Wenxiong Liao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenxiong Liao has a total of 55 co-authors. Here are the top co-authors by number of shared publications:
- Hongmin Cai
South China University of Technology, Guangzhou, China
10 shared publications - Tianming Liu
University of Georgia
8 shared publications - Dajiang Zhu
University of Texas at Arlington
7 shared publications - Quanzheng Li
Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA
7 shared publications - Xiang Li
Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, USA
7 shared publications - Zihao Wu
University of Georgia
7 shared publications
Latest Publications
- Open Access
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Leveraging the Hierarchical Structure of ICD Codes to Enhance Multimodal EHR Retrieval
Background: Similar EHR (Electronic Health Records) retrieval has the potential to support diagnostic decision-making by retrieving prior patient records with related clinical profiles. However, this task is hindered by two major challenges: the scarcity of annotated similar patient pairs, as well as the difficulty of fusing structured tables and unstructured clinical notes.Methods: To address these issues, we propose EHR-Retriever, an EHR retrieval framework that exploits the hierarchical semantics of ICD (International Classification of Diseases) codes as supervision signals and integrates multimodal information from tabular and unstructured sources. The framework contains a hierarchical diagnosis modeling module, which organizes the ICD code set into a diagnosis tree and encodes this tree into a diagnosis representation, and a cross-modal module termed X-Fuse, which explicitly models bidirectional interactions between structured and unstructured data via dual-channel cross-attention. Evaluation was conducted on MIMIC-III, with external evaluation on MIMIC-IV ICD-9 and ICD-10 subsets.Results: On MIMIC-III, EHR-Retriever achieved nDCG@10 of 0.7157, nDCG@5 of 0.7129, mJI@10 of 0.4432, mAP@10 of 0.6497, and mAP@5 of 0.6507. Component analyses showed that single-modality variants, simplified hierarchical diagnosis modeling, and serialization-based tabular-text fusion strategies reduced retrieval performance. On the MIMIC-IV ICD-9 query set, EHR-Retriever achieved the highest nDCG@10, nDCG@5, mAP@10, and mAP@5 among the evaluated methods, while BM25 achieved a slightly higher mJI@10. On the MIMIC-IV ICD-10 setting, EHR-Retriever achieved the highest values on all reported metrics.Conclusions: The results highlight the effectiveness of ICD code-driven multimodal integration and the model's ability to capture diagnostically relevant patterns in EHRs. The evaluation is based on MIMIC datasets and ICD-derived relevance metrics; deployment across heterogeneous EHR systems and privacy-preserving safeguards remain future work.
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessLeveraging the Hierarchical Structure of ICD Codes to Enhance Multimodal EHR Retrieval
Background: Similar EHR (Electronic Health Records) retrieval has the potential to support diagnostic decision-making by retrieving prior patient records with related clinical profiles. However, this task is hindered by two major challenges: the scarcity of annotated similar patient pairs, as well as the difficulty of fusing structured tables and unstructured clinical notes.Methods: To address these issues, we propose EHR-Retriever, an EHR retrieval framework that exploits the hierarchical semantics of ICD (International Classification of Diseases) codes as supervision signals and integrates multimodal information from tabular and unstructured sources. The framework contains a hierarchical diagnosis modeling module, which organizes the ICD code set into a diagnosis tree and encodes this tree into a diagnosis representation, and a cross-modal module termed X-Fuse, which explicitly models bidirectional interactions between structured and unstructured data via dual-channel cross-attention. Evaluation was conducted on MIMIC-III, with external evaluation on MIMIC-IV ICD-9 and ICD-10 subsets.Results: On MIMIC-III, EHR-Retriever achieved nDCG@10 of 0.7157, nDCG@5 of 0.7129, mJI@10 of 0.4432, mAP@10 of 0.6497, and mAP@5 of 0.6507. Component analyses showed that single-modality variants, simplified hierarchical diagnosis modeling, and serialization-based tabular-text fusion strategies reduced retrieval performance. On the MIMIC-IV ICD-9 query set, EHR-Retriever achieved the highest nDCG@10, nDCG@5, mAP@10, and mAP@5 among the evaluated methods, while BM25 achieved a slightly higher mJI@10. On the MIMIC-IV ICD-10 setting, EHR-Retriever achieved the highest values on all reported metrics.Conclusions: The results highlight the effectiveness of ICD code-driven multimodal integration and the model's ability to capture diagnostically relevant patterns in EHRs. The evaluation is based on MIMIC datasets and ICD-derived relevance metrics; deployment across heterogeneous EHR systems and privacy-preserving safeguards remain future work.
- Open Access
- 23 July 2026
- Published by Elsevier in SSRN Electronic Journal
Rank–Distortion Survival: Time-Scale Invariant Survival Modeling on Percentile Geometry
Survival analysis models censored time-to-event data, providing a fundamental basis for data-driven prediction and decision support in intelligent systems. Most existing survival models are formulated on time-domain primitives such as hazards or parametric event-time distributions, whose interpretation and validity depend intrinsically on the chosen time scale. When this time scale is misaligned with modeling assumptions, distributional misspecification can arise. In this paper, we introduce Rank–Distortion Survival (RDSurv), which formulates survival modeling in terms of failure rank rather than time. Each event time is mapped to a failure percentile, encoding its relative position within a reference population and preserving ordering under monotone time transformations. Covariate effects are then modeled as distortions of the percentile distribution on the unit interval, formalized as covariate-dependent transformation operators acting on ranks. This rank–distortion formulation yields a general representation of conditional survival distributions in which time enters only through a baseline percentile chart, and leads to a well-defined likelihood for right-censored data directly in percentile space. As a result, RDSurv provides individualized survival curves and an intrinsic time-scale-invariant risk ranking. Experiments on multiple benchmark survival datasets demonstrate that RDSurv consistently outperforms representative time-based survival models, while remaining stable under monotone transformations of the time axis. The source code will be made publicly available upon acceptance.
- 23 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessRank–Distortion Survival: Time-Scale Invariant Survival Modeling on Percentile Geometry
Survival analysis models censored time-to-event data, providing a fundamental basis for data-driven prediction and decision support in intelligent systems. Most existing survival models are formulated on time-domain primitives such as hazards or parametric event-time distributions, whose interpretation and validity depend intrinsically on the chosen time scale. When this time scale is misaligned with modeling assumptions, distributional misspecification can arise. In this paper, we introduce Rank–Distortion Survival (RDSurv), which formulates survival modeling in terms of failure rank rather than time. Each event time is mapped to a failure percentile, encoding its relative position within a reference population and preserving ordering under monotone time transformations. Covariate effects are then modeled as distortions of the percentile distribution on the unit interval, formalized as covariate-dependent transformation operators acting on ranks. This rank–distortion formulation yields a general representation of conditional survival distributions in which time enters only through a baseline percentile chart, and leads to a well-defined likelihood for right-censored data directly in percentile space. As a result, RDSurv provides individualized survival curves and an intrinsic time-scale-invariant risk ranking. Experiments on multiple benchmark survival datasets demonstrate that RDSurv consistently outperforms representative time-based survival models, while remaining stable under monotone transformations of the time axis. The source code will be made publicly available upon acceptance.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30  (7) , 5783-5793
GraphSTAR: Proximal Operator-Based Graph Neural Network Enhanced by Dynamic Graph Aggregation for Spatial Transcriptomics
Spatial transcriptomics technologies carry out advanced sequencing analysis of molecular profiles with a spatial context, providing multi-source information essential for elucidating biological regulatory mechanisms. Nonetheless, it poses challenges in the integration of raw spatial coordinates with high-dimensional gene expression profiles in their native feature space. While spatial-aware methods effectively aggregate molecular information from local spatial neighborhoods, they fail to explore the long-range relationships associated with gene expression data. To address this issue, this paper introduces a novel approach termed GraphSTAR that encodes both spatial and gene expression data into undirected graphs, characterizing the local spatial proximity and global transcriptional similarity, respectively. Through a graph aggregation process, GraphSTAR integrates these diverse data sources within a joint graph structure, effectively modeling both local neighborhood relationships and long-range functional associations. Subsequently, a reassembled graph neural network is established by incorporating the graph aggregation into the feed-forward propagation using proximal operators, progressively refining spatial-informed latent representation to decipher spatial expression patterns of genes. Extensive experiments on benchmark datasets demonstrate that GraphSTAR outperforms state-of-the-art methods in both spatial domain identification and cell-type annotation tasks.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30  (7) , 5783-5793
GraphSTAR: Proximal Operator-Based Graph Neural Network Enhanced by Dynamic Graph Aggregation for Spatial Transcriptomics
Spatial transcriptomics technologies carry out advanced sequencing analysis of molecular profiles with a spatial context, providing multi-source information essential for elucidating biological regulatory mechanisms. Nonetheless, it poses challenges in the integration of raw spatial coordinates with high-dimensional gene expression profiles in their native feature space. While spatial-aware methods effectively aggregate molecular information from local spatial neighborhoods, they fail to explore the long-range relationships associated with gene expression data. To address this issue, this paper introduces a novel approach termed GraphSTAR that encodes both spatial and gene expression data into undirected graphs, characterizing the local spatial proximity and global transcriptional similarity, respectively. Through a graph aggregation process, GraphSTAR integrates these diverse data sources within a joint graph structure, effectively modeling both local neighborhood relationships and long-range functional associations. Subsequently, a reassembled graph neural network is established by incorporating the graph aggregation into the feed-forward propagation using proximal operators, progressively refining spatial-informed latent representation to decipher spatial expression patterns of genes. Extensive experiments on benchmark datasets demonstrate that GraphSTAR outperforms state-of-the-art methods in both spatial domain identification and cell-type annotation tasks.
- Open Access
FunduSAM: A Specialized Deep Learning Model for Enhanced Optic Disc and Cup Segmentation in Fundus Images
The Segment Anything Model (SAM) has gained popularity as a versatile image segmentation method, thanks to its strong generalization capabilities across various domains. However, when applied to optic disc (OD) and optic cup (OC) segmentation tasks, SAM encounters challenges due to the complex structures, low contrast, and blurred boundaries typical of fundus images, leading to suboptimal performance. To overcome these challenges, we introduce a novel model, FunduSAM, which incorporates several Adapters into SAM to create a deep network specifically designed for OD and OC segmentation. The FunduSAM utilizes Adapter into each transformer block after encoder for parameter fine-tuning (PEFT). It enhances SAM's feature extraction capabilities by designing a Convolutional Block Attention Module (CBAM), addressing issues related to blurred boundaries and low contrast. Given the unique requirements of OD and OC segmentation, polar transformation is used to convert the original fundus OD images into a format better suited for training and evaluating FunduSAM. A joint loss is used to achieve structure preservation between the OD and OC, while accurate segmentation. Extensive experiments on the REFUGE dataset, comprising 1,200 fundus images, demonstrate the superior performance of FunduSAM compared to five mainstream approaches.
Open AccessFunduSAM: A Specialized Deep Learning Model for Enhanced Optic Disc and Cup Segmentation in Fundus Images
The Segment Anything Model (SAM) has gained popularity as a versatile image segmentation method, thanks to its strong generalization capabilities across various domains. However, when applied to optic disc (OD) and optic cup (OC) segmentation tasks, SAM encounters challenges due to the complex structures, low contrast, and blurred boundaries typical of fundus images, leading to suboptimal performance. To overcome these challenges, we introduce a novel model, FunduSAM, which incorporates several Adapters into SAM to create a deep network specifically designed for OD and OC segmentation. The FunduSAM utilizes Adapter into each transformer block after encoder for parameter fine-tuning (PEFT). It enhances SAM's feature extraction capabilities by designing a Convolutional Block Attention Module (CBAM), addressing issues related to blurred boundaries and low contrast. Given the unique requirements of OD and OC segmentation, polar transformation is used to convert the original fundus OD images into a format better suited for training and evaluating FunduSAM. A joint loss is used to achieve structure preservation between the OD and OC, while accurate segmentation. Extensive experiments on the REFUGE dataset, comprising 1,200 fundus images, demonstrate the superior performance of FunduSAM compared to five mainstream approaches.
- Top Cited Papers
- 30 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 11  (3) , 907-918
AugGPT: Leveraging ChatGPT for Text Data Augmentation
Text data augmentation is an effective strategy for overcoming the challenge of limited sample sizes in many natural language processing (NLP) tasks. This challenge is especially prominent in the few-shot learning (FSL) scenario, where the data in the target domain is generally much scarcer and of lowered quality. A natural and widely used strategy to mitigate such challenges is to perform data augmentation to better capture data invariance and increase the sample size. However, current text data augmentation methods either can’t ensure the correct labeling of the generated data (lacking faithfulness), or can’t ensure sufficient diversity in the generated data (lacking compactness), or both. Inspired by the recent success of large language models (LLM), especially the development of ChatGPT, we propose a text data augmentation approach based on ChatGPT (named ”AugGPT”). AugGPT rephrases each sentence in the training samples into multiple conceptually similar but semantically different samples. The augmented samples can then be used in downstream model training. Experiment results on multiple few-shot learning text classification tasks show the superior performance of the proposed AugGPT approach over state-of-the-art text data augmentation methods in terms of testing accuracy and distribution of the augmented samples.
- 30 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 11  (3) , 907-918
Top Cited PapersAugGPT: Leveraging ChatGPT for Text Data Augmentation
Text data augmentation is an effective strategy for overcoming the challenge of limited sample sizes in many natural language processing (NLP) tasks. This challenge is especially prominent in the few-shot learning (FSL) scenario, where the data in the target domain is generally much scarcer and of lowered quality. A natural and widely used strategy to mitigate such challenges is to perform data augmentation to better capture data invariance and increase the sample size. However, current text data augmentation methods either can’t ensure the correct labeling of the generated data (lacking faithfulness), or can’t ensure sufficient diversity in the generated data (lacking compactness), or both. Inspired by the recent success of large language models (LLM), especially the development of ChatGPT, we propose a text data augmentation approach based on ChatGPT (named ”AugGPT”). AugGPT rephrases each sentence in the training samples into multiple conceptually similar but semantically different samples. The augmented samples can then be used in downstream model training. Experiment results on multiple few-shot learning text classification tasks show the superior performance of the proposed AugGPT approach over state-of-the-art text data augmentation methods in terms of testing accuracy and distribution of the augmented samples.
- 27 September 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 304
Zero-shot relation triplet extraction as Next-Sentence Prediction
- 27 September 2024
- Published by Elsevier in Knowledge-Based Systems
- Vol. 304
Zero-shot relation triplet extraction as Next-Sentence Prediction
- Open Access
- 28 December 2023
- Published by JMIR Publications Inc. in JMIR Medical Education
- Vol. 9  (1) , e48904
Differentiating ChatGPT-Generated and Human-Written Medical Texts: Quantitative Study
Background: Large language models, such as ChatGPT, are capable of generating grammatically perfect and human-like text content, and a large number of ChatGPT-generated texts have appeared on the internet. However, medical texts, such as clinical notes and diagnoses, require rigorous validation, and erroneous medical content generated by ChatGPT could potentially lead to disinformation that poses significant harm to health care and the general public. Objective: This study is among the first on responsible artificial intelligence–generated content in medicine. We focus on analyzing the differences between medical texts written by human experts and those generated by ChatGPT and designing machine learning workflows to effectively detect and differentiate medical texts generated by ChatGPT. Methods: We first constructed a suite of data sets containing medical texts written by human experts and generated by ChatGPT. We analyzed the linguistic features of these 2 types of content and uncovered differences in vocabulary, parts-of-speech, dependency, sentiment, perplexity, and other aspects. Finally, we designed and implemented machine learning methods to detect medical text generated by ChatGPT. The data and code used in this paper are published on GitHub. Results: Medical texts written by humans were more concrete, more diverse, and typically contained more useful information, while medical texts generated by ChatGPT paid more attention to fluency and logic and usually expressed general terminologies rather than effective information specific to the context of the problem. A bidirectional encoder representations from transformers–based model effectively detected medical texts generated by ChatGPT, and the F1 score exceeded 95%. Conclusions: Although text generated by ChatGPT is grammatically perfect and human-like, the linguistic characteristics of generated medical texts were different from those written by human experts. Medical text generated by ChatGPT could be effectively detected by the proposed machine learning algorithms. This study provides a pathway toward trustworthy and accountable use of large language models in medicine.
- 28 December 2023
- Published by JMIR Publications Inc. in JMIR Medical Education
- Vol. 9  (1) , e48904
Open AccessDifferentiating ChatGPT-Generated and Human-Written Medical Texts: Quantitative Study
Background: Large language models, such as ChatGPT, are capable of generating grammatically perfect and human-like text content, and a large number of ChatGPT-generated texts have appeared on the internet. However, medical texts, such as clinical notes and diagnoses, require rigorous validation, and erroneous medical content generated by ChatGPT could potentially lead to disinformation that poses significant harm to health care and the general public. Objective: This study is among the first on responsible artificial intelligence–generated content in medicine. We focus on analyzing the differences between medical texts written by human experts and those generated by ChatGPT and designing machine learning workflows to effectively detect and differentiate medical texts generated by ChatGPT. Methods: We first constructed a suite of data sets containing medical texts written by human experts and generated by ChatGPT. We analyzed the linguistic features of these 2 types of content and uncovered differences in vocabulary, parts-of-speech, dependency, sentiment, perplexity, and other aspects. Finally, we designed and implemented machine learning methods to detect medical text generated by ChatGPT. The data and code used in this paper are published on GitHub. Results: Medical texts written by humans were more concrete, more diverse, and typically contained more useful information, while medical texts generated by ChatGPT paid more attention to fluency and logic and usually expressed general terminologies rather than effective information specific to the context of the problem. A bidirectional encoder representations from transformers–based model effectively detected medical texts generated by ChatGPT, and the F1 score exceeded 95%. Conclusions: Although text generated by ChatGPT is grammatically perfect and human-like, the linguistic characteristics of generated medical texts were different from those written by human experts. Medical text generated by ChatGPT could be effectively detected by the proposed machine learning algorithms. This study provides a pathway toward trustworthy and accountable use of large language models in medicine.
- 13 September 2023
- Published by Springer Nature
- p. 395-406
Multimodal Approaches for Alzheimer’s Detection Using Patients’ Speech and Transcript
- 13 September 2023
- Published by Springer Nature
- p. 395-406
Multimodal Approaches for Alzheimer’s Detection Using Patients’ Speech and Transcript
- Open Access
Exploring Multimodal Approaches for Alzheimer's Disease Detection Using Patient Speech Transcript and Audio Data
Alzheimer's disease (AD) is a common form of dementia that severely impacts patient health. As AD impairs the patient's language understanding and expression ability, the speech of AD patients can serve as an indicator of this disease. This study investigates various methods for detecting AD using patients' speech and transcripts data from the DementiaBank Pitt database. The proposed approach involves pre-trained language models and Graph Neural Network (GNN) that constructs a graph from the speech transcript, and extracts features using GNN for AD detection. Data augmentation techniques, including synonym replacement, GPT-based augmenter, and so on, were used to address the small dataset size. Audio data was also introduced, and WavLM model was used to extract audio features. These features were then fused with text features using various methods. Finally, a contrastive learning approach was attempted by converting speech transcripts back to audio and using it for contrastive learning with the original audio. We conducted intensive experiments and analysis on the above methods. Our findings shed light on the challenges and potential solutions in AD detection using speech and audio data.
Open AccessExploring Multimodal Approaches for Alzheimer's Disease Detection Using Patient Speech Transcript and Audio Data
Alzheimer's disease (AD) is a common form of dementia that severely impacts patient health. As AD impairs the patient's language understanding and expression ability, the speech of AD patients can serve as an indicator of this disease. This study investigates various methods for detecting AD using patients' speech and transcripts data from the DementiaBank Pitt database. The proposed approach involves pre-trained language models and Graph Neural Network (GNN) that constructs a graph from the speech transcript, and extracts features using GNN for AD detection. Data augmentation techniques, including synonym replacement, GPT-based augmenter, and so on, were used to address the small dataset size. Audio data was also introduced, and WavLM model was used to extract audio features. These features were then fused with text features using various methods. Finally, a contrastive learning approach was attempted by converting speech transcripts back to audio and using it for contrastive learning with the original audio. We conducted intensive experiments and analysis on the above methods. Our findings shed light on the challenges and potential solutions in AD detection using speech and audio data.
- Open Access
- 19 May 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Differentiating ChatGPT-Generated and Human-Written Medical Texts: Quantitative Study (Preprint)
BACKGROUND Large language models, such as ChatGPT, are capable of generating grammatically perfect and human-like text content, and a large number of ChatGPT-generated texts have appeared on the internet. However, medical texts, such as clinical notes and diagnoses, require rigorous validation, and erroneous medical content generated by ChatGPT could potentially lead to disinformation that poses significant harm to health care and the general public. OBJECTIVE This study is among the first on responsible artificial intelligence–generated content in medicine. We focus on analyzing the differences between medical texts written by human experts and those generated by ChatGPT and designing machine learning workflows to effectively detect and differentiate medical texts generated by ChatGPT. METHODS We first constructed a suite of data sets containing medical texts written by human experts and generated by ChatGPT. We analyzed the linguistic features of these 2 types of content and uncovered differences in vocabulary, parts-of-speech, dependency, sentiment, perplexity, and other aspects. Finally, we designed and implemented machine learning methods to detect medical text generated by ChatGPT. The data and code used in this paper are published on GitHub. RESULTS Medical texts written by humans were more concrete, more diverse, and typically contained more useful information, while medical texts generated by ChatGPT paid more attention to fluency and logic and usually expressed general terminologies rather than effective information specific to the context of the problem. A bidirectional encoder representations from transformers–based model effectively detected medical texts generated by ChatGPT, and the F1 score exceeded 95%. CONCLUSIONS Although text generated by ChatGPT is grammatically perfect and human-like, the linguistic characteristics of generated medical texts were different from those written by human experts. Medical text generated by ChatGPT could be effectively detected by the proposed machine learning algorithms. This study provides a pathway toward trustworthy and accountable use of large language models in medicine.
- 19 May 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessDifferentiating ChatGPT-Generated and Human-Written Medical Texts: Quantitative Study (Preprint)
BACKGROUND Large language models, such as ChatGPT, are capable of generating grammatically perfect and human-like text content, and a large number of ChatGPT-generated texts have appeared on the internet. However, medical texts, such as clinical notes and diagnoses, require rigorous validation, and erroneous medical content generated by ChatGPT could potentially lead to disinformation that poses significant harm to health care and the general public. OBJECTIVE This study is among the first on responsible artificial intelligence–generated content in medicine. We focus on analyzing the differences between medical texts written by human experts and those generated by ChatGPT and designing machine learning workflows to effectively detect and differentiate medical texts generated by ChatGPT. METHODS We first constructed a suite of data sets containing medical texts written by human experts and generated by ChatGPT. We analyzed the linguistic features of these 2 types of content and uncovered differences in vocabulary, parts-of-speech, dependency, sentiment, perplexity, and other aspects. Finally, we designed and implemented machine learning methods to detect medical text generated by ChatGPT. The data and code used in this paper are published on GitHub. RESULTS Medical texts written by humans were more concrete, more diverse, and typically contained more useful information, while medical texts generated by ChatGPT paid more attention to fluency and logic and usually expressed general terminologies rather than effective information specific to the context of the problem. A bidirectional encoder representations from transformers–based model effectively detected medical texts generated by ChatGPT, and the F1 score exceeded 95%. CONCLUSIONS Although text generated by ChatGPT is grammatically perfect and human-like, the linguistic characteristics of generated medical texts were different from those written by human experts. Medical text generated by ChatGPT could be effectively detected by the proposed machine learning algorithms. This study provides a pathway toward trustworthy and accountable use of large language models in medicine.