Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV
Background: A constantly growing body of literature shows that children from low- socioeconomic status (SES) backgrounds are at risk of cognitive developmental delays, poor health outcomes, and cognitive difficulties which lead to high rates of school drop-outs and struggles in other areas of life....
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English English |
| Published: |
Department of Human Biology
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613141049802752 |
|---|---|
| access_status_str | Open Access |
| author | Khobo, Isaac |
| author2 | Robertson, Frances |
| author_browse | Khobo, Isaac Robertson, Frances |
| author_facet | Robertson, Frances Khobo, Isaac |
| author_sort | Khobo, Isaac |
| collection | Thesis |
| description | Background: A constantly growing body of literature shows that children from low- socioeconomic status (SES) backgrounds are at risk of cognitive developmental delays, poor health outcomes, and cognitive difficulties which lead to high rates of school drop-outs and struggles in other areas of life. In sub-Saharan Africa, where the human immu- nodeficiency virus (HIV) is the most prevalent, low-SES households and communities are disproportionately affected by the disease and its effects on neurodevelopment. The ability to predict cognitive abilities or deficits from neuroimaging or other methods could make it easier to identify at-risk children who may benefit the most from targeted inter-ventions. This is of relevance in low-SES populations with relatively high rates of child-hood HIV that may affect neurodevelopment. Magnetic resonance (MR) imaging (MRI) is a versatile tool that can be used to measure a broad range of brain tissue properties giving rise to cognitive functions. For example, structural MRI (sMRI) can quantify brain volumes and other morphometrics, diffusion tensor imaging (DTI) can estimate the amount of nerve fibre damage, and proton MR spectroscopy (1H-MRS) can charac- terise the biochemical profile of grey and white matter (GM, WM).Research aims : The aims of this study were: First, collect evidence for what is known about the relationship between cognitive performance assessed by a comprehensive set of cognitive test batteries and brain changes measured with neuroimaging in children, adolescents, and youth living with HIV. Second, compare the predictive performance of penalised linear models (PLMs), support vector machines/regression (SVM/R), and de-cision tree ensembles (DTEs) in predicting continuous scores on cognitive tests, as well as categories of cognitive performance from multimodal neuroimaging in a cohort com-prising both children living with and without HIV. Third, determine whether multimodal MRI offers any predictive advantage compared to predicting future performance using cognitive scores at a younger age. Methods and materials : To address these aims, we first conducted a systematic liter-ature review and secondly a multimodal MRI neuroimaging and cognitive testing study of 132 children from low-SES backgrounds. For the review, we searched PubMed, Scopus, Web of Science, CINAHL, APA Psych Info & Psych Articles, and Academic Search Premier for studies published between 1 January 2006 and 31 October 2022. Inclusion criteria were studies that investigated a relationship between neuroimaging brain measures and cognitive test scores and included children (0–14 years), adolescents (15–18), and youth (19–26) living with HIV. For the neuroimaging and cognitive study, structural MRI, DTI, and 1H-MRS were ac-quired at ages 7 and 9 years. Cognitive performance was assessed using the Kaufman assessment battery for children, Beery-Buktenica developmental test of visual-motor in- tegrations, test of variables of attention, Purdue pegboard test, the Peabody picture vocabulary test, and semantic fluency test at both ages. PLMs, SVMs/R, and DTEs prediction models were implemented with Bayesian optimization and assessed with 10-fold cross validation (CV) and compared for their ability to predict continuous scores (regression) or categories of cognitive performance (classification). Poorer and better cognitive performance categories were identified with a hierarchical clustering algorithm. Regression performance was assessed via 10-fold CV errors, coefficient of determination (R2), and Pearson's r between predicted and actual values. For the classification models, 10-fold CV sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were obtained.Results : Evidence from the literature suggests that HIV may lead to alterations in the brain's structure, function, neurometabolism, and WM microstructure. Individual brain measures are linked to outcomes of short-term memory, processing speed, working memory, problem solving, and general intelligence quotients in children, adolescents, and youth living with and without. We could not find any studies linking multimodal MRI to cognitive performance in this population of young people. PLMs, SVMs/SVR, and DTEs performed poorly for the regression problem; the predic-tive models led to small training and fitting errors but high generalised CV errors. How-ever, using either multimodal MRI data or cognitive scores at age 7, we could predict auditory working memory (R2 = 0.45, r = 0.75), short-term memory (R2 = 0.43, r =0.62), visual-motor integration (R2 = 0.26, r = 0.39), and executive reasoning (R2 = 0.33, r = 0.27) scores at age 9 with moderate to strong Pearson's r. Classification of children into poorer or better performance categories was more successful than regression of the individual scores, with 0.75–0.81 AUC, 70–77% accuracies, 70–81% specificities, 71–79% sensitivities using historic multimodal MRI and cognitive scores. Historic multimodal MRI (AUC = 0.80, accuracy = 76%) was marginally better than cognitive scores (AUC= 0.75, accuracy = 70%) in classifying future overall cognitive performance.Conclusion: There were multimodal brain measures relevant in the prediction models, these included creatine and glutamate concentrations in midfrontal gray matter region, thalamus volume, diffusivity in the cingulum WM tract, cingulate gyrus area, and gyri-fication index of the parietal lobe. This suggests that multiple MRI modalities and fea-tures should be considered simultaneously to establish correlates of overall cognitive performance. The neural correlates we find could potentially be used to identify bi-omarkers of cognitive impairment, understand the developmental nature of cognitive plasticity, and enable the development of targeted interventions that can modulate brain networks associated with cognitive functions. |
| format | Thesis |
| id | oai:open.uct.ac.za:11427/41639 |
| institution | University of Cape Town (South Africa) |
| language | English eng |
| last_indexed | 2026-06-10T12:31:24.573Z |
| license_str | Not specified — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from UCTD — University of Cape Town Open Access Repository |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| publisher | Department of Human Biology |
| publisherStr | Department of Human Biology |
| record_format | dspace |
| source_str | UCTD — University of Cape Town Open Access Repository |
| spelling | oai:open.uct.ac.za:11427/41639 Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV Khobo, Isaac Robertson, Frances Meintjes, Ernesta HIV, MRI features Background: A constantly growing body of literature shows that children from low- socioeconomic status (SES) backgrounds are at risk of cognitive developmental delays, poor health outcomes, and cognitive difficulties which lead to high rates of school drop-outs and struggles in other areas of life. In sub-Saharan Africa, where the human immu- nodeficiency virus (HIV) is the most prevalent, low-SES households and communities are disproportionately affected by the disease and its effects on neurodevelopment. The ability to predict cognitive abilities or deficits from neuroimaging or other methods could make it easier to identify at-risk children who may benefit the most from targeted inter-ventions. This is of relevance in low-SES populations with relatively high rates of child-hood HIV that may affect neurodevelopment. Magnetic resonance (MR) imaging (MRI) is a versatile tool that can be used to measure a broad range of brain tissue properties giving rise to cognitive functions. For example, structural MRI (sMRI) can quantify brain volumes and other morphometrics, diffusion tensor imaging (DTI) can estimate the amount of nerve fibre damage, and proton MR spectroscopy (1H-MRS) can charac- terise the biochemical profile of grey and white matter (GM, WM).Research aims : The aims of this study were: First, collect evidence for what is known about the relationship between cognitive performance assessed by a comprehensive set of cognitive test batteries and brain changes measured with neuroimaging in children, adolescents, and youth living with HIV. Second, compare the predictive performance of penalised linear models (PLMs), support vector machines/regression (SVM/R), and de-cision tree ensembles (DTEs) in predicting continuous scores on cognitive tests, as well as categories of cognitive performance from multimodal neuroimaging in a cohort com-prising both children living with and without HIV. Third, determine whether multimodal MRI offers any predictive advantage compared to predicting future performance using cognitive scores at a younger age. Methods and materials : To address these aims, we first conducted a systematic liter-ature review and secondly a multimodal MRI neuroimaging and cognitive testing study of 132 children from low-SES backgrounds. For the review, we searched PubMed, Scopus, Web of Science, CINAHL, APA Psych Info & Psych Articles, and Academic Search Premier for studies published between 1 January 2006 and 31 October 2022. Inclusion criteria were studies that investigated a relationship between neuroimaging brain measures and cognitive test scores and included children (0–14 years), adolescents (15–18), and youth (19–26) living with HIV. For the neuroimaging and cognitive study, structural MRI, DTI, and 1H-MRS were ac-quired at ages 7 and 9 years. Cognitive performance was assessed using the Kaufman assessment battery for children, Beery-Buktenica developmental test of visual-motor in- tegrations, test of variables of attention, Purdue pegboard test, the Peabody picture vocabulary test, and semantic fluency test at both ages. PLMs, SVMs/R, and DTEs prediction models were implemented with Bayesian optimization and assessed with 10-fold cross validation (CV) and compared for their ability to predict continuous scores (regression) or categories of cognitive performance (classification). Poorer and better cognitive performance categories were identified with a hierarchical clustering algorithm. Regression performance was assessed via 10-fold CV errors, coefficient of determination (R2), and Pearson's r between predicted and actual values. For the classification models, 10-fold CV sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were obtained.Results : Evidence from the literature suggests that HIV may lead to alterations in the brain's structure, function, neurometabolism, and WM microstructure. Individual brain measures are linked to outcomes of short-term memory, processing speed, working memory, problem solving, and general intelligence quotients in children, adolescents, and youth living with and without. We could not find any studies linking multimodal MRI to cognitive performance in this population of young people. PLMs, SVMs/SVR, and DTEs performed poorly for the regression problem; the predic-tive models led to small training and fitting errors but high generalised CV errors. How-ever, using either multimodal MRI data or cognitive scores at age 7, we could predict auditory working memory (R2 = 0.45, r = 0.75), short-term memory (R2 = 0.43, r =0.62), visual-motor integration (R2 = 0.26, r = 0.39), and executive reasoning (R2 = 0.33, r = 0.27) scores at age 9 with moderate to strong Pearson's r. Classification of children into poorer or better performance categories was more successful than regression of the individual scores, with 0.75–0.81 AUC, 70–77% accuracies, 70–81% specificities, 71–79% sensitivities using historic multimodal MRI and cognitive scores. Historic multimodal MRI (AUC = 0.80, accuracy = 76%) was marginally better than cognitive scores (AUC= 0.75, accuracy = 70%) in classifying future overall cognitive performance.Conclusion: There were multimodal brain measures relevant in the prediction models, these included creatine and glutamate concentrations in midfrontal gray matter region, thalamus volume, diffusivity in the cingulum WM tract, cingulate gyrus area, and gyri-fication index of the parietal lobe. This suggests that multiple MRI modalities and fea-tures should be considered simultaneously to establish correlates of overall cognitive performance. The neural correlates we find could potentially be used to identify bi-omarkers of cognitive impairment, understand the developmental nature of cognitive plasticity, and enable the development of targeted interventions that can modulate brain networks associated with cognitive functions. 2025-08-28T11:33:02Z 2025-08-28T11:33:02Z 2025 2025-08-28T11:31:26Z Thesis / Dissertation Doctoral PhD http://hdl.handle.net/11427/41639 en eng application/pdf Department of Human Biology Faculty of Health Sciences University of Cape Town |
| spellingShingle | HIV, MRI features Khobo, Isaac Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV |
| thesis_degree_str | Doctoral |
| title | Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV |
| title_full | Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV |
| title_fullStr | Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV |
| title_full_unstemmed | Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV |
| title_short | Exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children: applying machine learning algorithms to brain MRI features to predict cognitive scores and performance categories of children living with and without HIV |
| title_sort | exploring the relationship between multimodal magnetic resonance neuroimaging and cognitive outcomes in children applying machine learning algorithms to brain mri features to predict cognitive scores and performance categories of children living with and without hiv |
| topic | HIV, MRI features |
| url | http://hdl.handle.net/11427/41639 |
| work_keys_str_mv | AT khoboisaac exploringtherelationshipbetweenmultimodalmagneticresonanceneuroimagingandcognitiveoutcomesinchildrenapplyingmachinelearningalgorithmstobrainmrifeaturestopredictcognitivescoresandperformancecategoriesofchildrenlivingwithandwithouthiv |