Assessing memory with mutual information and metric content estimates in ageing and dementia

Information theory has been successfully applied to the problem of decoding neurophysiologic data obtained from single cells recordings. Here we present an adaptation of mutual information measures to behavioural data in multiple choice classification tasks. We introduce the metric content index, which allows for a quantification of the tendency to perceive the stimuli as belonging to classes composed of homogeneous items. We show that this index allows deriving new insight in the interpretation of the possible patterns of errors produced by the subjects.We apply the method to the quantitative evaluation of knowledge about famous people in ageing and dementia. We find a significant increase of the mutual information estimate and of the metric content index in normal ageing. In opposition to this result, we show that in a sample of patients suffering from probable Alzheimer\'s Dementia, a further increase in the metric content index is associated with a decrement of the mutual information measure and of the overall performance level in the same classification task. Results are discussed in the context of the episodic vs semantic memory distinction.