Spin-off touts algorithms to speed processing


The reconstruction of CT data lags far behind acquisition, creating a gulf that widens with each new generation of scanners. The problem is most pronounced on the leading edge of clinical use: cardiovascular, fluoroscopic, and interventional applications.

The reconstruction of CT data lags far behind acquisition, creating a gulf that widens with each new generation of scanners. The problem is most pronounced on the leading edge of clinical use: cardiovascular, fluoroscopic, and interventional applications.

Vendors are trying to keep up by adding more powerful computing engines. These, however, add cost and raise operational issues, including reliability and heat dissipation.

Staff at InstaRecon, a start-up company in Illinois, believe they can do better. They have assembled a portfolio of image reconstruction algorithms particularly suited for CT scanners. The algorithms attack the problem at its source: the data. They speed up reconstruction by a factor of more than 20 without compromising image quality, according to Yoram Bresler, president, chief technology officer, and cofounder of InstaRecon.

The need is becoming ever more apparent. Current 64-detector row CT scanners acquire data at the rate of up to 300 slices per second. Siemens's new Somatom Definition, with its dual-source x-ray chains, bumps that number by a factor of two. The data processing capabilities of these scanners, however, lags far behind, reconstructing images at only about 20 slices per second, Bresler said.

True real-time reconstruction, at the same rate as the data are acquired, would need to happen 15 to 30 times faster than it does now. InstaRecon's technology can affordably provide this acceleration, according to Bresler.

"This will not only increase patient throughput in the scanner, but also enable new applications in cardiac and interventional CT," he said.

InstaRecon operates from a berth in the University of Illinois Research Park, an incubator for fledgling corporations founded to commercialize spin-offs from research done at the university. The company is now beating the corporate bushes of the medical imaging industry, looking for OEMs to incorporate its algorithms into high-end CT scanners and possibly MR systems, as well as SPECT and PET cameras.

Bresler contends that many applications that require special-purpose hardware to achieve necessary reconstruction speeds could be satisfied by InstaRecon's software-based solutions using standard off-the-shelf single- or multiprocessor PCs. Conversely, these algorithm, working in concert with specially designed hardware, could pull the most demanding applications, currently beyond the reach of medial imaging, within its grasp.

"If special-purpose hardware can do 'x' amount of throughput with the current algorithms, we will take that and do '10x' improvements to enable these new features," said Sanjiv Chopra, acting CEO of InstaRecon.

These solutions would be built into the higher end premium-type scanners. But InstaRecon's family of algorithms might also be harnessed to make emerging compact CT scanners, designed for cost-sensitive markets such as China, less costly.

"They offer the opportunity to use simpler reconstruction hardware, providing low-power, high-reliability solutions that run on dual-core or quadcore Pentium-type systems," Chopra said.

InstaRecon's algorithms are designed to tackle the most computationally intense operations. These are encountered when performing backprojection and reprojection computations. These are the two bottlenecks in tomographic reconstruction, according to Bresler.

Backprojection is the assignment of attenuation values to individual pixels along the path of an x-ray as it travels through the patient. Reprojection is the process of simulating what the scanner does, so as to ensure that the values determined by backprojection are correct. In CT, reprojection is used to correct flaws in the data, such as beam-hardening artifacts.

These processes can be speeded up through the use of iterative algorithms, such as those in InstaRecon's mathematical family. These algorithms go back and forth, repeating computational processes until errors are corrected. Applying them may offer the added advantage of cutting x-ray dose, as more computing can be traded for higher dose to get the same image quality, he said.

"If you can afford to do more computing, you can actually get by with lower x-ray dose," Bresler said.

InstaRecon's algorithms can be applied to SPECT, PET, and even security CT scanners and industrial manufacturing using nondestructive testing with digital x-rays. They could also have a positive effect on projection MRI, a mode seldom applied because the computation needs are so high, Bresler said.

"Projection MRI has some advantages in motion correction, but it is not very widely used now because of the computational expense," he said. "With our algorithm, it could become a more dominant model."

The acceleration provided by these algorithms translates proportionally to lesser requirements on the computational hardware necessary to achieve a given speed or resolution. The bottom line is a lower cost for the reconstruction engine or improved accuracy, resolution, and speed for the scanner.

Bresler and colleagues are now working with OEMs to determine the optimal trade-offs around image quality as a prelude to tweaking their algorithms to suit specific products.

Related Videos
Practical Insights on CT and MRI Neuroimaging and Reporting for Stroke Patients
Expediting the Management of Incidental Pulmonary Emboli on CT
Related Content
© 2023 MJH Life Sciences

All rights reserved.