When a new medical imaging device shall be purchased, one usually comes to a point where specsheets of the different vendors are compared. But when it comes to realistic performance parameters of CT machines, there are not many sources of information available (it's up to you whether sales agents count as a reliable source of information).

The Problem: Lack of Information

Regarding large bore CT scanners, the Centre for Evidence-based Purchasing (CEP), which can be found on the internet at, offers some useful reports, which should help in the purchasing process. One report contains comparative technical specifications.

However, specsheets have drawbacks, for at least three reasons:

We have decided to buy a Toshiba AquilionLB last year, for a number of reasons. In the process, we often had the feeling to decide in doubt - due to the poor information situation. Except for the CEP reports, the WWW was no great help either.

The Solution: Visit Our Site!

The situation has to be improved. On this and some more pages to come (on other topics), we want to present the results of our performance analyses regarding our own CT scanner. All measurements are done as objectively as possible, and the processes are (hopefully) described as clearly and completely as possible, so that they will hopefully be useful to you.

Image Reconstruction Performance - Our Results

No special tools are required for the tests, and you may easily reproduce them. If you want to be very accurate, use a stopwatch.

We scanned a phantom with our standard head & neck protocol (settings: see sidebar). We measured the time (in seconds) from the end of the scan to the point where the image counter on the scan console reached zero. (The image counter, counting down, displays the number of slices which still have to be reconstructed.) When the counter has reached zero, one can proceed with the images just acquired and export them to the TPS. Images can't be exported before they are reconstructed - a trivial truth.

Dividing the total reconstruction time (in seconds) by the number of slices gives the reconstruction time per slice (in seconds). The inverse is the number of reconstructed images per second (ips):

AquilionLB reconstruction performance
(data collected by Niko Metz)

One can see that we are far below the promised 4 images per second for useful FOVs, and that the specification "up to 10 images per second" is a joke, because up to is always true.

For typical radiotherapy planning situations, where the Field of View (FOV) is always larger than 400 mm, the number of images per second is 1 or less. If a different collimation like 2.0 mm x 16, and 2 mm reconstructed image thickness is chosen, the number of ips will be the same, for the same FOV. The FOV really is the dominant parameter regarding image reconstruction performance.


The impact of Toshiba's poor image reconstruction performance, in other words whether it bothers you or not, is strongly depenent on your way to work.

If time is at a premium, either because a patient with a full bladder is on the table or the patient wears an incomfortably tight thermoplastic mask, the performance result may even change your workflow, switching from a 1-day to a 2-day virtual simulation process, with all potential drawbacks (repositioning accuracy etc).

But if you choose the 2-day simulation workflow in the first place, poor image reconstruction performance possibly won't bother you much.

We are not satisfied with this situation. Coming from a Siemens Sensation Open scanner, where reconstruction at 500 mm FOV is extremely fast, we think this performance bottleneck on the AquilionLB scanner should be removed as soon as possible.

back to Medical Physics home

Valid XHTML 1.0 Strict