Aijuan Dong

Followers

Following

Co-authors

Public Views

Interests

Uploads

Papers by Aijuan Dong

YOLO-based Deep Learning to Automated Bacterial Colony Counting

Predictive Modeling and Deep Phenotyping of Obstructive Sleep Apnea and Associated Comorbidities through Natural Language Processing and Large Language Models

medRxiv (Cold Spring Harbor Laboratory), Apr 20, 2024

Obstructive Sleep Apnea (OSA) is a prevalent sleep disorder associated with serious health condit... more Obstructive Sleep Apnea (OSA) is a prevalent sleep disorder associated with serious health conditions. This project utilized large language models (LLMs) to develop lexicons for OSA sub-phenotypes. Our study found that LLMs can identify informative lexicons for OSA subphenotyping in simple patient cohorts, achieving wAUC scores of 0.9 or slightly higher. Among the six models studied, BioClinical BERT and BlueBERT outperformed the rest. Additionally, the developed lexicons exhibited some utility in predicting mortality risk (wAUC score of 0.86) and hospital readmission (wAUC score of 0.72). This work demonstrates the potential benefits of incorporating LLMs into healthcare. Data and Code Availability This paper uses the MIMIC-IV dataset (Johnson et al., 2023a), which is available on the PhysioNet repository (Johnson et al., 2023b). We plan to make the source code publicly available in the future. Institutional Review Board (IRB) This research does not require IRB approval.

Download

Ontology-driven annotation and access of educational video data

Automated Detection of Corona Cavities from SDO Images with YOLO

2021 IEEE Seventh International Conference on Multimedia Big Data (BigMM), 2021

Solar eruptions, such as filament eruptions and Corona Mass Ejections (CMEs), are primary drivers... more Solar eruptions, such as filament eruptions and Corona Mass Ejections (CMEs), are primary drivers of hazardous space weather that impacts the critical infrastructure on the Earth. Researchers now believe that corona cavities serve as launch pads for CMEs. In this study, we proposed, implemented, and evaluated a method for automated corona cavity detection using YOLOv5 framework. We applied our method to the data observed by the Solar Dynamics Observatory (SDO) during the period from 2010 to 2013. With 18, 512 x 512 images, we were able to build a model that achieved a precision rate of 0.964, a recall rate of 0.883, and a mAP@0.5 of 0.898. With GPU freely available on the Google Colaboratory, training the model takes about 40 minutes and the inference time per image is about 9 milliseconds. Experiments showed the proposed method is promising in real-time corona cavity detection, and potentially tracking CMEs and predicting space weather.

Detecting Key Poets and Communities by Constructing and Analyzing Tang Poet Social Networks

Proceedings of the 5th International Conference on Frontiers of Educational Technologies - ICFET 2019, 2019

Existing scholarship on Tang Poetry often focuses on linguistics or its cultural and social impac... more Existing scholarship on Tang Poetry often focuses on linguistics or its cultural and social impact of selected poets. In this paper, a framework was developed for constructing and analyzing Tang poet social networks. Specifically, we first built Tang poet social networks by inferring poet relationships embedded in Tang poems and then identified key poets and detected major sub-communities via social network analysis and visualization. The findings of this work, key poets and communities identified, will benefit humanities scholars and the framework developed can, with minor modifications, be reused by other scholars with similar research interests.

A Statistical Method for Constructing Tang Poet Social Networks

2020 IEEE 14th International Conference on Semantic Computing (ICSC), 2020

Existing scholarship on Tang Poetry often focuses on linguistics or the cultural and social impac... more Existing scholarship on Tang Poetry often focuses on linguistics or the cultural and social impact of selected poets. In this paper, a method was developed for constructing a Tang poet network based on natural language processing and statistical graphical learning. The proposed method used Tang poems from Quan Tang Shi as the data source and contained a few major steps: disambiguate poet names via China Biographical Database, build a document-poet matrix, and infer Tang poet networks via Poisson Graphical Lasso. The inferred Tang poet networks were then evaluated and discussed. The Tang poet networks constructed in this study will benefit humanity scholars and poetry lovers alike, and the method developed can, with minor modifications, be reused by other scholars with similar research interests in Chinese literature.

Deep Learning Enabled Predicting Modeling of Mortality of Diabetes Mellitus Patients

Proceedings of the Practice and Experience in Advanced Research Computing on Rise of the Machines (learning), 2019

Diabetes mellitus (DM) is a major public health concern that requires continuing medical care. It... more Diabetes mellitus (DM) is a major public health concern that requires continuing medical care. It is also a leading cause of other serious health complications associated with longer hospital stays and increased mortality rates. Fluctuation of blood glucose levels are easy to monitor. Physicians manage patients' blood glucose to prevent or slow the progress of diabetes. In this paper, the MIMIC-III data set is used to develop and train multiple models that aim to predict the mortality of DM patients. Our deep learning model of convolutional neural network produced a 0.885 AUC score, above all baseline models we constructed, which include decision trees, random forests, and fully connected neural networks. The inputs for each model were comprised of admission type, age, Elixhauser comorbidity score, blood glucose measurements, and blood glucose range. The results obtained from these models are valuable for physicians, patients, and insurance companies. By analyzing the features t...

Integrated Documentary Video Access for Education

Educational documentary videos play an important role in enriching learners' learning experience.... more Educational documentary videos play an important role in enriching learners' learning experience. However, due to unstructured and linear features of videos, documentary videos are much more difficult to access than text-based documents and have not been effectively utilized in education. We argue the effectiveness of documentary video access depends on how videos are annotated and what kinds of access mechanisms are supported. In this paper, we first propose an integrated multimedia annotation framework that explores temporal, spatial, and semantic relationships existing among various multimedia annotation data, then describe the process of building integrated documentary video learning materials based on this framework. To effectively utilize these integrated learning materials, we propose an integrated documentary video access platform that provides a hierarchical view of video content, presents information in multiple modes, augments video content by extracting relevant materials from external data sources and supports flexible search. To demonstrate our idea, we implement an experimental documentary video access platform. The platform receives positive comments during various demonstration sessions.

Download

A Parallel Attractor-tree Based Clustering Method

Data clustering methods have been proven to be a successful data mining technique. There are two ... more Data clustering methods have been proven to be a successful data mining technique. There are two kinds of clustering approaches: partitioning clustering and hierarchical clustering. Hierarchical clustering is more flexible than partitioning clustering. However, it is very expensive for large data sets. In this paper, we propose a parallel algorithm for attractor-tree based hierarchical clustering which is implemented using MPI (message passing interface). We run the parallel program on different numbers of machines and compared it with the sequential approach. Experiments show that our parallel approach speeds up the sequential clustering while producing a comparatively good clustering result.

Download

Ontology-Driven Annotation and Access of Presentation Video Data

The tremendous growth in video data calls for efficient and flexible access mechanisms. In this p... more The tremendous growth in video data calls for efficient and flexible access mechanisms. In this paper, we propose an ontology-driven framework for presentation video annotation and access. The goal is to integrate ontology into video systems to improve users’ video access experience. To realize ontology-driven video annotation, the first and foremost step is video segmentation. Current research in video segmentation has mainly focused on the visual and/or auditory modalities. In this paper, we investigate how to combine visual and textual information in the hierarchical segmentation of presentation video data. With average F-scores over 0.92, our experiments show that the proposed segmentation procedure is effective. After a video is segmented, video annotation data can be extracted. To extract annotation data from a video and its segments, and to organize them in a way that facilitates video access, we propose a multiontology based multimedia annotation model. In this model, a doma...

Download

Comparing Two Immersive Virtual Environments for Education

The Virtual Cell is a 3D multi-user educational game teaching principles of Cell Biology. The Geo... more

Feature Selection and Comparison in Medical Image Analysis

Automatic mammogram analysis is important in early breast cancer detection. In this paper, we pre... more Automatic mammogram analysis is important in early breast cancer detection. In this paper, we present a multiresolution approach to automated classification of mammograms using Gabor filters and compare texture-based features with area-based features in automatic mammographic analysis. Specifically, multi-resolution Gabor filters of different frequencies and orientations are used for texture study, while four statistical area-based features, i.e. mean, variance, skewness and krutosis, are used for intensity analysis. Experimental results show that Gabor filters are able to extract textual patterns of mammograms, statisticalbased feature selection and weighting can be used to further reduce the feature space without degrading the classification performance. Furthermore, feature comparison experiments show that texture-based features perform better than area-based features for automatic mammogram classification.

Download

Parallel Dynamic Fraud Detection on Market Basket Data

Exploring Instructional Computing Platforms for a Small Liberal Arts College: A Case Study with NoSQL

2016 International Conference on Computational Science and Computational Intelligence (CSCI), 2016

Large-scale computing technology, such as NoSQL database systems and cloud computing, has become ... more Large-scale computing technology, such as NoSQL database systems and cloud computing, has become an important knowledge area in computer science curriculum. In this study, we experimented and evaluated two instructional computing platforms—Amazon Elastic Compute Cloud (EC2) and Hood College Virtual Computing Lab (HOOD VCL)—for offering NoSQL database systems in a small liberal arts college. We tested four types of NoSQL systems: Redis, MongoDB, Cassandra, and Neo4j. With each system, students performed a few realistic lab exercises with either Amazon EC2 or Hood VCL. Based on student feedback, while Amazon EC2 and Hood VCL are comparable to students with respect to accessibility, easy setup and maintenance, and reliability, survey results show a majority of students (86%) are more likely to use Amazon EC2 for similar data processing course work in the future. In addition, 64% of students believe Amazon Web Services (AWS) are cost-effective and 82% of them believe that the AWS experi...

Parallel Implementation of Hybrid Clustering

Dmin, 2008

Integrated Documentary Video Access for Education

Computers and Their Applications, 2006

Ontology-Driven Annotation and Access of Presentation Video Data

Download

Educational documentary video segmentation and access through combination of visual, audio and text understanding

Proceedings of the Fifth IEEE International Symposium on Signal Processing and Information Technology, 2005., 2000

Educational documentary videos play an important role in enriching learning experience. However, ... more Educational documentary videos play an important role in enriching learning experience. However, due to unstructured and linear features, documentary videos are much more difficult to access than text-based documents and have not been effectively utilized. In this paper, we propose a multimodal, hierarchical documentary video segmentation procedure based on image, audio and text understanding. The coincidence of scene-level audio breaks

Detection of breast tumor candidates using marker-controlled watershed segmentation and morphological analysis

2012 IEEE Southwest Symposium on Image Analysis and Interpretation, 2012

Computer Aided Diagnosis (CAD) was approved to automate breast cancer detection with mammograms i... more Computer Aided Diagnosis (CAD) was approved to automate breast cancer detection with mammograms in 1998. But due to the great variability in tumor sizes and shapes, and underlying breast tissue structures, pattern recognition algorithms have a difficult time adapting to different situations. In this paper, a marker-controlled watershed segmentation algorithm was developed to locate breast mass tumor candidates. The approach

Ontology- Based Information Integration in Virtual Learning Environment

The 2005 IEEE/WIC/ACM International Conference on Web Intelligence (WI'05), 2005

Abstract A good virtual learning environment should deliver relevant learning materials to learne... more

Aijuan Dong

Uploads

Papers by Aijuan Dong

Log In