Anomalous gray matter patterns in specific reading comprehension deficit are independent of dyslexia

Sampling inequalities affect generalization of neuroimaging-based diagnostic classifiers in psychiatry

BACKGROUND

The development of machine learning models for aiding in the diagnosis of mental disorder is recognized as a significant breakthrough in the field of psychiatry. However, clinical practice of such models remains a challenge, with poor generalizability being a major limitation.

METHODS

Here, we conducted a pre-registered meta-research assessment on neuroimaging-based models in the psychiatric literature, quantitatively examining global and regional sampling issues over recent decades, from a view that has been relatively underexplored. A total of 476 studies (n = 118,137) were included in the current assessment. Based on these findings, we built a comprehensive 5-star rating system to quantitatively evaluate the quality of existing machine learning models for psychiatric diagnoses.

RESULTS

A global sampling inequality in these models was revealed quantitatively (sampling Gini coefficient (G) = 0.81, p < .01), varying across different countries (regions) (e.g., China, G = 0.47; the USA, G = 0.58; Germany, G = 0.78; the UK, G = 0.87). Furthermore, the severity of this sampling inequality was significantly predicted by national economic levels (β =  - 2.75, p < .001, R²_adj = 0.40; r =  - .84, 95% CI: - .41 to - .97), and was plausibly predictable for model performance, with higher sampling inequality for reporting higher classification accuracy. Further analyses showed that lack of independent testing (84.24% of models, 95% CI: 81.0-87.5%), improper cross-validation (51.68% of models, 95% CI: 47.2-56.2%), and poor technical transparency (87.8% of models, 95% CI: 84.9-90.8%)/availability (80.88% of models, 95% CI: 77.3-84.4%) are prevailing in current diagnostic classifiers despite improvements over time. Relating to these observations, model performances were found decreased in studies with independent cross-country sampling validations (all p < .001, BF₁₀ > 15). In light of this, we proposed a purpose-built quantitative assessment checklist, which demonstrated that the overall ratings of these models increased by publication year but were negatively associated with model performance.

CONCLUSIONS

Together, improving sampling economic equality and hence the quality of machine learning models may be a crucial facet to plausibly translating neuroimaging-based diagnostic classifiers into clinical practice.

Distinct neural correlates of poor decoding and poor comprehension in children with reading disability

Reading disability (RD) can manifest itself as a word decoding problem or a reading comprehension problem. In the current study, we identified 3 subtypes of RD: poor decoders (PD), poor comprehenders (PC), and poor-in-both (PB). We found that PD had greater deficits in meta-linguistic skills such as phonological awareness, orthographic skills, and morphological skills than PC, whereas PC had greater deficits in listening comprehension than PD. In the brain, we also found different patterns of deficits during an auditory rhyming judgment task using functional magnetic resonance imaging. PD showed less activation than PC and age controls in the left dorsal inferior frontal gyrus (IFG) and pre-supplementary motor area (SMA), brain activation of which was correlated with phonological awareness and working memory. In contrast, PC showed less activation in the left fusiform gyrus than PD and age controls, which was correlated with reading comprehension fluency and morphological skill. Last, PB showed both PD's and PC's deficits, as well as additional deficits in the bilateral lingual gyri. Our findings contribute to revealing different neural signatures of poor decoding and poor comprehension, which are distinct disorders but co-occur very often. These findings implicate possibility and necessity of precise diagnosis and individualized intervention.

Evaluation of Risk of Bias in Neuroimaging-Based Artificial Intelligence Models for Psychiatric Diagnosis: A Systematic Review

IMPORTANCE

Neuroimaging-based artificial intelligence (AI) diagnostic models have proliferated in psychiatry. However, their clinical applicability and reporting quality (ie, feasibility) for clinical practice have not been systematically evaluated.

OBJECTIVE

To systematically assess the risk of bias (ROB) and reporting quality of neuroimaging-based AI models for psychiatric diagnosis.

EVIDENCE REVIEW

PubMed was searched for peer-reviewed, full-length articles published between January 1, 1990, and March 16, 2022. Studies aimed at developing or validating neuroimaging-based AI models for clinical diagnosis of psychiatric disorders were included. Reference lists were further searched for suitable original studies. Data extraction followed the CHARMS (Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies) and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines. A closed-loop cross-sequential design was used for quality control. The PROBAST (Prediction Model Risk of Bias Assessment Tool) and modified CLEAR (Checklist for Evaluation of Image-Based Artificial Intelligence Reports) benchmarks were used to systematically evaluate ROB and reporting quality.

FINDINGS

A total of 517 studies presenting 555 AI models were included and evaluated. Of these models, 461 (83.1%; 95% CI, 80.0%-86.2%) were rated as having a high overall ROB based on the PROBAST. The ROB was particular high in the analysis domain, including inadequate sample size (398 of 555 models [71.7%; 95% CI, 68.0%-75.6%]), poor model performance examination (with 100% of models lacking calibration examination), and lack of handling data complexity (550 of 555 models [99.1%; 95% CI, 98.3%-99.9%]). None of the AI models was perceived to be applicable to clinical practices. Overall reporting completeness (ie, number of reported items/number of total items) for the AI models was 61.2% (95% CI, 60.6%-61.8%), and the completeness was poorest for the technical assessment domain with 39.9% (95% CI, 38.8%-41.1%).

CONCLUSIONS AND RELEVANCE

This systematic review found that the clinical applicability and feasibility of neuroimaging-based AI models for psychiatric diagnosis were challenged by a high ROB and poor reporting quality. Particularly in the analysis domain, ROB in AI diagnostic models should be addressed before clinical application.

Literacy-supporting skills in college students with specific reading comprehension deficit and developmental language disorder

Specific reading comprehension deficit (S-RCD) and developmental language disorder (DLD) are both commonly occurring developmental disorders of language. The ways in which these disorders do and do not overlap during childhood are a matter of debate (Nation & Norbury, 2005). Moreover, in both populations, the challenges faced by individuals in adulthood are understudied. Here, we combined data across cohorts of college students, and classified individuals with only S-RCD (n = 20), only DLD (n = 55), and co-occurring S-RCD and DLD (n = 13). Individuals with good language and reading skills, who matched those with S-RCD on decoding, comprised our typical language and reading group (TD; n = 20). Beyond the measures used for classification, group-level differences were identified in sentence-level reading fluency, phonological processing, verbal working memory, and rapid automatized naming. We found that skill profiles differed across groups; however, we found no evidence of weaknesses beyond the core deficit in reading comprehension observed in those with only S-RCD. In contrast, when S-RCD co-occurs with DLD, weaknesses are observed in phonological processing, as well as reading fluency and verbal working memory. These findings suggest that some adults with S-RCD have co-occurring DLD as a core weakness. These findings, as well as differences between individuals with S-RCD and DLD, are further discussed.

The Connectivity Fingerprints of Highly-Skilled and Disordered Reading Persist Across Cognitive Domains

The capacity to produce and understand written language is a uniquely human skill that exists on a continuum, and foundational to other facets of human cognition. Multivariate classifiers based on support vector machines (SVM) have provided much insight into the networks underlying reading skill beyond what traditional univariate methods can tell us. Shallow models like SVM require large amounts of data, and this problem is compounded when functional connections, which increase exponentially with network size, are predictors of interest. Data reduction using independent component analyses (ICA) mitigates this problem, but conventionally assumes linear relationships. Multilayer feedforward networks, in contrast, readily find optimal low-dimensional encodings of complex patterns that include complex nonlinear or conditional relationships. Samples of poor and highly-skilled young readers were selected from two open access data sets using rhyming and mental multiplication tasks, respectively. Functional connectivity was computed for the rhyming task within a functionally-defined reading network and used to train multilayer feedforward classifier models to simultaneously associate functional connectivity patterns with lexicality (word vs. pseudoword) and reading skill (poor vs. highly-skilled). Classifiers identified validation set lexicality with significantly better than chance accuracy, and reading skill with near-ceiling accuracy. Critically, a series of replications used pre-trained rhyming-task models to classify reading skill from mental multiplication task participants' connectivity with near-ceiling accuracy. The novel deep learning approach presented here provides the clearest demonstration to date that reading-skill dependent functional connectivity within the reading network influences brain processing dynamics across cognitive domains.

Categorical and Dimensional Diagnoses of Dyslexia: Are They Compatible?

Dyslexia is often assessed using categorical diagnoses, and subtypes of dyslexia are also recognized in a categorical fashion. Children may meet the criteria for dyslexia, and they may more specifically meet the criteria for a subtype of it, and thus get a diagnosis. This approach to diagnosis clashes with the actual distribution of reading performance in children (which is normal and continuous), and it has received criticism. This article offers a conceptual framework for conciliating these two positions. In short, the proposal is to use a set of multicomponent continuous assessments of reading, rather than thresholds. The proposal is explained using original data obtained from a sample of 30 children (age 7 to 11), tested in the United Kingdom. Using an assessment based on categorical-thresholds, only five children in our sample qualify for extra assistance, and only one may get a diagnosis of dyslexia, while with the mixed system proposed, a few additional children in the gray area would receive attention. This approach would not discard previous categorical approaches such as those distinguishing between surface and phonological dyslexia, but it would rather see these subtypes of dyslexia as the instance of a lower score on the continuum obtained on a single component of the multicomponent assessment.

A molecular-genetic and imaging-genetic approach to specific comprehension difficulties in children

Children with poor reading comprehension despite typical word reading skills were examined using neuropsychological, genetic, and neuroimaging data collected from the Genes, Reading and Dyslexia Study of 1432 Hispanic American and African American children. This unexpected poor comprehension was associated with profound deficits in vocabulary, when compared to children with comprehension skills consistent with their word reading. Those with specific comprehension difficulties were also more likely to have RU2Short alleles of READ1 regulatory variants of DCDC2, strongly associated with reading and language difficulties. Subjects with RU2Short alleles showed stronger resting state functional connectivity between the right insula/inferior frontal gyrus and the right supramarginal gyrus, even after controlling for potentially confounding variables including genetic ancestry and socioeconomic status. This multi-disciplinary approach advances the current understanding of specific reading comprehension difficulties, and suggests the need for interventions that are more appropriately tailored to the specific comprehension deficits of this group of children.

Effects of SYN1Q555X mutation on cortical gray matter microstructure

A new Q555X mutation on the SYN1 gene was recently found in several members of a family segregating dyslexia, epilepsy, and autism spectrum disorder. To describe the effects of this mutation on cortical gray matter microstructure, we performed a surface-based group study using novel diffusion and quantitative multiparametric imaging on 13 SYN1Q555X mutation carriers and 13 age- and sex-matched controls. Specifically, diffusion kurtosis imaging (DKI) and neurite orientation and dispersion and density imaging (NODDI) were used to analyze multi-shell diffusion data and obtain parametric maps sensitive to tissue structure, while quantitative metrics sensitive to tissue composition (T1, T2* and relative proton density [PD]) were obtained from a multi-echo variable flip angle FLASH acquisition. Results showed significant microstructural alterations in several regions usually involved in oral and written language as well as dyslexia. The most significant changes in these regions were lowered mean diffusivity and increased fractional anisotropy. This study is, to our knowledge, the first to successfully use diffusion imaging and multiparametric mapping to detect cortical anomalies in a group of subjects with a well-defined genotype linked to language impairments, epilepsy and autism spectrum disorder (ASD).

Brain basis of cognitive resilience: Prefrontal cortex predicts better reading comprehension in relation to decoding

OBJECTIVE

The ultimate goal of reading is to understand written text. To accomplish this, children must first master decoding, the ability to translate printed words into sounds. Although decoding and reading comprehension are highly interdependent, some children struggle to decode but comprehend well, whereas others with good decoding skills fail to comprehend. The neural basis underlying individual differences in this discrepancy between decoding and comprehension abilities is virtually unknown.

METHODS

We investigated the neural basis underlying reading discrepancy, defined as the difference between reading comprehension and decoding skills, in a three-part study: 1) The neuroanatomical basis of reading discrepancy in a cross-sectional sample of school-age children with a wide range of reading abilities (Experiment-1; n = 55); 2) Whether a discrepancy-related neural signature is present in beginning readers and predictive of future discrepancy (Experiment-2; n = 43); and 3) Whether discrepancy-related regions are part of a domain-general or a language specialized network, utilizing the 1000 Functional Connectome data and large-scale reverse inference from Neurosynth.org (Experiment-3).

RESULTS

Results converged onto the left dorsolateral prefrontal cortex (DLPFC), as related to having discrepantly higher reading comprehension relative to decoding ability. Increased gray matter volume (GMV) was associated with greater discrepancy (Experiment-1). Region-of-interest (ROI) analyses based on the left DLPFC cluster identified in Experiment-1 revealed that regional GMV within this ROI in beginning readers predicted discrepancy three years later (Experiment-2). This region was associated with the fronto-parietal network that is considered fundamental for working memory and cognitive control (Experiment-3).

INTERPRETATION

Processes related to the prefrontal cortex might be linked to reading discrepancy. The findings may be important for understanding cognitive resilience, which we operationalize as those individuals with greater higher-order reading skills such as reading comprehension compared to lower-order reading skills such as decoding skills. Our study provides insights into reading development, existing theories of reading, and cognitive processes that are potentially significant to a wide range of reading disorders.

Reading Profiles in Multi-Site Data With Missingness

Children with reading disability exhibit varied deficits in reading and cognitive abilities that contribute to their reading comprehension problems. Some children exhibit primary deficits in phonological processing, while others can exhibit deficits in oral language and executive functions that affect comprehension. This behavioral heterogeneity is problematic when missing data prevent the characterization of different reading profiles, which often occurs in retrospective data sharing initiatives without coordinated data collection. Here we show that reading profiles can be reliably identified based on Random Forest classification of incomplete behavioral datasets, after the missForest method is used to multiply impute missing values. Results from simulation analyses showed that reading profiles could be accurately classified across degrees of missingness (e.g., ∼5% classification error for 30% missingness across the sample). The application of missForest to a real multi-site dataset with missingness (n = 924) showed that reading disability profiles significantly and consistently differed in reading and cognitive abilities for cases with and without missing data. The results of validation analyses indicated that the reading profiles (cases with and without missing data) exhibited significant differences for an independent set of behavioral variables that were not used to classify reading profiles. Together, the results show how multiple imputation can be applied to the classification of cases with missing data and can increase the integrity of results from multi-site open access datasets.

The Importance of the Left Occipitotemporal Cortex in Developmental Dyslexia

Purpose of Review

Developmental dyslexia is characterized by an impaired acquisition of fluent and skilled reading ability. Numerous studies have explored the neural correlates of this neurodevelopmental disorder, with most classic accounts strongly focussing on left temporoparietal regions. We will review recent findings from structural and functional MRI studies that suggest a more important role of occipitotemporal cortex abnormalities in dyslexia.

Recent Findings

Recent findings highlight the role of the occipitotemporal cortex which exhibits functional as well as structural abnormalities in dyslexic readers and in children at risk for dyslexia and suggest a more central role for the occipitotemporal cortex in the pathophysiology of dyslexia.

Summary

We demonstrate the importance of the occipitotemporal cortex in for understanding impaired reading acquisition and point out how future research might enhance our understanding of functional and structural impairments in the reading network via large-scale data analysis approaches.

Neuropsychology of Learning Disabilities: The Past and the Future

Over the past 50 years, research on children and adults with learning disabilities has seen significant advances. Neuropsychological research historically focused on the administration of tests sensitive to brain dysfunction to identify putative neural mechanisms underlying learning disabilities that would serve as the basis for treatment. Led by research on classifying and identifying learning disabilities, four pivotal changes in research paradigms have produced a contemporary scientific, interdisciplinary, and international understanding of these disabilities. These changes are (1) the emergence of cognitive science, (2) the development of quantitative and molecular genetics, (3) the advent of noninvasive structural and functional neuroimaging, and (4) experimental trials of interventions focused on improving academic skills and addressing comorbid conditions. Implications for practice indicate a need to move neuropsychological assessment away from a primary focus on systematic, comprehensive assessment of cognitive skills toward more targeted performance-based assessments of academic achievement, comorbid conditions, and intervention response that lead directly to evidence-based treatment plans. Future research will continue to cross disciplinary boundaries to address questions regarding the interaction of neurobiological and contextual variables, the importance of individual differences in treatment response, and an expanded research base on (a) the most severe cases, (b) older people with LDs, and (c) domains of math problem solving, reading comprehension, and written expression. (JINS, 2017, 23, 930-940).

Common Brain Structure Findings Across Children with Varied Reading Disability Profiles

Dyslexia is a developmental disorder in reading that exhibits varied patterns of expression across children. Here we examined the degree to which different kinds of reading disabilities (defined as profiles or patterns of reading problems) contribute to brain morphology results in Jacobian determinant images that represent local brain shape and volume. A matched-pair brain morphometry approach was used to control for confounding from brain size and research site effects in this retrospective multi-site study of 134 children from eight different research sites. Parietal operculum, corona radiata, and internal capsule differences between cases and controls were consistently observed across children with evidence of classic dyslexia, specific comprehension deficit, and language learning disability. Thus, there can be common brain morphology findings across children with quite varied reading disability profiles that we hypothesize compound the developmental difficulties of children with unique reading disability profiles and reasons for their reading disability.