Hang Wu

Detection of blur artifacts in histopathological whole-slide images of endomyocardial biopsies

Histopathological whole-slide images (WSIs) have emerged as an objective and quantitative means for image-based disease diagnosis. However, WSIs may contain acquisition artifacts that affect downstream image feature extraction and quantitative disease diagnosis. We develop a method for detecting blur artifacts in WSIs using distributions of local blur metrics. As features, these distributions enable accurate classification of WSI regions as sharp or blurry. We evaluate our method using over 1000 portions of an endomyocardial biopsy (EMB) WSI. Results indicate that local blur metrics accurately detect blurry image regions.

Intelligent mortality reporting with FHIR

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.

Automated risk prediction for esophageal optical endomicroscopic images

Biomedical in vivo imaging has been playing an essential role in diagnoses and treatment in modern medicine. However, compared with the fast development of medical imaging systems, the medical imaging informatics, especially automated prediction, has not been fully explored. In our paper, we compared different feature extraction and classification methods for prediction pipeline to analyze in vivo endomicroscopic images, obtained from patients who are at risks for the development of gastric disease, esophageal adenocarcionoma. Extensive experiment results show that the selected feature representation and prediction algorithms achieved high accuracy in both binary and multi-class prediction tasks.

Post-surgical complication prediction in the presence of low-rank missing data.

The problem of missing data has made it difficult to analyze Electronic Health Records (EHR). In EHR data, the “missingness” often results from the low-rank property: each patient is considered a mixture of prototypical patients, and certain types of patients will have similar missing entries in their records. However, most existing methods to deal with missing data fail to capture this low-rank property of missing data. Hence we propose to use matrix factorization and matrix completion methods to perform prediction in the presence of missing data. We validated our methods in the task of post-surgical complication prediction and experimental results show that our method can improve the prediction accuracy significantly.

Improving Validity of Cause of Death on Death Certificates

Accurate reporting of causes of death on death certificates is essential to formulate appropriate disease control, prevention and emergency response by national health-protection institutions such as Center for disease prevention and control (CDC). In this study, we utilize knowledge from publicly available expert-formulated rules for the cause of death to determine the extent of discordance in the death certificates in national mortality data with the expert knowledge base. We also report the most commonly occurring invalid causal pairs which physicians put in the death certificates. We use sequence rule mining to find patterns that are most frequent on death certificates and compare them with the rules from the expert knowledge based. Based on our results, 20.1% of the common patterns derived from entries into death certificates were discordant. The most probable causes of these discordance or invalid rules are missing steps and non-specific ICD-10 codes on the death certificates.

Causes of death in the United States, 1999 to 2014

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 the unsupervised machine learning method “topic model” to study the United States death reporting data. Our model successfully groups morbidities based on their correlation, and reveals the temporal evolution of these groups from 1999 to 2014. This result is validated by existing literature, and provides a novel view that enables clinical practitioners to make more accurate healthcare decisions, and public health policymakers to make better policy.

Anticoagulation manager: Development of a clinical decision support mobile application for management of anticoagulants

Patients with certain clotting disorders or conditions have a greater risk of developing arterial or venous clots and downstream embolisms, strokes, and arterial insufficiency. These patients need prescription anticoagulant drugs to reduce the possibility of clot formation. However, historically, the clinical decision making workflow in determining the correct type and dosage of anticoagulant(s) is part science and part art. To address this problem, we developed Anticoagulation Manager, an intelligent clinical decision workflow management system on iOS-based mobile devices to help clinicians effectively choose the most appropriate and helpful follow-up clotting tests for patients with a common clotting profile. The app can provide physicians guidance to prescribe the most appropriate medication for patients in need of anticoagulant drugs. This intelligent app was jointly designed and developed by medical professionals in CDC and engineers at Georgia Tech, and will be evaluated by physicians for ease-of-use, robustness, flexibility, and scalability. Eventually, it will be deployed and shared in both physician community and developer community.