Sulci recognition for brain tumour patients

[MSh]

Hi

I am looking at the sulci identification with Brainvisa for brain tumour patients and have included the tumour mask in the brain segmentation step. The program has also found and labelled some folds in the tumour (shown in figure). Is there any way to improve the result?

Also, is there any way to check for recognition error (not compared to manual labelling but according to the model for example)?

Many thanks
MSh

[riviere]

Hi,
That's right, the tumor mask is taken into account for brain segmentation, but not for the cortical folds extraction and identification. This is something that we could perhaps improve - if the tumor mask is excluded (or considered as WM) when extracting the folds, it should just remove all sulci in the tumor zone.

Also, is there any way to check for recognition error (not compared to manual labelling but according to the model for example)?

I don't completely understand what you mean: if you don't have a manual labeling, it is not possible to compare the automatic labeling to a "ground truth" and thus detect errors. Or maybe you mean, get some local confidence measurements in the models output or posterior probablilities ? But it will not be a definitive error measurement. It is more or less possible using the ANN-based models, which provide all models potentials which can be considered as confidence measurements. Using the SPAM-based models, we have some posterior probabilities, but I think we have not made a tool to easily access them yet: you just have an output file with the posterior probabilities of each graph node.

Denis

[MSh]

Hi Denis

Thanks for the reply. What would you suggest then? Do I exclude the tumour mask from the brain? (does this not remove some genuine sulci?) Would you think this will be sorted out in the next release of the software?

Thanks
MSh