost identical (Fig.1) between CPU and GPU, while the reconstruction is done with (a,c) or without (b,d) present of field map and its gradients. The differences (e,f) between GPU and CPU are negligible (range of 10-4) compared with resulting image intensity (around 1) in (a-d). CONCLUSION We proposed a fast multi-GPU-based implementation of our MR imaging model reconstruction algorithm with susceptibility artifacts compensation. The speedup is up to two orders of magnitudes and increases along with number of GPUs. The significant speed improvement reduces computation times down to a clinically acceptable scale.
We demonstrate a label-free biosensor imaging approach that utilizes a photonic crystal surface t... more We demonstrate a label-free biosensor imaging approach that utilizes a photonic crystal surface to detect surface attachment of individual dielectric and metal nanoparticles.
Adhesion is a critical cellular process that contributes to migration, apoptosis, differentiation... more Adhesion is a critical cellular process that contributes to migration, apoptosis, differentiation, and division. It is followed by the redistribution of cellular materials at the cell membrane or at the cell-surface interface for cells interacting with surfaces, such as basement membranes. Dynamic and quantitative tracking of changes in cell adhesion mass redistribution is challenging because cells are rapidly moving, inhomogeneous, and nonequilibrium objects, whose physical and mechanical properties are difficult to measure or predict. Here, we report a novel biosensor based microscopy approach termed Photonic Crystal Enhanced Microscopy (PCEM) that enables the movement of cellular materials at the plasma membrane of individual live cells to be dynamically monitored and quantitatively imaged. PCEM utilizes a photonic crystal biosensor surface, which can be coated with arbitrary extracellular matrix materials to facilitate cellular interactions, within a modified brightfield microsc...
Non-receptor protein kinases FAK and Src play crucial roles in regulating cellular adhesions, gro... more Non-receptor protein kinases FAK and Src play crucial roles in regulating cellular adhesions, growth, migration and differentiation. However, it remains unclear how the activity of FAK and Src is regulated during the differentiation process from mesenchymal stem cells (MSCs) to bone cells. In this study, we used genetically encoded FAK and Src biosensors based on fluorescence resonance energy transfer (FRET) to monitor the FAK and Src activity in live cells during the differentiation process. The results revealed that the FAK activity increased after the induction of differentiation, which peaked around 20-27 days after induction. Meanwhile, the Src activity decreased continuously for 27 days after induction. Therefore, the results showed significant and differential changes of FAK and Src activity upon induction. This opposite trend between FAK and Src activation suggests novel and un-coupled Src/FAK functions during the osteoblastic differentiation process. These results should provide important information for the biochemical signals during the differentiation process of stem cells toward bone cells, which will advance our understanding of bone repair and tissue engineering.
FAK and Src signaling play important roles in cell differentiation, survival and migration. Howev... more FAK and Src signaling play important roles in cell differentiation, survival and migration. However, it remains unclear how FAK and Src activities are regulated at the initial stage of stem cell differentiation. We utilized fluorescence resonance energy transfer (FRET)-based FAK and Src biosensors to visualize these kinase activities at the plasma membrane of human mesenchymal stem cells (HMSCs) under the stimulation of osteogenic, myoblastic, or neural induction reagents. Our results indicate that the membrane FAK and Src activities are distinctively regulated by these differentiation induction reagents. FAK and Src activities were both up-regulated with positive feedback upon osteogenic induction, while myoblastic induction only activated Src, but not FAK. Neural induction, however, transiently activated FAK and subsequently Src, which triggered a negative feedback to partially inhibit FAK activity. These results unravel distinct regulation mechanisms of FAK and Src activities during HMSC fate decision, which should advance our understanding of stem cell differentiation in tissue engineering.
Conference proceedings : ... Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual Conference, 2014
We demonstrate a novel microscopy-based biosensing approach that utilizes a photonic crystal (PC)... more We demonstrate a novel microscopy-based biosensing approach that utilizes a photonic crystal (PC) surface to detect protein-protein binding with the functionalized nanoparticles as tags. This imaging approach utilizes the measurement of localized shifts in the resonant wavelength and resonant reflection magnitude from the PC biosensor in the presence of individual nanoparticles. Moreover, it substantially increases the sensitivity of the imaging approach through tunable localized surface plasmon resonant frequency of the nanoparticle matching with the resonance of the PC biosensor. Experimental demonstrations of photonic crystal enhanced microscopy (PCEM) imaging with single nanoparticle resolution are supported by Finite-Difference Time-Domain (FDTD) computer simulations. The ability to detect the surface adsorption of individual nanoparticles as tags offers a route to single molecule biosensing with photonic crystal biosensor in the future.
Uploads
Papers by Yue Zhuo