Rehabilitation of TBI is an exceedingly important public health goal not only because neurotrauma-related activity limitations can have significant impact upon life roles, but also because it affects interpersonal communication, as well as social participation in personal activities of daily living. In this context, our results and methodology hold implications for the systematic mapping of human neural impairment caused by this condition. Firstly, this work is of potential clinical relevance to the study of neural atrophy changes. Aside from identifying and describing connectomic patient profiles, the presented method can be used to generate suggestions for informing and guiding clinical interventions designed to ameliorate recuperation. Rapidly visualizing the longitudinal evolution of individual TBI cases using our diagrammatic tools can reveal how deficit patterns are influenced by lesion site, by relative sparing and redundancy within the distributed cortical system under scrutiny, as well as by the neural plastic changes that can occur with recovery.
Because of the high level of neuroanatomic information presented regarding the structural atrophy and reorganization of the brain in these depictions, they can be exploited to quickly delineate the function of specific WM fibers or cortical (sub-)regions. Thus, in addition to its relevance to the clinical field, our approach has potential applications to the formulation, validation, or information of basic science theories concerning perceptual learning and neural plasticity. It could also complement and extend information already gained from previous animal and human lesion studies. The detailed level of structural impairment description afforded by our technique can be combined with the task selection mechanism of our paradigm to construct more effective patient interventions. Such strategies can be used to explain occupational performance difficulties as well as to shed light upon existing or emerging compensatory rehabilitation techniques.
One possible caveat related to our method is that the presence of hemorrhage or certain other forms of pathology can modify local water diffusivity measures to an extent that may possibly obscure the presence of WM tracts in affected regions, as extracted using DTI tractography. However, it is not foreseeable that this drawback can influence our computed atrophy rates; this is because tissue permeation by blood in the acute TBI phase would be associated with decreased FA compared to the chronic phase. Consequently, acute hemorrhage may result in inappropriate (over-)estimation of WM recovery (due to increases in FA between the two time points), but not atrophy (which is related to decreases in FA). Thus, because our method focuses on the calculation of atrophy rates, the presence of hemorrhage is not expected to influence or results substantially. To state this in another way, the presence of hemorrhage results in decreased FA for some voxels at baseline. This, in turn, results in decreased fiber counts in hemorrhagic regions at acute baseline compared to the chronic stage (when hemorrhage is no longer present). Consequently, it is reasonable to expect the hemorrhage artifact-related percentage change in normalized fiber counts from acute baseline to chronic follow-up to be positive. However, because our technique is concerned with atrophy (i.e., negative percentage changes in normalized fiber counts), the effect of hemorrhage is unlikely to influence our results substantially.
Most methods for connectivity visualization rely on variations of graph theory to position network nodes, represent nodes using shapes, and to modulate edge properties (weight, color, thickness, etc.) according to metrics of connectivity. One popular method for node representation (Bassett et al., 2011
; Gerhard et al., 2011
; Yan et al., 2011
) involves positioning nodes at the 3D locations of brain regions. When viewed in three dimensions, this method allows one to associate network nodes with neuroanatomic landmarks. However, one disadvantage of this representation is that it can have so many nodes and edges that the relationships between them cannot be discerned (Sanz-Arigita et al., 2010
). By contrast, the approach proposed here orders nodes along the antero-posterior axis, thereby making all nodes equally visible to the viewer. Although 2D visualization of brain networks is also possible if the cortical surface is spread out as a 2D sheet (Palva et al., 2010
; Gerhard et al., 2011
), it remains somewhat difficult to associate locations on 2D cortical maps to 3D brain topology.
One difficulty involved in visualizing connectomes is the depiction of nodes using symbols representative of their properties. In Dosenbach et al. (2007
), Spoormaker et al. (2010
), Chen et al. (2011
), Ginestet and Simmons (2011
), and elsewhere, colored circles or spheres are used to this end. Although this offers the ability to differentiate groups of edges according to their properties, nodes, and edges can occasionally overlap. This can be partially addressed by changing node transparency (Ginestet and Simmons, 2011
), although this may not always be helpful for very dense networks. In our approach, by contrast, all nodes are visible and node properties are represented using concentric color-coded rings. Another difficulty refers to how best to encode network edge properties according to the degree of connectivity between nodes. Edges can be modulated by color (Cao and Slobounov, 2010
; van den Heuvel and Pol, 2010
) or thickness (He and Evans, 2010
; He et al., 2010
), and the top-down subdivision of the nervous system can be represented by concentric circles, e.g., brain to cortex to lobes to gyral/sulcal structures, as in (Holten, 2006
; Modha and Singh, 2010
). In the latter case, edges radiate from the center of a circle (the top structure) to the circumference (the lowest-level structures) and the circular shape of the connectivity graph is appealing and useful for 2D representation of relationships. One challenge that can be addressed by representing connectomes as circular diagrams is that of complexity and dimensionality reduction. In the hierarchical approach (Holten, 2006
; Modha and Singh, 2010
), this is done via top-down organization of connections; in ours, it can be achieved by displaying connections efferent from one hemisphere, one lobe, or from a single brain structure, thus allowing one to change the complexity of the connectogram.
It is important to emphasize that our workflow allows the user to use connectivity matrices that are generated using segmentation and connectivity matrix calculation methods other than our own. This feature may be preferable to investigators who wish to use some other standard for extracting connectivity information from brain imaging data. Naturally, the extent to which the connectogram functionality being proposed can be useful to investigators depends on how accurate the information contained in the connectivity matrix is. Consequently, although our own method for generating connectivity matrices has been validated, it can be expected that the results of our graphical workflow are only as accurate as the connectivity matrix itself. Unless our own method for generating connectivity matrices is being used, this is a factor beyond our control, which nevertheless leaves undiminished the potential use and novelty of our contribution.