Friday, March 2, 2012

ECR 2012 - workstation face-off

This year the face-off session at ECR for the fifth time with 6 participating companies. Aycan, GE Healthcare, Philips Healthcare, Siemens Healthcare, TeraRecon, and Vital Images.
The cases presented were prepared for four weeks preceding the conference. Two cases were presented with five minutes per company per case.
The first case was a cardiac case.
First task was to assess a four dimensional CT dataset. 3D visualization and assessment of the grafts in this patient was requested. Next analysis of the original RCA was requested with plaque analysis. Finally, LV function was performed.
Aycan, like all companies, showed automated segmentation of the heart.
The lima graft posed a bigger problem with full manual segmentation by Aycan. Overall the Aycan system proved to be less automated overall and required a lot of manual interaction and ran out of time.
Philips uses a model based approach where all 'components' are automatically segmented and colour coded. Suprisingly end systolic and end diastolic phases were selected manually while all phases were already segmented and thus this could be easily automated.
Terarecon also demonstrated their iPad integration that can be used for the communication with referring physicians. Their process is fast and automatic and they had 30 seconds left.
Siemens demonstrated the preprocessed workflow of syngo.via. Very automatic too with everything segmented and labelled. As an extra they showed the ability to do right ventricular analysis and had 20 seconds left.
Vital Images showed one click segmentation of the LIMA graft. The segmentation of the LV was a more manual process of defining the location of the axis of the LV after which the ejection fraction is computed.
GE Healthcare was the last to present the cardiac case. Heart segmentation contained a lot of remaining parts of the thorax and was not complete. Segmentation of the LIMA and RCA was manual and rather time consuming. The LV volume segmentation was automatic but required some manual adjustment to delete certain cavities from the volume. They managed to finish right in time.
Not supprisingly it was demonstrated, as always, that the LV function quantification varied between the different vendors.
The second case was an oncological case.
The tasks were to analyse a baseline and two followup CT scans. Specifically they were asked to define response, show simultaneous view of all timepoints and measure mass volume over different timepoints.
All workstations were up to this task, but the interesting thing was the variety of segmentation methods used.
Region growing segmentaion was shown by Terarecon which is a very straightforward way but requires more manual interaction.
Diameter definition segmentation was shown by Siemens that allows defining a rough centerline through the lesion with manual adjustment after the initial first segmentation.
A mouse over segmentation was shown by Vital Images which allowed the fastest segmentation with just a couple of mouse clicks.
Segmentation by GE was made using a edge based segmentation using a livewire which has to be performed at multiple levels to get the volume segmentation by interpolation.
Aycan required fully manual, polynomial segmentation of the lesion by defining a number of point on the borders in 2D at multiple slices. This required a lot of clicks and was the most manual way of all software companies. They were the only one running into time problems again.
Philips showed a new segmentation they call smart ROI which looks similar to the region growing algorithm but seems to be faster to operate.
Here also, the volume measurements were, off course, varying a lot. However, it was shown that the response reported by the different systems were similar.

There is no way to define a clear winner from sessions like this. All worstations perform their task and many issues are user and preference dependent. I would say though that based on what was demonstrated Aycan was the 'looser' however this might change when the price of the solutions is taken into account.


  1. We feel not bad as the "loser", when you consider the price and the size of the aycan organisation compared to all other vendors, who participated.

    At ECR 2013 Face-off aycan did accomplish all tasks within time (except the lung nodule CAD, which we don't support). Some vendors did not make it in time.

    aycan proved to fulfill complex, challanging clinical tasks at the Workstation Face-Off 2013.

    Stephan Popp
    aycan Digitalsysteme GmbH

  2. Fully agree with you, as I stated in my blogpost the price should be taken into account and face-off sessions, IMHO, don't have winners or losers since it is very difficult to compare workstations in that way. Unfortunately I was not at ECR this year so I can't comment on the face-off either.