For many neurodegenerative diseases, e.g. Parkinson’s disease, at the point of clinical manifestation already a substantial fraction of the relevant neurons have degenerated and mostcompensatory resources are already exhausted. The same is likely true also for cerebellar functioning in degenerative spinocerebellar ataxia (SCA).  The effectiveness of future interventions is thus critically dependent on the detection of early signs of the diseases, where subtle motor deficits can be an important predictor. In addition, a more detailed understanding of the early dysfunction might be important to develop optimized strategies in terms of rehabilitation training, in order to delay the manifestation of strong impairments, e.g. in cerebellar ataxia. In order to detect and quantify subtle signs of motor deficits we exploit methods from machine learning, combined with specific tasks that have a potential to reveal subtle motor deficits.

In recent work on spino-cerebellar ataxia (SCA) with Dr. M. Synofzik and Prof. L. Schöls  (HIH, Dept. for Neurodegeneration) we use a homogeneous cohort, containing only one genetically-characterized SCA subtype. By increasing the cognitive load exploiting dual-task paradigms, we try to detect subtle, otherwise concealed ataxia dysfunctions. Based on a number of predictors that we compute from a quantitative movements analysis using motion capture, we construct multi-variate models for the classification of preclinical movements patterns. Exploiting methods from machine learning, we can detect critically relevant predictors and can combine multiple predictors in an optimal way. Since cerebellar ataxias with homogenous subtype are rare in single European centres, we exploited a new system for movement analysis to study a large, very homogenous population of pre-symptomatic SCA participants in Holguin in Cuba (cooperation with Prof. L. Luis Velasquez-Perez ), where the prevalence of this disease form is increased.

In previous work we applied a similar approach for the detection of early movement symptoms in Parkinson’s disease (PD). In collaboration with Prof. D. Berg and Dr. I. Liepelt (HIH, Dept. for Neurodegeneration) patients and clinically healthy participants from a risk population were  recruited that was characterized by and  increase echogenicity of the substantia nigra, another pre-clinical marker for Parkinson’s disease. The participants were tested with a number of motor tests that seemed promising for the detection of preclinical signs of degeneration. Applying methods from machine learning (Support Vector Machines combined with recursive  feature elimination) we were able to construct a predictor from these tests that highly correlates with other pre-clinical markers, and which predicted correctly  several participant from the cohort at risk that developed clinically manifest PD in the following years. This result shows that subtle movement signs are a helpful preclinical marker for PD and the resulting set of critical features define a novel battery of critical tests for the assessment of such preclinical motor signs.