Due to the greater availability of CT/MRI perfusion imaging, it is widely accepted as a diagnostic tool for cerebrovascular disorders. Although the deconvolution method is generally employed for analysis, there is significant variation in the actual implementation of the deconvolution algorithm among analysis environments.

An identical set of the entire currently available CT/MRI perfusion data, which were obtained from different scanners in several institutes, was fed into the analyzing systems. We found that there was a substantial variation and discrepancy in the output [1,2]. Hence, it is necessary to find an optimal algorithm that produces an accurate and reproducible output.

The mean transit time (MTT) maps from an identical CT perfusion dataset that are computed by different workstations supplied by five vendors (A-E). There is substantial variation in the appearance of the maps; these variations include the definition of cerebral ischemia in the area of the left middle cerebral artery.

Having reviewed several algorithms, ASIST-Japan recommends the block-circulant singular value decomposition (bSVD) algorithm [3] for clinical use because it is essentially insensitive to the variations in the injection delay of contrast medium. Further, of all the other reviewed algorithms, this algorithm is the most robust to noise. We are now working with the vendors of scanners and analysis softwares to implement bSVD into their commercial systems.

Cerebral blood flow (CBF) maps obtained from different workstations supplied by three vendors (A-C); these workstations employ the bSVD algorithm. The images are displayed using the normalized window width (WW)/window level (WL) and a standard look-up table (a-LUT) recommended by ASIST-Japan. There is minimal variation in the parameter maps when they are computed by the same algorithm and are displayed under normalized conditions.