sulci recognition error

Questions about BrainVisa usage and installation

Moderators: riviere, denghien

csaiote
Posts: 4
Joined: Tue Dec 12, 2017 11:24 pm

sulci recognition error

Postby csaiote » Wed Jul 31, 2019 9:33 pm

Hi,

We’re trying to calculate sulci recognition errors with Morphologist 2015 and have some questions:

We initially ran the pipeline with the default options (SPAM, 2008 models, 3.1 graphs) and then ran the Recognition Error function from the expert options. Our results from this function seem strange after reading other posts on the forum. The results provide only false positive values for each sulcus, with 0 for size, False negative, True positive, and SI error=1 (example attached). The exception is the “unknown” sulcus label returning a value of 35000 for sulci size, and values for false negative, true positive, and SI error. Can this actually use SPAM results as we did?
We’re wondering whether we’re interpreting this wrong or if we are running the analysis incorrectly, but we tried also running the old ANN recognition and got similar results.

Also, how can we interpret the “auto_proba” .csv file for SPAM recognition, particularly the first column, of nodes?

Ultimately, we would like to get results analogous to what was done in the Regis et al. 2011 epilepsy paper.

Many thanks in advance,
Attachments
Recognition_Error.txt
(6.74 KiB) Downloaded 64 times

User avatar
riviere
Site Admin
Posts: 1358
Joined: Tue Jan 06, 2004 12:21 pm
Location: CEA NeuroSpin, Saint Aubin, France
Contact:

Re: sulci recognition error

Postby riviere » Wed Aug 07, 2019 10:21 am

Hi,
This old process was designed for the oldest models (ANN), and thus is using the ANN model file (not used in SPAM method) to provide a recognition energy which will not be useful to you. Anyway I think the problem is in the kind of labels the process is using: labels are stored in attributes in sulci graphs, and there are 2 possible attributes for labels (a historical choice which was meant to allow several labeling to coexist in order to save memory and disk space, but which actually made things confusing to users - one of the many historical wrong choices...). So "manual" labelings were stored in the attribute "name" in graphs, and automatic labelings were stored in the "label" attribute. Thus processes comparing labeling have to know, or be told, which attribute should be used for sulci labels. Here the process was designed to compare a manual labeling (using "name") and an automatic one (using "label"). I think you are comparing two automatic labelings, right ?
The simplest way to get around this problem right now is to transform one of the labelings into a "manual" one, using the process "Sulci Switch Manual Labels" before using the "Recognition Error" one. But I agree the process could / should be adapted.

Now the auto_proba.csv is an output of the SPAM recognition process. It provides for each vertex of the sulci graph its probability to get each label in the SPAM model, and summerizes the proba and lilkelyhood of the most probable label. The details should be found in Perrot 2002 paper.

Denis

csaiote
Posts: 4
Joined: Tue Dec 12, 2017 11:24 pm

Re: sulci recognition error

Postby csaiote » Thu Aug 22, 2019 5:30 pm

Thank you for your reply! We were able to run the "Recognition Error" after transforming the one of the labeling.

We have a follow-up question: using the ANN method, we would like to get the local energy values corresponding to each sulcus that show the similarity to the database. I see there is an energy_plot_file .nrj output but can't quite follow how it is organized. Is this what we are looking for?

Thank you again,
Catarina

User avatar
riviere
Site Admin
Posts: 1358
Joined: Tue Jan 06, 2004 12:21 pm
Location: CEA NeuroSpin, Saint Aubin, France
Contact:

Re: sulci recognition error

Postby riviere » Mon Sep 09, 2019 2:59 pm

Hi,
The energy_plot_file.nrj is a record of the global (whole hemisphere) energy across the optimization, it does not contain the local energies.
I don't remember any process providing this locally. It could be done of course because they are used during the identification process, but it's not ready to use right now.
Can you program a little bit of Python ? If yes I could probably provide some clues.
Denis


Return to “BrainVISA”

Who is online

Users browsing this forum: No registered users and 3 guests