Predicting MMSE Score from Finger-Tapping Measurement
This work addresses early dementia diagnosis for the elderly, but it is incremental as it applies existing methods to a new data type.
The paper tackles dementia diagnosis by predicting MMSE scores from finger-tapping measurements using a machine learning pipeline, achieving good prediction performance and identifying specific finger-tapping attributes as biomarkers.
Dementia is a leading cause of diseases for the elderly. Early diagnosis is very important for the elderly living with dementias. In this paper, we propose a method for dementia diagnosis by predicting MMSE score from finger-tapping measurement with machine learning pipeline. Based on measurement of finger tapping movement, the pipeline is first to select finger-tapping attributes with copula entropy and then to predict MMSE score from the selected attributes with predictive models. Experiments on real world data show that the predictive models such developed present good prediction performance. As a byproduct, the associations between certain finger-tapping attributes ('Number of taps', 'Average of intervals', and 'Frequency of taps' of both hands of bimanual in-phase task) and MMSE score are discovered with copula entropy, which may be interpreted as the biological relationship between cognitive ability and motor ability and therefore makes the predictive models explainable. The selected finger-tapping attributes can be considered as dementia biomarkers.