Hereditary Spastic Paraplegia (HSP) is a group of neurodegenerative disorders characterized by spactiicty and weakness of the legs.

HSP can be broadly divided into pure and complicated forms. While in both subtypesthe disease primarily impacts upper motor neurons resulting in spasticity of the legs and progressive lower limb weakness, in the pure form of HSP spasticity is usually isolated, whereas in the complicated form such as SPG11, other clinical conditions such as ataxia, intellectual disability, or hearing and vision abnormalities are also present. In the brain of SPG11 patients, atrophy of specific white matter pathways can be expected.

Contribution:

1. Image analysis and processing of T1-weighted, T2-weighted, and DTI data of HSP patients.

2. Investigating differences in brain morphology between patients with pure hereditary spastic paraplegia (pHSP), complicated HSP of type 11 (SPG11), and a group of healthy controls.

3. Development of a new approach to tensor based morphometry analysis (TBM). We used the full diffusion tensor information to drive the registration of the brain images of individual patients into a normative template for the population, to take advantage of more specific tissue characterization provided by DTI and analyze the deformation fields obtained from this registration (we refer to this method as D-TBM). This is in contrast to most previous TBM studies where the deformation fields are obtained from registration of T1-weighted images. D-TBM is a powerful approach to detect volumetric changes in white matter pathways with high sensitivity compared to existing methods.

4. Extensive evaluation of D-TBM compared to the classical TBM.

5. Application of random forests algorithm, a machine learning approach to classification of pure-HSP, SPG11, and controls.

Findings: For SPG11 patients, D-TBM analysis revealed selective atrophy of several long-range association and projection pathways, including the arcuate fasciculus, the inferior fronto-occipital fasciculus, the cingulum bundle, the corticospinal tract, and the corpus callosum that are not as evident using conventional methods. For pure-HSP where the differences are not as evident as SPG11, a machine learning approach identified lower fractional anisotropy and smaller size of corticospinal tract as key features in determining whether the subject belonged to the HSP class or control. D-TBM can be used for individual subject assessment and also longitudinal individual subject measures such as natural history studies. Using DTI metrics and features derived from D-TBM and a machine learning algorithm we were able to correctly classify all the controls subjects (²⁴⁄₂₄), all SPG11 (⁵⁄₅), and 12 of 14 Pure-HSP, achieving high sensitivity and specificity.

Papers/abstracts: Paper under preparation to be submitted to Human Brain Mapping (HBM). Abstract presented at the Eastern North American Region (ENAR) International Biometric Society 2016 Spring Meeting and ISMRM 2016.