Introduction
Tensor-Based Morphometry (TBM) is a cornerstone of modern neuroimaging, enabling researchers to map structural differences between populations or track changes over time. Traditionally, TBM relies on the Jacobian determinant—a scalar value representing local volume expansion or contraction. While effective for isotropic tissues, this approach has a significant limitation when applied to the brain's white matter: it discards directional information.
In fibrous tissues, structural changes are often anisotropic. For instance, a nerve fiber bundle might lengthen while simultaneously thinning, resulting in zero net volume change. Standard TBM would report "no change" in such a scenario, missing critical pathological or developmental markers. While "Generalized TBM" can analyze the full deformation matrix, its results are often difficult for clinicians to interpret. This paper introduces a framework that decomposes local volume changes into intuitive physical descriptors: changes in fiber length versus changes in bundle thickness.
Methodology
The proposed method "unlocks" the full Jacobian matrix by leveraging the brain's inherent architecture as a guide. The process involves three key steps:
- DTI Integration: Diffusion Tensor Imaging (DTI) is used to identify the local orientation of fiber bundles ($e_1$).
- Local Reference Frames: Instead of analyzing the Jacobian matrix in standard Cartesian (XYZ) coordinates, the analysis is rotated into a local reference frame aligned with the identified fiber orientation.
- Decomposition into Intuitive Descriptors: After factoring out the rotation required to align a subject to a template (using the "Preservation of Principal Direction" or PPD strategy), the remaining deformation is represented as an upper-triangular matrix. This allows the total volume change ($det(J)$) to be split into two physically meaningful components:
- $s_1$: Expansion or compression along the fiber (change in length).
- $s_{23}$: Expansion or shrinkage perpendicular to the fiber (change in cross-sectional area or "thickness").
Key Insights: Anatomy-Aware Mathematics
A central insight of this work is that the Gram-Schmidt orthonormalization used in DTI reorientation is mathematically equivalent to a QR-decomposition of the Jacobian. This provides a rigorous foundation for separating volume changes into "along" and "across" components.
By moving beyond simple volume metrics, the method demonstrates that "no volume change" does not equate to "no structural change." By separating length from thickness, researchers can detect compensatory changes that the Jacobian determinant typically masks. This provides a language for pathology—such as distinguishing between axonal loss and myelin degradation—that is far more accessible to clinicians than abstract deformation tensors.
Experimental Results
The authors validated their method using the IXI brain database, specifically examining gender differences in a cohort of 35 subjects aged 65 and older. High-dimensional tensor-based registration was used to create a population-specific DTI template.
The results highlighted two major advantages of the new framework:
- Explanatory Power: In regions where standard TBM found differences, the new method explained the underlying cause. For example, specific clusters were revealed to be driven primarily by changes in fiber length ($s_1$).
- Increased Sensitivity: Crucially, the proposed method identified new clusters of significant difference that were invisible to standard TBM. These regions likely underwent anisotropic changes (e.g., stretching in one direction while shrinking in another) that left the total volume unchanged, thus evading detection by the Jacobian determinant.
Conclusion and Future Work
By grounding morphometry in the physical reality of fibrous tissues, this framework offers a more sensitive and interpretable tool for neuroimaging. While the current study focused on the primary fiber direction ($e_1$) due to noise in secondary directions, future work may refine these measurements. Additionally, the authors suggest this framework could be applied to other fibrous structures, such as cardiac laminar structures, to explore how fiber bundles bend, twist, or thin during disease progression.
Summary updated