... IEEE-EMBS International Conference on Biomedical and Health Informatics. IEEE-EMBS International Conference on Biomedical and Health Informatics最新文献

The Centers for Disease Control estimate that 1.6 to 3.8 million concussions occur in sports and recreational activities annually. Studies have shown that concussions increase the risk of future injuries and mild cognitive disorders. Despite extensive research on sports related concussion risk factors, the factors which are most predictive of concussion outcome and recovery time course remain unknown. In order to overcome the issue of physician bias and to identify the factors which can best predict concussion diagnosis, we propose a multi-variate logistic regression based analysis. We demonstrate our results on a dataset with 126 subjects (ages 12-31). Our results indicate that among 322 features, our model selected 27-29 features which included a history of playing sports, history of a previous concussion, drowsiness, nausea, trouble focusing as measured by a common symptom list, and oculomotor function. The features picked using our model were found to be highly predictive of concussions and gave a prediction performance accuracy greater than 90%, Matthews correlation coefficient greater than 0.8 and the area under the curve greater than 0.95.

Optical Endomicroscopy (OE) is a newly-emerged biomedical imaging modality that can help physicians make real-time clinical decisions about patients' grade of dysplasia. However, the performance of applying medical imaging classification for computer-aided diagnosis is primarily limited by the lack of labeled images. To improve the classification performance, we propose a semi-supervised learning algorithm that can incorporate large sets of unlabeled images. Our real-world endo-microscopic imaging datasets consist of 425 labeled images and 2,826 unlabeled ones. With semi-supervised learning algorithms, we improved multi-class classification performance over supervised learning algorithms by around 10% in all evaluation metrics, namely precision, recall, F1 score and Cohen-Kappa statistics.

One pressing need in the area of public health is timely, accurate, and complete reporting of deaths and the conditions leading up to them. Fast Healthcare Interoperability Resources (FHIR) is a new HL7 interoperability standard for electronic health record (EHR), while Sustainable Medical Applications and Reusable Technologies (SMART)-on-FHIR enables third-party app development that can work "out of the box". This research demonstrates the feasibility of developing SMART-on-FHIR applications to enable medical professionals to perform timely and accurate death reporting within multiple different jurisdictions of US. We explored how the information on a standard certificate of death can be mapped to resources defined in the FHIR standard (DSTU2). We also demonstrated analytics for potentially improving the accuracy and completeness of mortality reporting data.

With the increasing use of digital technologies, online social networks are emerging as major means of communication. Recently, social networks such as Facebook and Twitter are also being used by consumers, care providers (physicians, hospitals), and government agencies to share health related information. The asymmetric user network and the short message size have made Twitter particularly popular for propagating health related content on the Web. Besides tweeting on their own, users can choose to retweet particular tweets from other users (even if they do not follow them on Twitter.) Thus, a tweet can diffuse through the Twitter network via the follower-friend connections. In this paper, we report results of a pilot study we conducted to quantitatively assess how health related tweets diffuse in the directed follower-friend Twitter graph through the retweeting activity. Our effort includes (1). development of a retweet collection and Twitter retweet graph formation framework and (2). a preliminary analysis of retweet graphs and associated diffusion metrics for health tweets. Given the ambiguous nature (due to polysemy and sarcasm) of health relatedness of tweets collected with keyword based matches, our initial study is limited to ≈ 200 health related tweets (which were manually verified to be on health topics) each with at least 25 retweets. To our knowledge, this is first attempt to study health information diffusion on Twitter through retweet graph analysis.

Among critically-ill patients, hypotension represents a failure in compensatory mechanisms and may lead to organ hypoperfusion and failure. In this work, we adopt a data-driven approach for phenotype discovery and visualization of patient similarity and cohort structure in the intensive care unit (ICU). We used Hierarchical Dirichlet Process (HDP) as a nonparametric topic modeling technique to automatically learn a d-dimensional feature representation of patients that captures the latent "topic" structure of diseases, symptoms, medications, and findings documented in hospital discharge summaries. We then used the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm to convert the d-dimensional latent structure learned from HDP into a matrix of pairwise similarities for visualizing patient similarity and cohort structure. Using discharge summaries of a large patient cohort from the MIMIC II database, we evaluated the clinical utility of the discovered topic structure in phenotyping critically-ill patients who experienced hypotensive episodes. Our results indicate that the approach is able to reveal clinically interpretable clustering structure within our cohort and may potentially provide valuable insights to better understand the association between disease phenotypes and outcomes.

Standards for scoring adverse effects after radiation therapy (RT) is crucial for integrated, consistent, and accurate analysis of toxicity results at large scale and across multiple studies. This project aims to investigate the usage of the three most commonly used standards in published RT clinical studies by developing a text-mining based analysis method. We develop and compare two text-mining methods, one based on regular expressions and one based on Naïve Bayes Classifier, to analyze published full articles in terms of their adoption of standards in RT. The full dataset includes published articles identified in MEDLINE between January 2010 and August 2015. A radiation oncology physician reviewed all the articles in the training/validation subset and produced the usage trending data manually as gold standard for validation. The regular-expression based method reported classifications and overall usage trends that are comparable to those of the domain expert. The CTCAE standard is becoming the overall most commonly used standards over time, but the pace of adoption seems very slow. Further examination of the results indicates that the usage vary by disease type. It suggests that further efforts are needed to improve and harmonize the standards for adverse effects scoring in RT research community.

Modern intensive care units (ICUs) collect large volumes of data in monitoring critically ill patients. Clinicians in the ICUs face the challenge of interpreting large volumes of high-dimensional data to diagnose and treat patients. In this work, we explore the use of Hierarchical Dirichlet Processes (HDP) as a Bayesian nonparametric framework to infer patients' states of health by combining multiple sources of data. In particular, we employ HDP to combine clinical time series and text from the nursing progress notes in a probabilistic topic modeling framework for patient risk stratification. Given a patient cohort, we use HDP to infer latent "topics" shared across multimodal patient data from the entire cohort. Each topic is modeled as a multinomial distribution over a vocabulary of codewords, defined over heterogeneous data sources. We evaluate the clinical utility of the learned topic structure using the first 24-hour ICU data from over 17,000 adult patients in the MIMIC-II database to estimate patients' risks of in-hospital mortality. Our results demonstrate that our approach provides a viable framework for combining different data modalities to model patient's states of health, and can potentially be used to generate alerts to identify patients at high risk of hospital mortality.

Cardiac allograft rejection is one major limitation for long-term survival for patients with heart transplants. The endomyocardial biopsy is one gold standard to screen heart rejection for patients that have heart transplantation. However, manual identification of heart rejection is expensive and time-consuming. With the development of imaging processing techniques and machine learning tools, automatic prediction of heart rejection using whole-slide images is one promising approach to improve the care of patients with heart transplants. In this paper, we first develop a histopathological whole-slide image processing pipeline to extract features automatically. Then, we construct deep neural networks with and without regularization and dropout to classify the patients into nonrejection and rejection respectively. Our results show that neural networks with regularization and dropout can significantly reduce overfitting and achieve more stable accuracies.

Statistical methods have been widely used in studies of public health. Although useful in clinical research and public health policy making, these methods could not find correlation among health conditions automatically, or capture the temporal evolution of causes of death correctly. To cope with two challenges above, we implement an unsupervised machine learning model, termed topic models, to investigate the mortality data of the United States. Our model successfully groups morbidities based on their correlation, and reveals the temporal evolution of these groups from 1999 to 2014, which are also validated by existing literature. This work could provide a novel view for clinical practitioners to provide more accurate healthcare service, and for public health policymakers to make better policy.

Inadequate sleep affects health in multiple ways. Unobtrusive ambulatory methods to monitor long-term sleep patterns in large populations could be useful for health and policy decisions. This paper presents an algorithm that uses multimodal data from smartphones and wearable technologies to detect sleep/wake state and sleep episode on/offset. We collected 5580 days of multimodal data and applied recurrent neural networks for sleep/wake classification, followed by cross-correlation-based template matching for sleep episode on/offset detection. The method achieved a sleep/wake classification accuracy of 96.5%, and sleep episode on/offset detection F1 scores of 0.85 and 0.82, respectively, with mean errors of 5.3 and 5.5 min, respectively, when compared with sleep/wake state and sleep episode on/offset assessed using actigraphy and sleep diaries.