[Gate-users] Re: [Gate-devel] Strange behavior of Gate_v3.1.1 and Gate_v3.1.2

[Sebastien JAN]

Hi,

Please, to try to understand what happens, could you send me your 
benchmarks results (PET & SPECT) for your 2 configurations (v3.1.1 & 
v3.1.2) ?
And could you tell me what is your instalation set-up (gcc version, 
linux distrib....) ?

Thanks
Sebastien JAN


> Dear Developers,
>
> We are concerned that there is something wrong with Gate_v3.1.2. We 
> have been running simulations for quite a long time using Gate_v3.1.1 
> and recently we have installed the new version of Geant (v4.9.0_p01 ) 
> and ROOT (v5.14.00) together with Gate_v3.1.2. The results we obtain 
> running the same simulations as before, totally defer from each other. 
> The simulation parameters are as follows:
>
> Cubic water phantom 30x30x30cm, not voxelized or voxelized in 1cm or 
> 0.4cm cubic voxels, center in (0,0,0)
> The source is in the center (0,0,0), slightly translated with 1e-7cm 
> to avoid setting up the source on the intersect between voxels
> Source activity is 1e6 bq.
> The simulation Run is 10s long.
> We are using General Electric scanner.
>
> We are interested in the total singles and coincidences produced by 
> the simulations. The number of singles we obtain from both versions 
> match quite well. But the problem is that the number coincidences 
> (true scattered and true unscattered) differ from each other by an 
> order of magnitude, which in our opinion is quite disturbing. 
> Performing one and the same simulation (with exactly the same mac 
> files in exactly the same directory) with one and the same initial 
> random seeds gives a huge difference in the number of coincidences, 
> which obviously is statistically significant. An example is shown below:
>
> *** Gate_v3.1.1 ***
> Phantom: WaterBox, 1.0cm Vox
> Singles: 708469
> Random Coins.: 109
> True Unscattered Coins.: 27304
> True Scattered Coins.: 17191
> Total: 44604
> Events: 9995991
>
> *** Gate_v3.1.2 ***
> Phantom: WaterBox, 1.0cm Vox
> Singles: 717897
> Random Coins.: 111
> True Unscattered Coins.: 2497
> True Scattered Coins.: 5355
> Total: 7963
> Events: 9998927
>
> Some time ago we have also noticed 2 more problems, this time with 
> Gate_v3.1.1:
>
> 1) We have prepared simulation files (again we were interested in the 
> number of singles and coincidences produced). Running the simulation 
> gave us a set of results (let's call them 'Results1'). Then, changing 
> ONLY the line:
> /control/verbose 0
> to:
> /control/verbose 2
> Gave us another set of results (let's call them 'Results2'). It was 
> quite interesting to find that Results1 had nothing in common with 
> Results2. We have retested this many times and everytime the 'anomaly' 
> was observed. What has the verbosity to do with the number of singles 
> and coincidences produced by one and the same simulation?
>
> 2) This problem is quite similar to the one mentioned above. This time 
> the ONLY thing we've changed in the mac files, was the name of the 
> ROOT output file. Again the two sets of results we obtained had 
> nothing in common. This was also retested many times. Investigating 
> the 'anomaly' we found out that the set of results we obtain each time 
> depends on the length of the ROOT output filename.
>
> All of the problems above where observed on two different machines.
>
> Has anyone noticed problems, similar to those above? Is there a 
> version of Gate that is known to be stable?
> Any suggestions and support will be appreciated.
>
> Georgi Gerganov & Yordan Madzhunkov
> Bulgaria, Sofia University
>
>------------------------------------------------------------------------
>
>_______________________________________________
>Gate-devel mailing list
>Gate-devel at lists.healthgrid.org
>http://lists.healthgrid.org/mailman/listinfo/gate-devel
>  
>