Jing Mei
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jing Mei has a total of 92 co-authors. Here are the top co-authors by number of shared publications:
Guotong Xie
IBM Research, Beijing, China
13 shared publications- Yiqin Yu
IBM Research China, Beijing, China
12 shared publications Haifeng Liu
IBM Research China Beijing China
11 shared publicationsXiang Li
IBM Research China Beijing China
10 shared publicationsLiu Haifeng
IBM Research China, Beijing, China
5 shared publications- Pengwei Hu
Chinese Academy of Sciences, Xinjiang Technical Institute of Physics and Chemistry, Urumqi, China
5 shared publications
Latest Publications
- Open Access
Medical visual question answering using joint self-supervised learning
Visual Question Answering (VQA) becomes one of the most active research problems in the medical imaging domain. A well-known VQA challenge is the intrinsic diversity between the image and text modalities, and in the medical VQA task, there is another critical problem relying on the limited size of labelled image-question-answer data. In this study we propose an encoder-decoder framework that leverages the image-text joint representation learned from large-scaled medical image-caption data and adapted to the small-sized medical VQA task. The encoder embeds across the image-text dual modalities with self-attention mechanism and is independently pre-trained on the large-scaled medical image-caption dataset by multiple self-supervised learning tasks. Then the decoder is connected to the top of the encoder and fine-tuned using the small-sized medical VQA dataset. The experiment results present that our proposed method achieves better performance comparing with the baseline and SOTA methods.
Open AccessMedical visual question answering using joint self-supervised learning
Visual Question Answering (VQA) becomes one of the most active research problems in the medical imaging domain. A well-known VQA challenge is the intrinsic diversity between the image and text modalities, and in the medical VQA task, there is another critical problem relying on the limited size of labelled image-question-answer data. In this study we propose an encoder-decoder framework that leverages the image-text joint representation learned from large-scaled medical image-caption data and adapted to the small-sized medical VQA task. The encoder embeds across the image-text dual modalities with self-attention mechanism and is independently pre-trained on the large-scaled medical image-caption dataset by multiple self-supervised learning tasks. Then the decoder is connected to the top of the encoder and fine-tuned using the small-sized medical VQA dataset. The experiment results present that our proposed method achieves better performance comparing with the baseline and SOTA methods.
- 1 January 2023
- Published by AHFE International in Applied human factors and ergonomics international
- Vol. 77 (77)
Design strategy of digital outdoor study manual based on children's learning behavior guidance
Applied Human Factors and Ergonomics International
- 1 January 2023
- Published by AHFE International in Applied human factors and ergonomics international
- Vol. 77 (77)
Design strategy of digital outdoor study manual based on children's learning behavior guidance
Applied Human Factors and Ergonomics International
- Open Access
- 4 November 2021
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 21 (S1) , 1-13
Learning from low-rank multimodal representations for predicting disease-drug associations
Background: Disease-drug associations provide essential information for drug discovery and disease treatment. Many disease-drug associations remain unobserved or unknown, and trials to confirm these associations are time-consuming and expensive. To better understand and explore these valuable associations, it would be useful to develop computational methods for predicting unobserved disease-drug associations. With the advent of various datasets describing diseases and drugs, it has become more feasible to build a model describing the potential correlation between disease and drugs. Results: In this work, we propose a new prediction method, called LMFDA, which works in several stages. First, it studies the drug chemical structure, disease MeSH descriptors, disease-related phenotypic terms, and drug-drug interactions. On this basis, similarity networks of different sources are constructed to enrich the representation of drugs and diseases. Based on the fused disease similarity network and drug similarity network, LMFDA calculated the association score of each pair of diseases and drugs in the database. This method achieves good performance on Fdataset and Cdataset, AUROCs were 91.6% and 92.1% respectively, higher than many of the existing computational models. Conclusions: The novelty of LMFDA lies in the introduction of multimodal fusion using low-rank tensors to fuse multiple similar networks and combine matrix complement technology to predict potential association. We have demonstrated that LMFDA can display excellent network integration ability for accurate disease-drug association inferring and achieve substantial improvement over the advanced approach. Overall, experimental results on two real-world networks dataset demonstrate that LMFDA able to delivers an excellent detecting performance. Results also suggest that perfecting similar networks with as much domain knowledge as possible is a promising direction for drug repositioning.
- 4 November 2021
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 21 (S1) , 1-13
Open AccessLearning from low-rank multimodal representations for predicting disease-drug associations
Background: Disease-drug associations provide essential information for drug discovery and disease treatment. Many disease-drug associations remain unobserved or unknown, and trials to confirm these associations are time-consuming and expensive. To better understand and explore these valuable associations, it would be useful to develop computational methods for predicting unobserved disease-drug associations. With the advent of various datasets describing diseases and drugs, it has become more feasible to build a model describing the potential correlation between disease and drugs. Results: In this work, we propose a new prediction method, called LMFDA, which works in several stages. First, it studies the drug chemical structure, disease MeSH descriptors, disease-related phenotypic terms, and drug-drug interactions. On this basis, similarity networks of different sources are constructed to enrich the representation of drugs and diseases. Based on the fused disease similarity network and drug similarity network, LMFDA calculated the association score of each pair of diseases and drugs in the database. This method achieves good performance on Fdataset and Cdataset, AUROCs were 91.6% and 92.1% respectively, higher than many of the existing computational models. Conclusions: The novelty of LMFDA lies in the introduction of multimodal fusion using low-rank tensors to fuse multiple similar networks and combine matrix complement technology to predict potential association. We have demonstrated that LMFDA can display excellent network integration ability for accurate disease-drug association inferring and achieve substantial improvement over the advanced approach. Overall, experimental results on two real-world networks dataset demonstrate that LMFDA able to delivers an excellent detecting performance. Results also suggest that perfecting similar networks with as much domain knowledge as possible is a promising direction for drug repositioning.
- 1 March 2021
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 115, 103686
Disease network delineates the disease progression profile of cardiovascular diseases
- 1 March 2021
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 115, 103686
Disease network delineates the disease progression profile of cardiovascular diseases
- Open Access
Disease Network Delineates the Disease Progression Profile of Cardiovascular Diseases
As Electronic Health Records (EHR) data accumulated explosively in recent years, the tremendous amount of patient clinical data provided opportunities to discover real world evidence. In this study, a graphical disease network, named progressive cardiovascular disease network (progCDN), was built based on EHR data from 14.3 million patients 1 to delineate the progression profiles of cardiovascular diseases (CVD). The network depicted the dominant diseases in CVD development, such as the heart failure and coronary arteriosclerosis. Novel progression relationships were also discovered, such as the progression path from long QT syndrome to major depression. In addition, three age-group progCDNs identified a series of age-associated disease progression paths and important successor diseases with age bias. Furthermore, we extracted a list of salient features to build a series of disease risk models based on the progression pairs in the disease network. The progCDN network can be further used to validate or explore novel disease relationships in real world data. Features with sufficient abundance and high correlation can be widely applied to train disease risk models when using EHR data.
Open AccessDisease Network Delineates the Disease Progression Profile of Cardiovascular Diseases
As Electronic Health Records (EHR) data accumulated explosively in recent years, the tremendous amount of patient clinical data provided opportunities to discover real world evidence. In this study, a graphical disease network, named progressive cardiovascular disease network (progCDN), was built based on EHR data from 14.3 million patients 1 to delineate the progression profiles of cardiovascular diseases (CVD). The network depicted the dominant diseases in CVD development, such as the heart failure and coronary arteriosclerosis. Novel progression relationships were also discovered, such as the progression path from long QT syndrome to major depression. In addition, three age-group progCDNs identified a series of age-associated disease progression paths and important successor diseases with age bias. Furthermore, we extracted a list of salient features to build a series of disease risk models based on the progression pairs in the disease network. The progCDN network can be further used to validate or explore novel disease relationships in real world data. Features with sufficient abundance and high correlation can be widely applied to train disease risk models when using EHR data.
- Open Access
- 21 August 2020
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 22 (8) , e19995
Chinese Public’s Engagement in Preventive and Intervening Health Behaviors During the Early Breakout of COVID-19: Cross-Sectional Study
Background: Since January 2020, the coronavirus disease (COVID-19) swept over China and then the world, causing a global public health crisis. People’s adoption of preventive and intervening behaviors is critical in curbing the spread of the virus. Objective: The aim of this study is to evaluate Chinese people’s adoption of health behaviors in responding to COVID-19 and to identify key determinants for their engagement. Methods: An anonymous online questionnaire was distributed in early February 2020 among Mainland Chinese (18 years or older) to examine their engagement in preventive behaviors (eg, frequent handwashing, wearing masks, staying at home) and intervening behaviors (eg, advising family to wash hands frequently), and to explore potential determinants for their adoption of these health behaviors. Results: Out of 2949 participants, 55.3% (n=1629) reported frequent engagement in preventive health behaviors, and over 84% (n=2493) performed at least one intervening health behavior. Greater engagement in preventive behaviors was found among participants who received higher education, were married, reported fewer barriers and greater benefits of engagement, reported greater self-efficacy and emotional support, had greater patient-centered communication before, had a greater media literacy level, and had greater new media and traditional media use for COVID-19 news. Greater engagement in intervening behaviors was observed among participants who were married, had lower income, reported greater benefits of health behaviors, had greater patient-centered communication before, had a lower media literacy level, and had a greater new media and traditional media use for COVID-19 news. Conclusions: Participants’ engagement in coronavirus-related preventive and intervening behaviors was overall high, and the associations varied across demographic and psychosocial variables. Hence, customized health interventions that address the determinants for health behaviors are needed to improve people’s adherence to coronavirus-related behavior guidelines.
- 21 August 2020
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 22 (8) , e19995
Open AccessChinese Public’s Engagement in Preventive and Intervening Health Behaviors During the Early Breakout of COVID-19: Cross-Sectional Study
Background: Since January 2020, the coronavirus disease (COVID-19) swept over China and then the world, causing a global public health crisis. People’s adoption of preventive and intervening behaviors is critical in curbing the spread of the virus. Objective: The aim of this study is to evaluate Chinese people’s adoption of health behaviors in responding to COVID-19 and to identify key determinants for their engagement. Methods: An anonymous online questionnaire was distributed in early February 2020 among Mainland Chinese (18 years or older) to examine their engagement in preventive behaviors (eg, frequent handwashing, wearing masks, staying at home) and intervening behaviors (eg, advising family to wash hands frequently), and to explore potential determinants for their adoption of these health behaviors. Results: Out of 2949 participants, 55.3% (n=1629) reported frequent engagement in preventive health behaviors, and over 84% (n=2493) performed at least one intervening health behavior. Greater engagement in preventive behaviors was found among participants who received higher education, were married, reported fewer barriers and greater benefits of engagement, reported greater self-efficacy and emotional support, had greater patient-centered communication before, had a greater media literacy level, and had greater new media and traditional media use for COVID-19 news. Greater engagement in intervening behaviors was observed among participants who were married, had lower income, reported greater benefits of health behaviors, had greater patient-centered communication before, had a lower media literacy level, and had a greater new media and traditional media use for COVID-19 news. Conclusions: Participants’ engagement in coronavirus-related preventive and intervening behaviors was overall high, and the associations varied across demographic and psychosocial variables. Hence, customized health interventions that address the determinants for health behaviors are needed to improve people’s adherence to coronavirus-related behavior guidelines.
- Open Access
Dynamic Knowledge Distillation for Black-box Hypothesis Transfer Learning
In real world applications like healthcare, it is usually difficult to build a machine learning prediction model that works universally well across different institutions. At the same time, the available model is often proprietary, i.e., neither the model parameter nor the data set used for model training is accessible. In consequence, leveraging the knowledge hidden in the available model (aka. the hypothesis) and adapting it to a local data set becomes extremely challenging. Motivated by this situation, in this paper we aim to address such a specific case within the hypothesis transfer learning framework, in which 1) the source hypothesis is a black-box model and 2) the source domain data is unavailable. In particular, we introduce a novel algorithm called dynamic knowledge distillation for hypothesis transfer learning (dkdHTL). In this method, we use knowledge distillation with instance-wise weighting mechanism to adaptively transfer the "dark" knowledge from the source hypothesis to the target domain.The weighting coefficients of the distillation loss and the standard loss are determined by the consistency between the predicted probability of the source hypothesis and the target ground-truth label.Empirical results on both transfer learning benchmark datasets and a healthcare dataset demonstrate the effectiveness of our method.
Open AccessDynamic Knowledge Distillation for Black-box Hypothesis Transfer Learning
In real world applications like healthcare, it is usually difficult to build a machine learning prediction model that works universally well across different institutions. At the same time, the available model is often proprietary, i.e., neither the model parameter nor the data set used for model training is accessible. In consequence, leveraging the knowledge hidden in the available model (aka. the hypothesis) and adapting it to a local data set becomes extremely challenging. Motivated by this situation, in this paper we aim to address such a specific case within the hypothesis transfer learning framework, in which 1) the source hypothesis is a black-box model and 2) the source domain data is unavailable. In particular, we introduce a novel algorithm called dynamic knowledge distillation for hypothesis transfer learning (dkdHTL). In this method, we use knowledge distillation with instance-wise weighting mechanism to adaptively transfer the "dark" knowledge from the source hypothesis to the target domain.The weighting coefficients of the distillation loss and the standard loss are determined by the consistency between the predicted probability of the source hypothesis and the target ground-truth label.Empirical results on both transfer learning benchmark datasets and a healthcare dataset demonstrate the effectiveness of our method.
- Open Access
- 22 May 2020
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 22 (5) , e15817
A Cross-Cultural Comparison of an Extended Planned Risk Information Seeking Model on Mental Health Among College Students: Cross-Sectional Study
Background: Approximately 42.5 million adults have been affected by mental illness in the United States in 2013, and 173 million people have been affected by a diagnosable psychiatric disorder in China. An increasing number of people tend to seek health information on the Web, and it is important to understand the factors associated with individuals’ mental health information seeking. Identifying factors associated with mental health information seeking may influence the disease progression of potential patients. Objective: This study aimed to test the planned risk information seeking model (PRISM) in China and the United States with a chronic disease, mental illness, and two additional factors, ie, media use and cultural identity, among college students. Methods: Data were collected in both countries using the same online survey through a survey management program (Qualtrics). In China, college instructors distributed the survey link among university students, and it was also posted on a leading social media site called Sina Weibo. In the United States, the data were collected in a college-wide survey pool in a large Northwestern university. Results: The final sample size was 235 for the Chinese sample and 241 for the US sample. Media use was significantly associated with mental health information–seeking intentions in the Chinese sample (P<.001), and cultural identity was significantly associated with intentions in both samples (China: P=.02; United States: P<.001). The extended PRISM had a better model fit than the original PRISM. Conclusions: Cultural identity and media use should be considered when evaluating the process of mental health information seeking or when designing interventions to address mental health information seeking.
- 22 May 2020
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 22 (5) , e15817
Open AccessA Cross-Cultural Comparison of an Extended Planned Risk Information Seeking Model on Mental Health Among College Students: Cross-Sectional Study
Background: Approximately 42.5 million adults have been affected by mental illness in the United States in 2013, and 173 million people have been affected by a diagnosable psychiatric disorder in China. An increasing number of people tend to seek health information on the Web, and it is important to understand the factors associated with individuals’ mental health information seeking. Identifying factors associated with mental health information seeking may influence the disease progression of potential patients. Objective: This study aimed to test the planned risk information seeking model (PRISM) in China and the United States with a chronic disease, mental illness, and two additional factors, ie, media use and cultural identity, among college students. Methods: Data were collected in both countries using the same online survey through a survey management program (Qualtrics). In China, college instructors distributed the survey link among university students, and it was also posted on a leading social media site called Sina Weibo. In the United States, the data were collected in a college-wide survey pool in a large Northwestern university. Results: The final sample size was 235 for the Chinese sample and 241 for the US sample. Media use was significantly associated with mental health information–seeking intentions in the Chinese sample (P<.001), and cultural identity was significantly associated with intentions in both samples (China: P=.02; United States: P<.001). The extended PRISM had a better model fit than the original PRISM. Conclusions: Cultural identity and media use should be considered when evaluating the process of mental health information seeking or when designing interventions to address mental health information seeking.
- Open Access
- 11 May 2020
- Published by JMIR Publications Inc. in JMIR Preprints
Chinese Public’s Engagement in Preventive and Intervening Health Behaviors During the Early Breakout of COVID-19: Cross-Sectional Study (Preprint)
BACKGROUND Since January 2020, the coronavirus disease (COVID-19) swept over China and then the world, causing a global public health crisis. People’s adoption of preventive and intervening behaviors is critical in curbing the spread of the virus. OBJECTIVE The aim of this study is to evaluate Chinese people’s adoption of health behaviors in responding to COVID-19 and to identify key determinants for their engagement. METHODS An anonymous online questionnaire was distributed in early February 2020 among Mainland Chinese (18 years or older) to examine their engagement in preventive behaviors (eg, frequent handwashing, wearing masks, staying at home) and intervening behaviors (eg, advising family to wash hands frequently), and to explore potential determinants for their adoption of these health behaviors. RESULTS Out of 2949 participants, 55.3% (n=1629) reported frequent engagement in preventive health behaviors, and over 84% (n=2493) performed at least one intervening health behavior. Greater engagement in preventive behaviors was found among participants who received higher education, were married, reported fewer barriers and greater benefits of engagement, reported greater self-efficacy and emotional support, had greater patient-centered communication before, had a greater media literacy level, and had greater new media and traditional media use for COVID-19 news. Greater engagement in intervening behaviors was observed among participants who were married, had lower income, reported greater benefits of health behaviors, had greater patient-centered communication before, had a lower media literacy level, and had a greater new media and traditional media use for COVID-19 news. CONCLUSIONS Participants’ engagement in coronavirus-related preventive and intervening behaviors was overall high, and the associations varied across demographic and psychosocial variables. Hence, customized health interventions that address the determinants for health behaviors are needed to improve people’s adherence to coronavirus-related behavior guidelines.
- 11 May 2020
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessChinese Public’s Engagement in Preventive and Intervening Health Behaviors During the Early Breakout of COVID-19: Cross-Sectional Study (Preprint)
BACKGROUND Since January 2020, the coronavirus disease (COVID-19) swept over China and then the world, causing a global public health crisis. People’s adoption of preventive and intervening behaviors is critical in curbing the spread of the virus. OBJECTIVE The aim of this study is to evaluate Chinese people’s adoption of health behaviors in responding to COVID-19 and to identify key determinants for their engagement. METHODS An anonymous online questionnaire was distributed in early February 2020 among Mainland Chinese (18 years or older) to examine their engagement in preventive behaviors (eg, frequent handwashing, wearing masks, staying at home) and intervening behaviors (eg, advising family to wash hands frequently), and to explore potential determinants for their adoption of these health behaviors. RESULTS Out of 2949 participants, 55.3% (n=1629) reported frequent engagement in preventive health behaviors, and over 84% (n=2493) performed at least one intervening health behavior. Greater engagement in preventive behaviors was found among participants who received higher education, were married, reported fewer barriers and greater benefits of engagement, reported greater self-efficacy and emotional support, had greater patient-centered communication before, had a greater media literacy level, and had greater new media and traditional media use for COVID-19 news. Greater engagement in intervening behaviors was observed among participants who were married, had lower income, reported greater benefits of health behaviors, had greater patient-centered communication before, had a lower media literacy level, and had a greater new media and traditional media use for COVID-19 news. CONCLUSIONS Participants’ engagement in coronavirus-related preventive and intervening behaviors was overall high, and the associations varied across demographic and psychosocial variables. Hence, customized health interventions that address the determinants for health behaviors are needed to improve people’s adherence to coronavirus-related behavior guidelines.
- 3 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 24 (9) , 2663-2670
Aceso: PICO-Guided Evidence Summarization on Medical Literature
Evidence-Based Medicine (EBM) aims to apply the best available evidence gained from scientific methods to clinical decision making. A generally accepted criterion to formulate evidence is to use the PICO framework, where PICO stands for Problem/Population, Intervention, Comparison, and Outcome. Automatic extraction of PICO-related sentences from medical literature is crucial to the success of many EBM applications. In this work, we present our Aceso system, which automatically generates PICO-based evidence summaries from medical literature. In Aceso $^{1}$ , we adopt an active learning paradigm, which helps to minimize the cost of manual labeling and to optimize the quality of summarization with limited labeled data. An UMLS2Vec model is proposed to learn a vector representation of medical concepts in UMLS $^{2}$ , and we fuse the embedding of medical knowledge with textual features in summarization. The evaluation shows that our approach is better on identifying PICO sentences against state-of-the-art studies and outperforms baseline methods on producing high-quality evidence summaries.- 3 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 24 (9) , 2663-2670
Aceso: PICO-Guided Evidence Summarization on Medical Literature
Evidence-Based Medicine (EBM) aims to apply the best available evidence gained from scientific methods to clinical decision making. A generally accepted criterion to formulate evidence is to use the PICO framework, where PICO stands for Problem/Population, Intervention, Comparison, and Outcome. Automatic extraction of PICO-related sentences from medical literature is crucial to the success of many EBM applications. In this work, we present our Aceso system, which automatically generates PICO-based evidence summaries from medical literature. In Aceso $^{1}$ , we adopt an active learning paradigm, which helps to minimize the cost of manual labeling and to optimize the quality of summarization with limited labeled data. An UMLS2Vec model is proposed to learn a vector representation of medical concepts in UMLS $^{2}$ , and we fuse the embedding of medical knowledge with textual features in summarization. The evaluation shows that our approach is better on identifying PICO sentences against state-of-the-art studies and outperforms baseline methods on producing high-quality evidence summaries.