Image reconstruction in x-ray computer tomography (CT) scanners lags far behind the data acquisition. For example a whole-body scan using the latest 64-slice medical CT scanners, with sub-millimeter slice thickness, which takes about 10 seconds to acquire, requires about 20 times as long to reconstruct. This computational lag occurs despite the use of expensive special-purpose computing hardware. Faster image reconstruction is critical for life-threatening trauma cases, and is key to enhancing the use of x-ray CT as a dynamic real-time imaging modality for cardiac imaging, fluoroscopy and interventional applications. Furthermore, faster reconstruction is needed to enable new applications using computationally demanding iterative reconstruction methods that can overcome metal artifacts, improve image quality, and reduce the x-ray dose required to achieve acceptable image quality. Similarly, faster reconstruction is desired in CT security imaging, especially for scanning of checked luggage at airports. To date, acceleration of image reconstruction in CT scanners has been achieved only by scaling the computing hardware. However, because of the ever-increasing speed demands, simply scaling the hardware (parallelizing, upsizing, or using more processors) carries a prohibitive price tag. The objective of this project is to achieve very large speed-ups through the use of more clever image reconstruction algorithms (i.e. more clever mathematics), which were developed and patented at the University of Illinois and have been licensed to InstaRecon. These algorithms reduce the mathematical operation counts for the reconstruction by factors of 10 to 50 for 512 W 512 pixel images typical in medical applications. We propose to develop, evaluate, and validate a hardware prototype of an ultra-fast algorithmically-accelerated image reconstruction engine for three- dimensional cone-beam CT scanners. The hardware platform will be a reconfigurable field-programmable object array (FPOA), which offers an attractive tradeoff between cost, speed, and flexibility.
Specific aims of this project are prototypes of (i) an ultra-fast algorithmically accelerated hardware backprojector for the 3D circular imaging scan geometry;(ii) a fast complete software reconstruction algorithm for the 3D helical cone beam geometry, for the so-called long object problem, applicable to diagnostic imaging;and (iii) an ultra-fast algorithmically accelerated complete hardware reconstructor for the 3D helical cone beam long object geometry.
We aim to provide a speed-up of at least 20W relative to two benchmarks: (i) conventional algorithms implemented on comparable hardware resources, and (ii) current best-in class commercial CT reconstruction rates. These speed-ups can be used to implement more sophisticated algorithms to produce better image quality and for low-dose imaging. Stringent control of image quality by both objective and subjective measures and experiments will be applied throughout the course of the project to ensure that the unprecedented speed-up is achieved while maintaining pristine image quality.
