Reconfigurable acceleration of 3D image registration
read more
Citations
A Hybrid Similarity Measure Framework for Multimodal Medical Image Registration
Customisable Multi-Processor Acceleration of Inductive Logic Programming.
Architectural and algorithmic design for embedded medical imaging
Hybrid FPGA : architecture and interface
References
Nonrigid registration using free-form deformations: application to breast MR images
Medical image registration
Accuracy-Guaranteed Bit-Width Optimization
Difference-Based Partial Reconfiguration
FPGA-based computation of free-form deformations in medical image registration
Related Papers (5)
Frequently Asked Questions (12)
Q2. How many pixels are required for the final SSD?
For single interpolated output value which updates the final SSD, 1 pixel from target image and 8 pixels from source image are required.
Q3. How many parallel cores can the authors use to evaluate 30 different transformations?
With 30 parallel cores on chip, the authors can evaluate 30 different transformations concurrently, resulting in a throughput of 1.168ms per evaluation.
Q4. What are the main reasons why IR is becoming more popular in practical applications?
Despite the flexibility in software implementations, the improving resolution in imaging equipments, the increasing amount of images to be processed and the need of real time diagnosing ability make dedicated hardware platforms increasingly attractive in practical IR applications as higher computing power is required.
Q5. How many steps are required to advance the projected coordinated?
3.For | −→ xd| ≤ 0.5, 1/| −→ xd| accumulation steps are required to advance the projected coordinated to the next integral pixelcube.
Q6. What are the main applications of reconfigurable platforms?
Reconfigurable platforms have been widely adopted in accelerating different computational intensive digital signal processing algorithms.
Q7. How many times of the processing speed is required to prevent the hardware from being idle?
To prevent the hardware from being idle, the memory bandwidth in random access mode must be at least 9 times of the processing speed which is not realistic in most environments.
Q8. What can be done to improve the efficiency of the cache?
Under the limited cache size, the authors can improve cache efficiency by capturing application specific information, such as custom cache design for predictable access pattern.
Q9. What is the time for an evaluation process in a fully pipelined architecture?
The time for an evaluation process in a fully pipelined architecture isTeva = N × ρ × (Mtarget + Msource)/f, (8)where Mtarget and Msource are the number of clock cycles for memory access reading data from target and source images.
Q10. How many pixel values can be cached?
When 2 < | −→ xd|, no previously fetched pixel values can be reused along the X direction and thus no cache can help reducing the external memory reference.
Q11. What is the probability of advancing to the next integral pixel cube?
As it is impossible to stay in the previously projected pixel cube, the cache system for | −→ xd| < 0.5 is not applicable in this case.
Q12. What is the main purpose of this paper?
This paper describes a novel approach for optimizing a reconfigurable accelerator for an IR kernel derived from the ITK package, that takes into account data and platform dependent parameters.