Selected work

Representative Papers

MemoCare: An Interactive Multimodal Mobile System for Automated Cognitive Screening

Oct 07, 2026

This study addresses the lack of multimodal mobile cognitive screening systems by developing an automated English-Vietnamese bilingual screening platform using React Native, integrating voice, spatial, touch, and drawing tasks. We propose a novel multimodal automated scoring architecture that combines deterministic natural language processing rules with a ShuffleNetV2 convolutional neural network consensus mechanism to achieve standardized cross-lingual assessment. Experimental results demonstrate that all software tests were successfully passed, the drawing module attained a balanced accuracy of 91.33%, and clinical experts awarded a comprehensive rating of 4 out of 5. These findings validate the effectiveness and clinical applicability of the proposed system for cross-lingual cognitive screening.

0 citationsRead paper

Trusting the Inverse: Reliability-Aware Mapping for Simulation-Based Microstructure Estimation in Diffusion MRI

Sep 30, 2026

This study addresses the inherent ambiguity in diffusion MRI microstructure parameter estimation and the lack of voxel-wise reliability characterization by proposing a reliability-aware framework based on out-of-distribution signals, local mismatches, and parameter degeneracy. By constructing a geometrically realistic Monte Carlo dictionary constrained by electron microscopy data, this work designs a complementary three-dimensional scoring system to quantify estimation uncertainty, achieving the first effective decoupling of extrapolation errors, interpolation errors, and degeneracy while overcoming traditional point-estimation limitations. Experiments demonstrate that the reliability index correlates significantly with actual errors in synthetic data (ρ=−0.742), and over half of the voxels in in vivo rat and human brain data yield highly reliable estimates, thereby validating the effectiveness of the proposed method.

0 citationsRead paper

ProteinPNet: Prototypical Part Networks for Concept Learning in Spatial Proteomics

Dec 02, 2025

Resolving the spatial heterogeneity of the tumor microenvironment (TME) is critical for precision oncology, yet existing methods struggle to learn biologically grounded, discriminative, and interpretable prototypes directly from spatial proteomics data. To address this, we propose an end-to-end learnable Prototype-Part Network that jointly integrates supervised contrastive learning, graph-structured modeling, and morphological analysis to automatically discover and interpretably model spatial functional modules within the TME. Evaluated on both synthetic benchmarks and real-world spatial proteomics data from non-small cell lung cancer, our method robustly identifies immune infiltration patterns and tissue modularity features highly concordant with histopathological subtypes. Notably, it achieves the first supervised prototype learning of spatial motifs in the TME—recurring, biologically meaningful spatial configurations of protein expression. This establishes a novel, mechanism-driven paradigm for discovering spatially resolved biomarkers with direct biological interpretability.

0 citationsRead paper

Predicting Cognition from fMRI:A Comparative Study of Graph, Transformer, and Kernel Models Across Task and Rest Conditions

Jul 28, 2025

This study addresses the challenge of predicting individual cognitive abilities from fMRI data to uncover underlying neurobiological mechanisms and advance precision medicine and early detection of neuropsychiatric disorders. Methodologically, we systematically benchmark graph neural networks (GNNs), Transformer-GNNs, and kernel ridge regression (KRR) on both resting-state and task-based fMRI, and propose a multimodal GNN framework integrating structural connectivity (SC) and functional connectivity (FC). Results show that task-based fMRI yields significantly higher predictive accuracy than resting-state fMRI; the SC-FC–integrated GNN achieves superior and more stable performance across most metrics; Transformer-GNN excels in dynamic modeling for task-based data but underperforms in resting-state settings; notably, its accuracy is not statistically distinguishable from FC-only KRR. This work establishes a reproducible benchmark framework for multimodal brain imaging and provides mechanistic insights into connectome-based cognitive prediction.

0 citationsRead paper

Meta Learning not to Learn: Robustly Informing Meta-Learning under Nuisance-Varying Families

Mar 06, 2025

In multi-source heterogeneous tasks, the coexistence of causal and spurious features leads to out-of-distribution (OOD) generalization failure. Method: We propose a collaborative modeling framework of positive and negative inductive biases: (i) guiding models to learn task-invariant causal features (“what to learn”), and (ii) explicitly suppressing spurious correlations (“what not to learn”). We theoretically prove that existing knowledge fusion methods fail under distributionally robust objectives; accordingly, we design RIME—a novel algorithm integrating causal inference, distributionally robust optimization, meta-learning (a MAML variant), and adversarial disentanglement—to jointly optimize both biases under a family of spurious-feature shifts. Contribution/Results: Experiments demonstrate that RIME significantly improves generalization stability in real-world OOD settings—e.g., cross-hospital medical image prognosis prediction—and achieves state-of-the-art performance in distributionally robust meta-learning.

0 citationsRead paper

Recent publications

Latest Papers

MemoCare: An Interactive Multimodal Mobile System for Automated Cognitive Screening

Oct 07, 2026

This study addresses the lack of multimodal mobile cognitive screening systems by developing an automated English-Vietnamese bilingual screening platform using React Native, integrating voice, spatial, touch, and drawing tasks. We propose a novel multimodal automated scoring architecture that combines deterministic natural language processing rules with a ShuffleNetV2 convolutional neural network consensus mechanism to achieve standardized cross-lingual assessment. Experimental results demonstrate that all software tests were successfully passed, the drawing module attained a balanced accuracy of 91.33%, and clinical experts awarded a comprehensive rating of 4 out of 5. These findings validate the effectiveness and clinical applicability of the proposed system for cross-lingual cognitive screening.

0 citationsRead paper

Trusting the Inverse: Reliability-Aware Mapping for Simulation-Based Microstructure Estimation in Diffusion MRI

Sep 30, 2026

This study addresses the inherent ambiguity in diffusion MRI microstructure parameter estimation and the lack of voxel-wise reliability characterization by proposing a reliability-aware framework based on out-of-distribution signals, local mismatches, and parameter degeneracy. By constructing a geometrically realistic Monte Carlo dictionary constrained by electron microscopy data, this work designs a complementary three-dimensional scoring system to quantify estimation uncertainty, achieving the first effective decoupling of extrapolation errors, interpolation errors, and degeneracy while overcoming traditional point-estimation limitations. Experiments demonstrate that the reliability index correlates significantly with actual errors in synthetic data (ρ=−0.742), and over half of the voxels in in vivo rat and human brain data yield highly reliable estimates, thereby validating the effectiveness of the proposed method.

0 citationsRead paper

ProteinPNet: Prototypical Part Networks for Concept Learning in Spatial Proteomics

Dec 02, 2025

Resolving the spatial heterogeneity of the tumor microenvironment (TME) is critical for precision oncology, yet existing methods struggle to learn biologically grounded, discriminative, and interpretable prototypes directly from spatial proteomics data. To address this, we propose an end-to-end learnable Prototype-Part Network that jointly integrates supervised contrastive learning, graph-structured modeling, and morphological analysis to automatically discover and interpretably model spatial functional modules within the TME. Evaluated on both synthetic benchmarks and real-world spatial proteomics data from non-small cell lung cancer, our method robustly identifies immune infiltration patterns and tissue modularity features highly concordant with histopathological subtypes. Notably, it achieves the first supervised prototype learning of spatial motifs in the TME—recurring, biologically meaningful spatial configurations of protein expression. This establishes a novel, mechanism-driven paradigm for discovering spatially resolved biomarkers with direct biological interpretability.

0 citationsRead paper

Predicting Cognition from fMRI:A Comparative Study of Graph, Transformer, and Kernel Models Across Task and Rest Conditions

Jul 28, 2025

This study addresses the challenge of predicting individual cognitive abilities from fMRI data to uncover underlying neurobiological mechanisms and advance precision medicine and early detection of neuropsychiatric disorders. Methodologically, we systematically benchmark graph neural networks (GNNs), Transformer-GNNs, and kernel ridge regression (KRR) on both resting-state and task-based fMRI, and propose a multimodal GNN framework integrating structural connectivity (SC) and functional connectivity (FC). Results show that task-based fMRI yields significantly higher predictive accuracy than resting-state fMRI; the SC-FC–integrated GNN achieves superior and more stable performance across most metrics; Transformer-GNN excels in dynamic modeling for task-based data but underperforms in resting-state settings; notably, its accuracy is not statistically distinguishable from FC-only KRR. This work establishes a reproducible benchmark framework for multimodal brain imaging and provides mechanistic insights into connectome-based cognitive prediction.

0 citationsRead paper

Meta Learning not to Learn: Robustly Informing Meta-Learning under Nuisance-Varying Families

Mar 06, 2025

In multi-source heterogeneous tasks, the coexistence of causal and spurious features leads to out-of-distribution (OOD) generalization failure. Method: We propose a collaborative modeling framework of positive and negative inductive biases: (i) guiding models to learn task-invariant causal features (“what to learn”), and (ii) explicitly suppressing spurious correlations (“what not to learn”). We theoretically prove that existing knowledge fusion methods fail under distributionally robust objectives; accordingly, we design RIME—a novel algorithm integrating causal inference, distributionally robust optimization, meta-learning (a MAML variant), and adversarial disentanglement—to jointly optimize both biases under a family of spurious-feature shifts. Contribution/Results: Experiments demonstrate that RIME significantly improves generalization stability in real-world OOD settings—e.g., cross-hospital medical image prognosis prediction—and achieves state-of-the-art performance in distributionally robust meta-learning.

0 citationsRead paper