Learning temporal statistics for sensory predictions in mild cognitive impairment

Abstract

Training is known to improve performance in a variety of perceptual and cognitive skills. However, there is accumulating evidence that mere exposure (i.e. without supervised training) to regularities (i.e. patterns that co-occur in the environment) facilitates our ability to learn contingencies that allow us to interpret the current scene and make predictions about future events. Recent neuroimaging studies have implicated fronto-striatal and medial temporal lobe brain regions in the learning of spatial and temporal statistics. Here, we ask whether patients with mild cognitive impairment due to Alzheimer's disease (MCI-AD) that are characterized by hippocampal dysfunction are able to learn temporal regularities and predict upcoming events. We tested the ability of MCI-AD patients and age-matched controls to predict the orientation of a test stimulus following exposure to sequences of leftwards or rightwards orientated gratings. Our results demonstrate that exposure to temporal sequences without feedback facilitates the ability to predict an upcoming stimulus in both MCI-AD patients and controls. However, our fMRI results demonstrate that MCI-AD patients recruit an alternate circuit to hippocampus to succeed in learning of predictive structures. In particular, we observed stronger learning-dependent activations for structured sequences in frontal, subcortical and cerebellar regions for patients compared to age-matched controls. Thus, our findings suggest a cortico-striatal-cerebellar network that may mediate the ability for predictive learning despite hippocampal dysfunction in MCI-AD.

Classifying Cognitive Profiles Using Machine Learning with Privileged Information in Mild Cognitive Impairment

“…The cognitive and fMRI data used in this study were collected in the context of two behavioral and fMRI studies (Baker et al, 2015; Luft et al, 2015, 2016) in which the participants were asked to predict the orientation of a test stimulus following exposure to structured sequence of leftwards and rightwards oriented gratings, and no feedback were given. Both studies aimed to (1) test whether training on structured temporal sequences improves the ability to predict upcoming sensory events and (2) identify brain regions that support the ability of using implicit knowledge about the past for predicting future.…”

“…In Luft et al (2016), regions-of-interest (ROIs) were identified by applying whole-brain general linear model analysis with a voxel-wise mixed-design three-way ANOVA, that is,…”

“…Table 1 in Luft et al (2016) listed all brain regions showing significant interaction between session, sequence, and group. For the study presented in this paper, we combined two ROIs in the frontal region (Superior Frontal Gyrus, SFG, on the right hemisphere and Medial Frontal Gyrus, MFG, on the left hemisphere) and two ROIs in the cerebellar region (Cerebellar Lingual and Cullmen ROIs in both hemispheres).…”

“…Among these participants, nine MCI patients and nine healthy controls come from the cohort reported in Luft et al (2015). The remaining 16 healthy controls come from the cohort reported in Luft et al (2016). The size of that cohort is 20.…”

“…The remaining 17 participants are from unpublished studies but they participated exactly the same experiments as other participants. Note that all neuroimaging data used in this study are reported either in Luft et al (2015) or in Luft et al (2016). …”