Altered global brain network topology as a trait marker in patients with anorexia nervosa

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.

Disrupted longitudinal restoration of brain connectivity during weight normalization in severe anorexia nervosa

Altered intrinsic brain connectivity of patients with anorexia nervosa has been observed in the acute phase of the disorder, but it remains unclear to what extent these alterations recover during weight normalization. In this study, we used functional imaging data from three time points to probe longitudinal changes in intrinsic connectivity patterns in patients with severe anorexia nervosa (BMI ≤ 15.5 kg/m²) over the course of weight normalization. At three distinct stages of inpatient treatment, we examined resting-state functional connectivity in 27 women with severe anorexia nervosa and 40 closely matched healthy controls. Using network-based statistics and graph-theoretic measures, we examined differences in global network strength, subnetworks with altered intrinsic connectivity, and global network topology. Patients with severe anorexia nervosa showed weakened intrinsic connectivity and altered network topology which did not recover during treatment. The persistent disruption of brain networks suggests sustained alterations of information processing in weight-recovered severe anorexia nervosa.

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.

Individual behavioral trajectories shape whole-brain connectivity in mice

It is widely assumed that our actions shape our brains and that the resulting connections determine who we are. To test this idea in a reductionist setting, in which genes and environment are controlled, we investigated differences in neuroanatomy and structural covariance by ex vivo structural magnetic resonance imaging in mice whose behavioral activity was continuously tracked for 3 months in a large, enriched environment. We confirmed that environmental enrichment increases mouse hippocampal volumes. Stratifying the enriched group according to individual longitudinal behavioral trajectories, however, revealed striking differences in mouse brain structural covariance in continuously highly active mice compared to those whose trajectories showed signs of habituating activity. Network-based statistics identified distinct subnetworks of murine structural covariance underlying these differences in behavioral activity. Together, these results reveal that differentiated behavioral trajectories of mice in an enriched environment are associated with differences in brain connectivity.

Complex functional brain network properties in anorexia nervosa

BACKGROUND

Anorexia nervosa (AN) is a disorder characterized by an incapacitating fear of weight gain and by a disturbance in the way the body is experienced, facets that motivate dangerous weight loss behaviors. Multimodal neuroimaging studies highlight atypical neural activity in brain networks involved in interoceptive awareness and reward processing.

METHODS

The current study used resting-state neuroimaging to model the architecture of large-scale functional brain networks and characterize network properties of individual brain regions to clinical measures. Resting-state neuroimaging was conducted in 62 adolescents, 22 (21 female) with a history of AN and 40 (39 female) healthy controls (HCs). Sensorimotor and basal ganglia regions, as part of a 165-region whole-brain network, were investigated. Subject-specific functional brain networks were computed to index centrality. A contrast analysis within the general linear model covarying for age was performed. Correlations between network properties and behavioral measures were conducted (significance q < .05).

RESULTS

Compared to HCs, AN had lower connectivity from sensorimotor regions, and greater connectivity from the left caudate nucleus to the right postcentral gyrus. AN demonstrated lower sensorimotor centrality, but higher basal ganglia centrality. Sensorimotor connectivity dyads and centrality exhibited negative correlations with body dissatisfaction and drive for thinness, two essential features of AN.

CONCLUSIONS

These findings suggest that AN is associated with greater communication from the basal ganglia, and lower information propagation in sensorimotor cortices. This is consistent with the clinical presentation of AN, where individuals exhibit patterns of rigid habitual behavior that is not responsive to bodily needs, and seem "disconnected" from their bodies.

Altered White Matter Connectivity in Young Acutely Underweight Patients With Anorexia Nervosa

OBJECTIVE

Reductions of gray matter volume and cortical thickness in anorexia nervosa (AN) are well documented. However, findings regarding the integrity of white matter (WM) as studied via diffusion weighted imaging (DWI) are remarkably heterogeneous, and WM connectivity has been examined only in small samples using a limited number of regions of interest. The present study investigated whole-brain WM connectivity for the first time in a large sample of acutely underweight patients with AN.

METHOD

DWI data from predominantly adolescent patients with acute AN (n = 96, mean age = 16.3 years) and age-matched healthy control participants (n = 96, mean age = 17.2 years) were analyzed. WM connectivity networks were generated from fiber-tractography-derived streamlines connecting 233 cortical/subcortical regions. To identify group differences, network-based statistic was used while taking head motion, WM, and ventricular volume into account.

RESULTS

Patients with AN were characterized by 6 WM subnetworks with abnormal architecture, as indicated by increased fractional anisotropy located primarily in parietal-occipital regions and accompanied by reduced radial diffusivity. Group differences based on number of streamlines reached only nominal significance.

CONCLUSION

Our study reveals pronounced alterations in the WM connectome in young patients with AN. In contrast to known reductions in gray matter in the acutely underweight state of AN, this pattern does not necessarily indicate a deterioration of the WM network. Future studies using advanced MRI sequences will have to clarify interrelations with axonal packing or myelination, and whether the changes should be considered a consequence of undernutrition or a vulnerability for developing or maintaining AN.

Brain networks in eating disorders: a systematic review of graph theory studies

PURPOSE

Recent evidence from neuroimaging research has shown that eating disorders (EDs) are characterized by alterations in interconnected neural systems, whose characteristics can be usefully described by connectomics tools. The present paper aimed to review the neuroimaging literature in EDs employing connectomic tools, and, specifically, graph theory analysis.

METHODS

A systematic review of the literature was conducted to identify studies employing graph theory analysis on patients with eating disorders published before the 22nd of June 2020.

RESULTS

Twelve studies were included in the systematic review. Ten of them address anorexia nervosa (AN) (AN = 199; acute AN = 85, weight recovered AN with acute diagnosis = 24; fully recovered AN = 90). The remaining two articles address patients with bulimia nervosa (BN) (BN = 48). Global and regional unbalance in segregation and integration properties were described in both disorders.

DISCUSSION

The literature concerning the use of connectomics tools in EDs evidenced the presence of alterations in the topological characteristics of brain networks at a global and at a regional level. Changes in local characteristics involve areas that have been demonstrated to be crucial in the neurobiology and pathophysiology of EDs. Regional imbalances in network properties seem to reflect on global patterns.

LEVEL OF EVIDENCE

Level I, systematic review.

Shift Toward Randomness in Brain Networks of Patients With Anorexia Nervosa: The Role of Malnutrition

No study to date investigated structural white matter (WM) connectome characteristics in patients with anorexia nervosa (AN). Previous research in AN found evidence of imbalances in global and regional connectomic brain architecture and highlighted a role of malnutrition in determining structural brain changes. The aim of our study was to explore the characteristics of the WM network architecture in a sample of patients with AN. Thirty-six patients with AN and 36 healthy women underwent magnetic resonance imaging to obtain a high-resolution three-dimensional T1-weighted anatomical image and a diffusion tensor imaging scan. Probabilistic tractography data were extracted and analyzed in their network properties through graph theory tools. In comparison to healthy women, patients with AN showed lower global network segregation (normalized clustering: p = 0.029), an imbalance between global network integration and segregation (i.e., lower small-worldness: p = 0.031), and the loss of some of the most integrative and influential hubs. Both clustering and small-worldness correlated with the lowest lifetime body mass index. A significant relationship was found between the average regional loss of cortical volume and changes in network properties of brain nodes: the more the difference in the cortical volume of brain areas, the more the increase in the centrality of corresponding nodes in the whole brain, and the decrease in clustering and efficiency of the nodes of parietal cortex. Our findings showed an unbalanced connectome wiring in AN patients, which seems to be influenced by malnutrition and loss of cortical volume. The role of this rearrangement in the maintenance and prognosis of AN and its reversibility with clinical improvement needs to be established by future studies.

Study protocol of comprehensive risk evaluation for anorexia nervosa in twins (CREAT): a study of discordant monozygotic twins with anorexia nervosa

BACKGROUND

Anorexia nervosa (AN) is a severe disorder, for which genetic evidence suggests psychiatric as well as metabolic origins. AN has high somatic and psychiatric comorbidities, broad impact on quality of life, and elevated mortality. Risk factor studies of AN have focused on differences between acutely ill and recovered individuals. Such comparisons often yield ambiguous conclusions, as alterations could reflect different effects depending on the comparison. Whereas differences found in acutely ill patients could reflect state effects that are due to acute starvation or acute disease-specific factors, they could also reflect underlying traits. Observations in recovered individuals could reflect either an underlying trait or a "scar" due to lasting effects of sustained undernutrition and illness. The co-twin control design (i.e., monozygotic [MZ] twins who are discordant for AN and MZ concordant control twin pairs) affords at least partial disambiguation of these effects.

METHODS

Comprehensive Risk Evaluation for Anorexia nervosa in Twins (CREAT) will be the largest and most comprehensive investigation of twins who are discordant for AN to date. CREAT utilizes a co-twin control design that includes endocrinological, neurocognitive, neuroimaging, genomic, and multi-omic approaches coupled with an experimental component that explores the impact of an overnight fast on most measured parameters.

DISCUSSION

The multimodal longitudinal twin assessment of the CREAT study will help to disambiguate state, trait, and "scar" effects, and thereby enable a deeper understanding of the contribution of genetics, epigenetics, cognitive functions, brain structure and function, metabolism, endocrinology, microbiology, and immunology to the etiology and maintenance of AN.

Neuroimaging and eating disorders

PURPOSE OF REVIEW

Eating disorders are severe psychiatric disorders with a suspected complex biopsychosocial cause. The purpose of this review is to synthesize the recent literature on brain imaging in eating disorders.

RECENT FINDINGS

Food restriction as well as binge eating and purging behaviors are associated with lower regional brain volumes or cortical thickness, but those changes largely return to normal with normalization of weight and eating behavior. Computational modeling has started to identify patterns of structural and functional imaging data that classify eating disorder subtypes, which could be used in the future, diagnostically and to better understand disorder-specific psychopathology. The prediction error model, a computational approach to assess dopamine-related brain reward function, helped support a brain-based model for anorexia nervosa. In that model, the conscious motivation to restrict conflicts with body signals that stimulate eating. This conflict causes anxiety and drives a vicious cycle of food restriction.

SUMMARY

Novel brain research supports the notion that eating disorders have distinct neurobiological underpinnings. This new knowledge can be used to describe disease models to patients and develop novel treatments.

Functional Connectivity Patterns and the Role of 5-HTTLPR Polymorphism on Network Architecture in Female Patients With Anorexia Nervosa

Introduction

Recent neuroimaging studies suggest that anorexia nervosa (AN) symptoms emerge from failures in the relationships between spatially distributed networks that support different cognitive, emotional, and somatosensory functions. The 5-HTTLPR genotype has been shown to modulate all these abilities in AN, as well as the connectivity patterns between brain regions that support their functioning. This study aims at exploring the presence of any difference in functional connectome properties between AN patients and healthy controls (HC) by means of graph theory tools. The effect of 5-HTTLPR genotype on regional and global network characteristics in AN and HC was also explored.

Methods

A sample of 74 subjects (38 HC, 36 AN) underwent a resting state functional magnetic resonance imaging and was genotyped for 5-HTTLPR polymorphism. Comparisons of network properties were made between the AN and HC groups and, within each group, between 5-HTTLPR carriers of low-functioning alleles and carriers of the long–long genotype.

Results

Patients with AN displayed lower network clustering than HC (p = 0.04 at Mann–Whitney U test). Based on both degree and betweenness, a different distribution of network hubs emerged in the two groups. In particular, the anterior part of the anterior cingulate cortex was a hub only in the patient group. A correlation emerged between differences in brain volumes between patients and HC and differences in degree values of basal ganglia, nodes in the insula, and those in the parietal cortex. Carriers of the short allele of the 5-HTTLPR polymorphism were characterized by lower small-world properties (p = 0.027) and modularity (p = 0.031) in the patient group, and a trend toward higher modularity (p = 0.033) and small-world values (p = 0.123) in the HC group.

Discussion

Patients with AN showed differences in hubs distribution, providing evidence of the presence of a different functional architectural backbone in this group. Since some correlation emerged between different degree values of nodes and differences in volumes, further longitudinal studies are warranted to better understand the role of malnutrition on brain network architecture. The opposite effects of 5-HTTLPR polymorphism on global network characteristics in the two groups suggest an interaction of the short allele and malnutrition in modulating brain network properties.

Functional Correlates of Resting-State Connectivity in the Default Mode Network of Heroin Users on Methadone Treatment and Medication-Free Therapeutic Community Program

The treatment of heroin addiction is a complex process involving changes in addictive behavior and brain functioning. The goal of this study was to explore the brain default mode network (DMN) functional connectivity using resting-state functional magnetic resonance imaging (rs-fMRI) and decision-making performance based on the Cambridge gambling task in heroin-dependent individuals undergoing methadone treatment (MT, n = 11) and medication-free faith-based therapeutic community program (TC, n = 11). The DMN involved the medial prefrontal cortex (mPFC), left inferior parietal lobe (IPL_L), right inferior parietal lobe (IPL_R), and posterior cingulate cortex (PCC) subregions for all participants in both the MT and TC groups. Compared with MT, TC had an increased functional connectivity in IPL_L-IPL_R and IPL_R-PCC and decreased functional connectivity in mPFC-IPL_L and IPL_L-PCC. Both groups exhibited no significant difference in the regional rs-fMRI metric [i.e., amplitude of low-frequency fluctuation (ALFF)]. In the analysis of the neural correlates for decision-making performance, risk adjustment was positively associated with ALFF in IPL_L for all participants considering the group effects. The involvement of IPL in decision-making performance and treatment response among heroin-dependent patients warrants further investigation.