Nonlinear discriminant function analysis differentiates neurodegenerative changes in presymptomatic Huntington's disease: Cross sectional data from Image-HD.

Abstract

Introduction: We investigated two measures of neural integrity, T1 weighted volumetric measures and diffusion tensor imaging (DTI), exploring their combined potential to differentiate pre-diagnosis HD (pre-HD) from healthy controls. Method(s): 34 symp-HD, 32 pre-HD and 35 controls participated in IMAGE-HD. Participants underwent DTI and T1 weighted imaging using a Siemens 3 Tesla scanner. Volumetric and DTI measures [mean diffusivity (MD) and fractional anisotropy (FA)] were calculated. Quadratic discriminant function analysis (QDFA) was performed to differentiate pre- HD from controls, while considering volumetric, MD and FA measures in various regions of interest. We adopted a stepwise approach, considering each measure separately from within the caudate, pallidum and putamen, followed by simultaneous consideration of each. Finally, we tested for improvements to this model, by including: (1) extra regions (thalamus and accumbens), (2) motor scores, and (3) motor and neurocognitive scores. Cross-validation analyses were conducted. Result(s): All three groups differed (p < 0.01) in volume, MD and FA within the caudate and putamen. QDFA demonstrated that the best differentiator of pre-HD from controls is the model incorporating both volumetric and diffusion data from basal ganglia regions and thalamus (95.5%), compared with models that present each measure separately. During cross-validation, the combined model with just basal ganglia regions, was 73.1% accurate. Inclusion of motor, or motor and neurocognitive scores did not improve discriminatory accuracy. Conclusion(s): Our QDFA findings provide evidence that combined multi-modal imaging measures may be a sensitive means of developing a neuroimaging based biomarker that can accurately classify pre-HD from controls early in disease trajectory.