An estimation of a resected cancer lodge localization after breast tumor surgery is a challenging task during radiotherapy planning. Knowledge about the tumor lodge position and shape could improve the radiation dose ...
详细信息
ISBN:
(纸本)9788362065301
An estimation of a resected cancer lodge localization after breast tumor surgery is a challenging task during radiotherapy planning. Knowledge about the tumor lodge position and shape could improve the radiation dose distribution. However, the tumor no longer exists after the surgery, but information about its position is available in the 3D image acquired before the surgery. Therefore, image registration algorithms can be used to estimate the tumor lodge localization and potentially improve the radiotherapy planning. In this work, we evaluate different variants of a Demons image registration algorithm. The nonparametric Demons algorithms are compared to a parametric registration procedure, the B-Splines free-form deformations. The results are evaluated using a target registration error and a medical expert visual inspection. The results show that for small deformations, the diffeomorphic, symmetric Demons are the most reliable, but for larger deformations, parametric B-Splines free-form deformations provide better results. Results demonstrate that there is still a place for a specialized algorithm development.
Low power technologies for VLSI allows tightly coupled complex processing parts near video sensors, making so-called smart sensors. They can be miniaturized for semi-automatic in vivo human body exploration. However s...
Low power technologies for VLSI allows tightly coupled complex processing parts near video sensors, making so-called smart sensors. They can be miniaturized for semi-automatic in vivo human body exploration. However smart sensors designs are more and more complexes to be simulated using traditional methods, new approaches are now emerging.
This paper presents the design and implementation of a joint interpolation-based QR decomposition and lattice reduction processor for the MIMO detection in 4 x 4 multiple-input multiple-output (MIMO) orthogonal freque...
详细信息
This paper presents the design and implementation of a joint interpolation-based QR decomposition and lattice reduction processor for the MIMO detection in 4 x 4 multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems. The proposed algorithm considers the coherence bandwidth in the OFDM spectrum to reduce the computational complexity of the QR decomposition and lattice reduction. This study also proposes a MIMO preprocessing architecture and a time scheduling algorithm for allocating the tasks of the processing elements. The hardware analysis results show that the proposed design method yields the smallest area and processing time (AT) product compared to the baseline architectures under most channel environments. The proposed processor was designed and implemented in TSMC 90nm 1P9M CMOS technology. The proposed processor achieves at most 6.592 M matrix/s with 135.14 MHz clock speed and 220.68 K gates.
Compute imageprocessing algorithm on block or tile is a solution to realize an efficient parallelization on multi-processor system. In this session a GPU implementation of Block-Matching and 3-Dimensional filter (BM3...
详细信息
Compute imageprocessing algorithm on block or tile is a solution to realize an efficient parallelization on multi-processor system. In this session a GPU implementation of Block-Matching and 3-Dimensional filter (BM3D) denoising algorithmm is compared to a CPU implementation. Also a method to determine optimal 2D image tile sizing using constraint programming is implemented on a multi-core processor to optimize the system performance.
Fully integrated CMOS frequency-modulated continuous-wave radar ICs are under development, in which computing FFTs cost a significant amount of energy. In this paper we introduce a power-efficient FFT solution which e...
详细信息
This session proposes four papers, that, although all related to imageprocessing, offer a great variety of algorithms, application domains and implementation targets. The algorithms include matrix processing, optical...
详细信息
ISBN:
(纸本)9781509030859
This session proposes four papers, that, although all related to imageprocessing, offer a great variety of algorithms, application domains and implementation targets. The algorithms include matrix processing, optical flow, the computation of image features with the Histogram of Oriented Gradients and hyperspectral imaging. The implementation targets include Intel processors, ASIPs, FPGAs and a Massively Parallel Processor Array. The session can thus appeal to all researchers interested in the implementation of computationally intensive imageprocessing algorithms on a variety of platforms.
Being able to estimate power and energy consumptions at high level in embedded systems design flows is crucial for nowadays complex applications and systems where low level approaches are not usable. Early estimates, ...
Being able to estimate power and energy consumptions at high level in embedded systems design flows is crucial for nowadays complex applications and systems where low level approaches are not usable. Early estimates, when good, naturally lead to decisive optimization opportunities. The session will discuss two important points in this context: the modeling of operating systems energy overhead and the refinement of behavior analysis for power estimation.
Deep learning has achieved outstanding performance in many fields such as image classification and target recognition. Recently multiple research efforts are focusing on deep learning to medical imageprocessing. Whil...
详细信息
ISBN:
(纸本)9781538607909
Deep learning has achieved outstanding performance in many fields such as image classification and target recognition. Recently multiple research efforts are focusing on deep learning to medical imageprocessing. While it is common in imageprocessing to apply transfer learning for problems with small sample sizes, the statistics of histopathological stains are known to be very different from the photographic RGB images in common deep learning imaging tasks such as imagenet and MIT Places. This paper evaluates the performance of fine-tuned models on Haematoxylin and Eosin(H&E) histopathology stain data. Furthermore, to analyze the performance of different deep learning architectures on these domains, we compare three convolutional neural network(CNN) architectures in various settings. Finally, the impact of the size of the context of training samples is evaluated. We use the BreaKHis dataset consisting of H&E stained microscopical scans of breast cancer tissue [1]. Our results show that fine-tuned architectures perform favorably over neural networks that are trained from scratch in terms of accuracy and patient rate.
imageprocessing algorithms in today electronics market pose the need for increasing computation capabilities at a limited power budget. Modern applications in the robotics and cyber-physical systems domains require i...
详细信息
imageprocessing algorithms in today electronics market pose the need for increasing computation capabilities at a limited power budget. Modern applications in the robotics and cyber-physical systems domains require image acquisition, analysis and information extraction to be executed on embedded low-power, portable and autonomous devices, requiring novel architecture solutions to be studied and implemented. The papers in this session present three design cases targeting edge-cutting applications, exploiting heterogeneous embedded computing platforms.
This paper presents a design method of reversible integer quaternionic paraunitary filter banks (Int-Q-PUFB) using the adder-based distributed arithmetic (DA(Sigma)) for implementation multiplier block-lifting structu...
详细信息
ISBN:
(纸本)9788362065301
This paper presents a design method of reversible integer quaternionic paraunitary filter banks (Int-Q-PUFB) using the adder-based distributed arithmetic (DA(Sigma)) for implementation multiplier block-lifting structure modules. The proposed quaternion multiplier (Q-MUL) and 8-channel Int-Q-PUFB processors are implemented on the FPGA Xilinx Zynq 7010. The total magnitude response of analysis-synthesis system based on the given Int-Q-PUFB shows that the 8-channel 8 x 24 Int-Q-PUFB is perfect reconstruction filter bank for finite precision. Compared to known solutions of Int-Q-PUFB using block-lifting structure based on the CORDIC devices and ROM-based distributed arithmetic the given DA(Sigma)-based Int-Q-PUFB have more less implementation complexity and latency.
暂无评论