Paola Scifo, MR scientist in Scientific Institute H San Raffaele, Milan, Italy
Group objectives:
Our group aims at developing a complementary alternative to the
coordinate-based point of view underlying most of the brain mapping approaches.
This alternative is based on structural strategies, where the word "structural"
stands for brain models and data representations made up of a set of
entities linked by various relations. While this structural point of view
was deeply embedded in Neurosciences before the advent of neuroimaging
(architectonic parcellations of the cortical surface, structural models
of macaque visual system, etc.), the warping-based spatial normalization
paradigm leads to represent brain mapping results as fuzzy images.
This "fuzzyness", which stems from the voxel-based averaging process performed to compare subjects,
may result in some loss of information relatively to the brain organization.
Furthermore, testing the putative structural models proposed by cognitive neuroscientists
using probabilistic maps is sometimes confusing.
We think that converting the individual data into structural representations
before performing the group analysis process may overcome some of the problems.
Our alternative structural framework for brain mapping relies on a few
generic ideas:
Convert raw images into structural representations (cortical folds and their neighborhood relationships, pieces of fiber bundles, activation clusters, etc...);
Merge several representations coming from the same individual;
Match such structural representations with syntactically corresponding structural models (sulci, known fiber bundles, cognitive areas, etc...). These models come from a database of the current knowledge, which could include concurrent points of view;
Infer new similarities across a set of individuals to improve the current structural models.
Here is an illustration of a possible resulting scheme:
The goal underlying this framework is the design of Artificial Intelligence methods performing the structural model
inference supposed to stem from brain imaging experiments. Such methods
would rely on a specific kind of Computer Vision dedicated to volumetric images
of the brain. They could greatly support human neuroscientists who have unfortunately not
been endowed with a captor for volumetric images.
