Recognition Error Function
Recognition Error Function
Hi, Can someone direct me to documentation that will help me interpret the output of the Recognition Error Function? For example, how do I interpret the values reported for the False Positive indices? Or, is there a range of Global Mass Error values that provides confidence that sulci have been correctly labeled. Thanks, Mark
- riviere
- Site Admin
- Posts: 1361
- Joined: Tue Jan 06, 2004 12:21 pm
- Location: CEA NeuroSpin, Saint Aubin, France
- Contact:
Re: Recognition Error Function
Hi Mark,
Sorry, I quite forgot your post and have not answered...
The process you are referring to compares two labelings for a given (identical, it must have exactly the same structure) sulcal graph, one is supposed to be a "manual" labeling taken as grouns truth, and an automatic labeling.
The reported errors are percentages of the size of a sulcus (voxels from the sulcal skeleton) which match between the two loabelings. False positives are, for a given sulcus of label "l", nodes that have been labelled "l" but that are not labelled "l" in the manual labeling. False negative are nodes that are labelled "l" in the manual labeling but are not in the automatic one. The "gloabl mass error" is the mean of these figures on all sulci, weighted by the size of sulci, which, in short, captures only false positive errors. The "SI error" is a global similarity index which takes both error types into account. You can find their exact definition in:
M. Perrot, D. Rivière, A. Tucholka, and J.-F. Mangin. Joint Bayesian Cortical Sulci Recognition and Spatial Normalization. In Proc. 21th IPMI, LNCS-5636, Williamsburg, VA, pages 176-187, July 2009. Springer Verlag.
Denis
Sorry, I quite forgot your post and have not answered...
The process you are referring to compares two labelings for a given (identical, it must have exactly the same structure) sulcal graph, one is supposed to be a "manual" labeling taken as grouns truth, and an automatic labeling.
The reported errors are percentages of the size of a sulcus (voxels from the sulcal skeleton) which match between the two loabelings. False positives are, for a given sulcus of label "l", nodes that have been labelled "l" but that are not labelled "l" in the manual labeling. False negative are nodes that are labelled "l" in the manual labeling but are not in the automatic one. The "gloabl mass error" is the mean of these figures on all sulci, weighted by the size of sulci, which, in short, captures only false positive errors. The "SI error" is a global similarity index which takes both error types into account. You can find their exact definition in:
M. Perrot, D. Rivière, A. Tucholka, and J.-F. Mangin. Joint Bayesian Cortical Sulci Recognition and Spatial Normalization. In Proc. 21th IPMI, LNCS-5636, Williamsburg, VA, pages 176-187, July 2009. Springer Verlag.
Denis