Geaur Rahman
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Geaur Rahman has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Zahidul Islam
School of Computing, Mathematics and Engineering, Charles Sturt University, Bathurst, NSW 2795, Australia
10 shared publications - Minh Chau
Teaching Innovation in Radiography Group Charles Sturt University Wagga Wagga New South Wales Australia
4 shared publications - Tanmoy Debnath
School of Computing, Mathematics and Engineering, Charles Sturt University, Bathurst, NSW 2795, Australia
4 shared publications - Anisur Rahman
La Trobe Business School, La Trobe University, Bundoora, Australia
2 shared publications - Mohammad Zavid Parvez
Charles Sturt University
2 shared publications - Sabih Ur Rehman
School of Computing, Mathematics and Engineering, Charles Sturt University, Port Macquarie, NSW, Australia
2 shared publications
Latest Publications
- Open Access
- 17 September 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 129
A Sparse-Attention Fusion Network integrating heterogeneous multimodal features for Parkinson’s disease classification
- 17 September 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 129
Open AccessA Sparse-Attention Fusion Network integrating heterogeneous multimodal features for Parkinson’s disease classification
- Open Access
Regional Brain Volume Variation Across Adulthood: A Cross-Sectional MRI Analysis of Age, Sex, and Hemispheric Asymmetry
Distinguishing normative cross-sectional structural age differences from early neurodegeneration requires region-specific characterisation of cross-sectional age-related variation, yet findings remain sensitive to modelling strategy and intracranial volume (ICV) correction. T1-weighted MRI from 187 healthy adults (97 female, 90 male; aged 16.22–81.48 years) in the AgeRisk dataset were segmented using Vol2Brain. Multiple regression models applied to 19 predefined brain regions tested linear and quadratic age, sex, ICV, and signal-to-noise ratio (FDR-corrected within predictor families); hemispheric asymmetry indices were Bonferroni-corrected. Three hypothesis domains were preregistered: age associations, sex differences, and hemispheric asymmetry. Negative cross-sectional age associations were observed in subcortical (hippocampus, amygdala, caudate, putamen, and nucleus accumbens) and cortical regions (insula, temporal and occipital lobes; all ). The parietal lobe showed significant quadratic curvature (); total cortical grey matter showed no age effect. White matter exhibited a positive linear rather than the hypothesised inverted-U relationship. After ICV adjustment, females showed greater grey-matter-dominant volumes and males greater cerebrospinal fluid volume (both ). Sex-by-age interactions in four subcortical regions indicated shallower negative cross-sectional age associations in females (all ). Age was associated with reduced parietal and temporal hemispheric asymmetry (both ). Findings confirm regionally heterogeneous ageing, ICV-adjusted sex differences, and a selective negative cross-sectional association between age and hemispheric asymmetry, consistent with the broader lifespan literature.Open AccessRegional Brain Volume Variation Across Adulthood: A Cross-Sectional MRI Analysis of Age, Sex, and Hemispheric Asymmetry
Distinguishing normative cross-sectional structural age differences from early neurodegeneration requires region-specific characterisation of cross-sectional age-related variation, yet findings remain sensitive to modelling strategy and intracranial volume (ICV) correction. T1-weighted MRI from 187 healthy adults (97 female, 90 male; aged 16.22–81.48 years) in the AgeRisk dataset were segmented using Vol2Brain. Multiple regression models applied to 19 predefined brain regions tested linear and quadratic age, sex, ICV, and signal-to-noise ratio (FDR-corrected within predictor families); hemispheric asymmetry indices were Bonferroni-corrected. Three hypothesis domains were preregistered: age associations, sex differences, and hemispheric asymmetry. Negative cross-sectional age associations were observed in subcortical (hippocampus, amygdala, caudate, putamen, and nucleus accumbens) and cortical regions (insula, temporal and occipital lobes; all ). The parietal lobe showed significant quadratic curvature (); total cortical grey matter showed no age effect. White matter exhibited a positive linear rather than the hypothesised inverted-U relationship. After ICV adjustment, females showed greater grey-matter-dominant volumes and males greater cerebrospinal fluid volume (both ). Sex-by-age interactions in four subcortical regions indicated shallower negative cross-sectional age associations in females (all ). Age was associated with reduced parietal and temporal hemispheric asymmetry (both ). Findings confirm regionally heterogeneous ageing, ICV-adjusted sex differences, and a selective negative cross-sectional association between age and hemispheric asymmetry, consistent with the broader lifespan literature. - Open Access
- 27 June 2026
- Published by Elsevier in SSRN Electronic Journal
A Federated Multimodal Co-Attention Framework for Privacy-Preserving Healthcare Analytics
Accurate, interpretable, and privacy-preserving diagnosis of thoracic diseases remains a criticalchallenge for healthcare organizations seeking to improve clinical decision-making whileprotecting patient data. Conventional deep learning models that rely on a single data modality,such as chest radiographs or radiology reports, often fail to capture complementary clinicalinformation and provide limited interpretability. Although multimodal artificial intelligence(AI) approaches have demonstrated improved diagnostic performance by integrating visualand textual information, most require centralized data aggregation, creating concerns regardingpatient privacy, regulatory compliance, and deployment across distributed healthcare environments.This study proposes FedCoAttention, a federated multimodal co-attention framework forhealthcare analytics and clinical decision support. The framework integrates pretrained DistilledBidirectional Encoder Representations from Transformers (DistilBERT) and Residual Network-50 (ResNet50) encoders through a transformer-based co-attention mechanism that captures bidirectionalinteractions between textual tokens and image patches. Collaborative model trainingis achieved using Federated Averaging (FedAvg), enabling multiple healthcare institutions tojointly develop predictive models without sharing raw patient data. To support transparency andclinician trust, the framework incorporates Patch Importance Maps and Gradient-weighted ClassActivation Mapping (Grad-CAM) visualizations that align clinically relevant report contentwith salient radiographic regions. Experimental evaluation on the Indiana Chest X-ray dataset,consisting of Normal, Cardiomegaly, and Atelectasis cases, demonstrates strong predictiveperformance and stable federated convergence. The proposed framework achieved a best globalaccuracy of 98.2%, a macro F1-score of 0.96, and a weighted F1-score of 0.98. Across threeindependent federated training runs, the model attained a mean accuracy of 97.7% ± 0.4%,indicating robust and reproducible performance. The findings demonstrate the potential offederated multimodal analytics to support privacy-preserving clinical decision support acrossdistributed healthcare organizations while enhancing diagnostic accuracy, interpretability, andscalability.
- 27 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA Federated Multimodal Co-Attention Framework for Privacy-Preserving Healthcare Analytics
Accurate, interpretable, and privacy-preserving diagnosis of thoracic diseases remains a criticalchallenge for healthcare organizations seeking to improve clinical decision-making whileprotecting patient data. Conventional deep learning models that rely on a single data modality,such as chest radiographs or radiology reports, often fail to capture complementary clinicalinformation and provide limited interpretability. Although multimodal artificial intelligence(AI) approaches have demonstrated improved diagnostic performance by integrating visualand textual information, most require centralized data aggregation, creating concerns regardingpatient privacy, regulatory compliance, and deployment across distributed healthcare environments.This study proposes FedCoAttention, a federated multimodal co-attention framework forhealthcare analytics and clinical decision support. The framework integrates pretrained DistilledBidirectional Encoder Representations from Transformers (DistilBERT) and Residual Network-50 (ResNet50) encoders through a transformer-based co-attention mechanism that captures bidirectionalinteractions between textual tokens and image patches. Collaborative model trainingis achieved using Federated Averaging (FedAvg), enabling multiple healthcare institutions tojointly develop predictive models without sharing raw patient data. To support transparency andclinician trust, the framework incorporates Patch Importance Maps and Gradient-weighted ClassActivation Mapping (Grad-CAM) visualizations that align clinically relevant report contentwith salient radiographic regions. Experimental evaluation on the Indiana Chest X-ray dataset,consisting of Normal, Cardiomegaly, and Atelectasis cases, demonstrates strong predictiveperformance and stable federated convergence. The proposed framework achieved a best globalaccuracy of 98.2%, a macro F1-score of 0.96, and a weighted F1-score of 0.98. Across threeindependent federated training runs, the model attained a mean accuracy of 97.7% ± 0.4%,indicating robust and reproducible performance. The findings demonstrate the potential offederated multimodal analytics to support privacy-preserving clinical decision support acrossdistributed healthcare organizations while enhancing diagnostic accuracy, interpretability, andscalability.
- Open Access
- 18 June 2026
- Published by American Psychological Association (APA) in Neuropsychology
- Vol. 40 (8) , 791-811
Integrity in six key brain regions predicts numeric working memory performance in 30,000 middle-aged and older adults: A UK Biobank magnetic resonance imaging study.
This study investigated the neuroanatomical correlates of working memory using the UK Biobank Numeric Memory Test (NMT; a visual backward digit-span task) in a large sample of middle-aged and older adults. Volumes of brain regions previously implicated in digit-span tasks were hypothesized to positively correlate with NMT performance; however, some differences were expected due to visual NMT format (e.g., reduced relevance for language regions). A secondary aim was to generate the first normative data set for the NMT. As part of the UK Biobank study, 30,640 healthy adults aged 51-80 completed both NMT and brain magnetic resonance imaging. Linear fixed-effects modeling assessed whether brain volumes (25 regions of interest) could predict NMT performance after accounting for age, sex, education, and head size. Normative data were generated as percentile tables stratified by age/sex/education. Linear fixed-effects modeling revealed that better NMT performance was predicted by larger volumes of total cerebellar cortex, bilateral superior temporal, bilateral insula, left inferior parietal, left hippocampus, and left lateral occipital (p range: <.001-.048), with a moderate effect size (f² = 0.072). Both linear fixed-effects and normative data revealed that NMT performance decreases with age but increases with education (both p < .001), with no discernible sex differences. The findings support the idea that greater neurostructural integrity correlates with better cognitive performance. While future multimodal longitudinal studies are needed to confirm structure-function associations and more diverse ethnic/socioeconomic groups are needed to improve generalizability, these findings could inform clinical decision making and targeted treatment for working memory dysfunction. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
- 18 June 2026
- Published by American Psychological Association (APA) in Neuropsychology
- Vol. 40 (8) , 791-811
Open AccessIntegrity in six key brain regions predicts numeric working memory performance in 30,000 middle-aged and older adults: A UK Biobank magnetic resonance imaging study.
This study investigated the neuroanatomical correlates of working memory using the UK Biobank Numeric Memory Test (NMT; a visual backward digit-span task) in a large sample of middle-aged and older adults. Volumes of brain regions previously implicated in digit-span tasks were hypothesized to positively correlate with NMT performance; however, some differences were expected due to visual NMT format (e.g., reduced relevance for language regions). A secondary aim was to generate the first normative data set for the NMT. As part of the UK Biobank study, 30,640 healthy adults aged 51-80 completed both NMT and brain magnetic resonance imaging. Linear fixed-effects modeling assessed whether brain volumes (25 regions of interest) could predict NMT performance after accounting for age, sex, education, and head size. Normative data were generated as percentile tables stratified by age/sex/education. Linear fixed-effects modeling revealed that better NMT performance was predicted by larger volumes of total cerebellar cortex, bilateral superior temporal, bilateral insula, left inferior parietal, left hippocampus, and left lateral occipital (p range: <.001-.048), with a moderate effect size (f² = 0.072). Both linear fixed-effects and normative data revealed that NMT performance decreases with age but increases with education (both p < .001), with no discernible sex differences. The findings support the idea that greater neurostructural integrity correlates with better cognitive performance. While future multimodal longitudinal studies are needed to confirm structure-function associations and more diverse ethnic/socioeconomic groups are needed to improve generalizability, these findings could inform clinical decision making and targeted treatment for working memory dysfunction. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
- Open Access
- 4 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 7 (7) , 4207-4220
A Semisupervised Domain Adaptation Framework Using Dynamic Distribution Alignment and Manifold Regularization
Domain adaptation (DA) aims to build classifiers for a target domain (TD) by utilising knowledge from a source domain (SD). However, due to two main issues: lack of training data in TD and distribution divergence between the domains, DA can be a highly challenging problem in reality, especially as the existing methods often only prioritise one issue at a time. In this paper, we present a method called SDM that builds a classifier for TD having some unlabelled data by leveraging knowledge from SD having huge labelled data. We first build a classifier applying an existing classification algorithm, such as SVM, on the training data having only SD data. The classifier is then iteratively adjusted by updating the training data, adding TD data and deleting SD data in each iteration. We alleviate both issues by simultaneously minimising the structural risk and joint distributions across domains, and maximising the manifold consistency underneath marginal distributions. Furthermore, while existing methods rely on a user-defined threshold for adding and deleting data, we propose an automatic strategy to determine them. We evaluate the effectiveness of SDM by comparing its performances with fifteen state-of-the-art methods on seven publicly available datasets. We also assess the performance of SDM under certain challenging conditions, such as SD and TD datasets may have noisy and missing values. Our experimental results, including statistical significance tests, demonstrate a clear superiority of SDM over the state-of-the-art methods used in this study. SDM’s source code is publicly available on GitHub at https://github.com/grahman20/SDM.
- 4 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 7 (7) , 4207-4220
Open AccessA Semisupervised Domain Adaptation Framework Using Dynamic Distribution Alignment and Manifold Regularization
Domain adaptation (DA) aims to build classifiers for a target domain (TD) by utilising knowledge from a source domain (SD). However, due to two main issues: lack of training data in TD and distribution divergence between the domains, DA can be a highly challenging problem in reality, especially as the existing methods often only prioritise one issue at a time. In this paper, we present a method called SDM that builds a classifier for TD having some unlabelled data by leveraging knowledge from SD having huge labelled data. We first build a classifier applying an existing classification algorithm, such as SVM, on the training data having only SD data. The classifier is then iteratively adjusted by updating the training data, adding TD data and deleting SD data in each iteration. We alleviate both issues by simultaneously minimising the structural risk and joint distributions across domains, and maximising the manifold consistency underneath marginal distributions. Furthermore, while existing methods rely on a user-defined threshold for adding and deleting data, we propose an automatic strategy to determine them. We evaluate the effectiveness of SDM by comparing its performances with fifteen state-of-the-art methods on seven publicly available datasets. We also assess the performance of SDM under certain challenging conditions, such as SD and TD datasets may have noisy and missing values. Our experimental results, including statistical significance tests, demonstrate a clear superiority of SDM over the state-of-the-art methods used in this study. SDM’s source code is publicly available on GitHub at https://github.com/grahman20/SDM.
- Open Access
- 1 January 2026
- Published by Wiley in Human Behavior and Emerging Technologies
- Vol. 2026 (1)
Data‐Driven and Explainable Discovery of Autism Spectrum Disorder Patterns Across Age Groups
Autism Spectrum Disorder (ASD) presents a complex and diverse challenge, with difficulties in social communication and restricted interests manifesting differently among children, adolescents and adults. The traditional ASD diagnosis is often time consuming and resource intensive, delaying access to crucial interventions. Early detection of ASD is vital, as it enables timely referral for comprehensive assessment and supports interventions during key developmental periods, greatly enhancing long‐term outcomes. To address these challenges, this research presents three complementary studies examining ASD detection across children, adolescents and adults. Using the Autism Spectrum Quotient (AQ) short‐form dataset encompassing behavioural, communication and self‐reported indicators, each study applies decision tree machine learning models to identify behavioural screening patterns within each age group. Study 1 focuses on children, highlighting observable social and behavioural cues; Study 2 investigates adolescents, revealing transitional communication and self‐awareness patterns; and Study 3 explores adults, emphasising subtle self‐reported and behavioural markers. Across studies, interpretable models achieved strong classification accuracy while revealing developmental variations in ASD‐related traits. Collectively, these findings enhance understanding of age‐specific patterns in ASD expression, support earlier and more individualised screening and provide a foundation for designing tailored interventions that align with developmental milestones and improve overall care quality.
- 1 January 2026
- Published by Wiley in Human Behavior and Emerging Technologies
- Vol. 2026 (1)
Open AccessData‐Driven and Explainable Discovery of Autism Spectrum Disorder Patterns Across Age Groups
Autism Spectrum Disorder (ASD) presents a complex and diverse challenge, with difficulties in social communication and restricted interests manifesting differently among children, adolescents and adults. The traditional ASD diagnosis is often time consuming and resource intensive, delaying access to crucial interventions. Early detection of ASD is vital, as it enables timely referral for comprehensive assessment and supports interventions during key developmental periods, greatly enhancing long‐term outcomes. To address these challenges, this research presents three complementary studies examining ASD detection across children, adolescents and adults. Using the Autism Spectrum Quotient (AQ) short‐form dataset encompassing behavioural, communication and self‐reported indicators, each study applies decision tree machine learning models to identify behavioural screening patterns within each age group. Study 1 focuses on children, highlighting observable social and behavioural cues; Study 2 investigates adolescents, revealing transitional communication and self‐awareness patterns; and Study 3 explores adults, emphasising subtle self‐reported and behavioural markers. Across studies, interpretable models achieved strong classification accuracy while revealing developmental variations in ASD‐related traits. Collectively, these findings enhance understanding of age‐specific patterns in ASD expression, support earlier and more individualised screening and provide a foundation for designing tailored interventions that align with developmental milestones and improve overall care quality.
- Open Access
- 31 January 2025
- Published by Elsevier in Radiography
- Vol. 31 (2) , 102878
A scoping review of automatic and semi-automatic MRI segmentation in human brain imaging
- 31 January 2025
- Published by Elsevier in Radiography
- Vol. 31 (2) , 102878
Open AccessA scoping review of automatic and semi-automatic MRI segmentation in human brain imaging
- 6 December 2024
- Published by Springer Nature in Earth Science Informatics
- Vol. 18 (1) , 1-23
Analysis of data cleansing methods for improving meteorological data quality: a case study
- 6 December 2024
- Published by Springer Nature in Earth Science Informatics
- Vol. 18 (1) , 1-23
Analysis of data cleansing methods for improving meteorological data quality: a case study
- 20 November 2024
- Published by Springer Nature
- p. 404-416
Real-Time Human Activity Recognition Using Non-intrusive Sensing and Continual Learning
- 20 November 2024
- Published by Springer Nature
- p. 404-416
Real-Time Human Activity Recognition Using Non-intrusive Sensing and Continual Learning
- Open Access
- 13 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 133030-133044
OutCenTR: A Method for Predicting Exploits of Cyber Vulnerabilities in High Dimensional Datasets
While an ever-growing number of vulnerabilities are reported every day, all the reported vulnerabilities are not all the same, as some are more targeted than others. Correctly estimating the likelihood of a vulnerability being exploited is a critical task for system administrators in cyber security management as it can help them prioritise and patch the right vulnerabilities. However, due to three key issues: unavailability of labeled training data, high dimensionality, and class imbalance in datasets, the prediction of vulnerabilities can be a challenging problem in practice, especially as the existing methods often only prioritise one issue at a time. In this paper, we propose a method called OutCenTR that can predict the likelihood of a vulnerability being exploited by addressing all the issues concurrently. In OutCenTR, the unavailability of labeled training data is addressed by considering a semi-supervised approach which requires only a few labeled data, the high dimensionality issue is addressed by identifying and removing insignificant features, and the class imbalance issue is addressed by introducing context-based distinguishability scores between records. OutCenTR first determines important features in datasets and then makes use of an existing algorithm to build a classifier by considering only the important features. The classifier is not only effective for predicting exploit of vulnerabilities but also valuable for general-purpose outlier detection. We evaluate the effectiveness of OutCenTR by comparing its performance with the performance of five state-of-the-art methods on four publicly available datasets and twelve synthetic datasets. The methods are evaluated in terms of five criteria, namely: Precision, Recall, F1 Score, ROC, and Execution time. Our initial experimental results clearly indicate that the proposed method, OutCenTR outperforms existing methods.
- 13 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 133030-133044
Open AccessOutCenTR: A Method for Predicting Exploits of Cyber Vulnerabilities in High Dimensional Datasets
While an ever-growing number of vulnerabilities are reported every day, all the reported vulnerabilities are not all the same, as some are more targeted than others. Correctly estimating the likelihood of a vulnerability being exploited is a critical task for system administrators in cyber security management as it can help them prioritise and patch the right vulnerabilities. However, due to three key issues: unavailability of labeled training data, high dimensionality, and class imbalance in datasets, the prediction of vulnerabilities can be a challenging problem in practice, especially as the existing methods often only prioritise one issue at a time. In this paper, we propose a method called OutCenTR that can predict the likelihood of a vulnerability being exploited by addressing all the issues concurrently. In OutCenTR, the unavailability of labeled training data is addressed by considering a semi-supervised approach which requires only a few labeled data, the high dimensionality issue is addressed by identifying and removing insignificant features, and the class imbalance issue is addressed by introducing context-based distinguishability scores between records. OutCenTR first determines important features in datasets and then makes use of an existing algorithm to build a classifier by considering only the important features. The classifier is not only effective for predicting exploit of vulnerabilities but also valuable for general-purpose outlier detection. We evaluate the effectiveness of OutCenTR by comparing its performance with the performance of five state-of-the-art methods on four publicly available datasets and twelve synthetic datasets. The methods are evaluated in terms of five criteria, namely: Precision, Recall, F1 Score, ROC, and Execution time. Our initial experimental results clearly indicate that the proposed method, OutCenTR outperforms existing methods.