Memorias de investigación
Other publications:
Restating Parkinson's disease severity indices by means of non-motor criteria

Research Areas
  • Artificial intelligence,
  • Insanity senile (alzheimer, parkinson)

Is it possible to set the motor staging of a Parkinson?s disease (PD) patient just in terms of just the cognitive symptoms? This is the kickoff question for an in-depth study of the comparison between motor and non-motor indices throughout this report. The most classical severity index for PD patients is the Hoehn & Yahr (H-Y) severity index, whose values range from 1 (mild) to 5 (severest). In the first part of this report, we explore the possible translations of this index using other non-motor indices, like SCOPA-Cognitive scale, non-motor symptoms index, psychiatric complications or autonomous capacities. These translations are computed as an optimization process that could in most cases be tackled using exhaustive search. Not only are complete indices compared, but also individual items from the non-motor indices are selected to create subscales that are fitter for the translation. The translation accuracy of these results varies from 45% to 62%. Another recently proposed severity index is the clinical impression of severity index, or CISI-PD. A similar translation analysis is tackled in the second part of this report. Since this is a continous index, two different binning policies are proposed to categorize its values: one suggested in the original work that proposed the index and another one identified here again by solving an optimization problem. Results using this last encoding show the different comparisons between the CISI-PD index and two non-motor indices to be more accurate. Individual items are also selected using a powerful optimization algorithm: an estimation of distribution algorithm or EDA. Lastly, the third part of the report includes a fully supervised classification task coupled with a feature subset selection process. We jointly configure a data matrix using all individual non-motor items from the different severity indices. The supervised class variable is therefore either the H-Y or the CISI-PD severity score. We use a wrapper item selection to identify subsets of highly relevant items capable of predicting up to 92% of the cases in the dataset. This way, some of the non-motor items subsets are clearly related to specific motor index scores, suggesting a link between motor and non-motor symptoms.
Universidad Politécnica de Madrid
Publication type
Informe Técnico

Research Group, Departaments and Institutes related
  • Creador: Departamento: Inteligencia Artificial