Remapping the cognitive and neural profiles of children who struggle at school

Abstract

Our understanding of learning difficulties largely comes from children with specific diagnoses or individuals selected from community/clinical samples according to strict inclusion criteria. Applying strict exclusionary criteria overemphasizes within group homogeneity and between group differences, and fails to capture comorbidity. Here, we identify cognitive profiles in a large heterogeneous sample of struggling learners, using unsupervised machine learning in the form of an artificial neural network. Children were referred to the Centre for Attention Learning and Memory (CALM) by health and education professionals, irrespective of diagnosis or comorbidity, for problems in attention, memory, language, or poor school progress (n = 530). Children completed a battery of cognitive and learning assessments, underwent a structural MRI scan, and their parents completed behavior questionnaires. Within the network we could identify four groups of children: (a) children with broad cognitive difficulties, and severe reading, spelling and maths problems; (b) children with age-typical cognitive abilities and learning profiles; (c) children with working memory problems; and (d) children with phonological difficulties. Despite their contrasting cognitive profiles, the learning profiles for the latter two groups did not differ: both were around 1 SD below age-expected levels on all learning measures. Importantly a child's cognitive profile was not predicted by diagnosis or referral reason. We also constructed whole-brain structural connectomes for children from these four groupings (n = 184), alongside an additional group of typically developing children (n = 36), and identified distinct patterns of brain organization for each group. This study represents a novel move toward identifying data-driven neurocognitive dimensions underlying learning-related difficulties in a representative sample of poor learners.

Keywords: cognitive development; education; learning difficulties; machine learning.

Figure 1

CONSORT diagram showing recruitment avenues and exclusions

Figure 2

Overview of processing steps to reconstruct a white matter connectome from diffusion-weighted and T1-weighted MRI data

Figure 3

Weight distributions from the self-organizing map, split by task. For each task the map depicts high weights (i.e., good performance) as yellow squares and low weights (i.e., poor performance) as black squares. The Pearson correlation between the weight distributions can be seen in the bottom-right matrix

Figure 4

The distributions of children’s best matching unit (BMU) within the map. This is first shown for all children and then for children categorized by referral reason and diagnosis. Beneath each plot the statistic indicates whether the BMUs are evenly scattered or grouped

Figure 5

The top panel shows the distributions of children assigned to each of the four clusters. Beneath each map the statistic indicates that all four clusters occupy a nonrandom set of nodes within the map. Beneath the maps the cognitive profile of each cluster is shown, ordered by cluster number. The scale indicates performance as a z score relative to age expected levels. The dots indicate individual children with the shade indicating the child’s consistency within that cluster over the 1,000 iterations—the darker the shade the more consistent the child

Figure 6

Regions with consistent significant differences in node degree between Cluster 1 and the control groups (blue) and Cluster 4 and the control groups (red)