Data Management and Analytics for Medicine and Healthcare

Drug-induced liver injury (DILI) is an uncommon but important and challenging adverse drug event developed following the use of drugs, both prescription and over-the-counter. Early detection of DILI cases can greatly improve the patient care as discontinuing the offending drugs is essential for the care of DILI cases. An online resource, LiverTox, has been established to provide up-to-date, comprehensive clinical information on DILI in the form of case reports. In this study, we explored the use of the Human Phenotype Ontology (HPO) to annotate case reports with HPO terms and to predict a ranked list of phenotype categories (describing patient outcomes) that is most closely matched to the HPO annotations that are attached to the case report. The prediction performance based on our method was found to be good to excellent for 67% of case reports included in this study, i.e., the phenotype category that was assigned to the report was among the Top 3 predicted phenotype category descriptions. Future directions would be to incorporate other annotations, laboratory findings, and the exploration of other semantic-based methods for case report retrieval and ranking.

In clinical decision making, information on the treatment of patients that show similar medical conditions and symptoms to the current case, is one of most relevant information sources to create a good, evidence-based treatment plan. However, the retrieval of similar cases is still challenging and automatic support is missing. The reasons are two-fold: First, the query formulation is difficult since multiple criteria need to be selected and specified in short query phrases. Second, the discrete storage of multimedia patient records makes the retrieval and summary of a patient history extremely difficult. In this paper, we present a retrieval system for electronic health records (EHR). More specifically, a retrieval platform for EHRs for supporting clinical decision making in treatment of cervical spine defects with the information extracted from textual data of patient records is implemented as prototype. The patient cases are classified according to cervical spine defect classes, while the classification relies upon rules obtained from the corresponding defect classification schema and guidelines. In a retrospective study, the classifier is applied to clinical documents and the classification results are evaluated.

Microscopic image analysis of nuclei in pathology images generates tremendous amount of spatially derived data to support biomedical research and potential diagnosis. Such spatial data can be managed by traditional SQL based spatial databases and queried by SQL for spatial relationships. However, traditional spatial databases are designed for structured data with limited expressibility, which is difficult to support queries for complex visual patterns. Moreover, SQL based queries are not intuitive for biomedical researchers or pathologists.

In this paper, we investigate the expressive power of visual query for spatial databases and propose an effective yet general Iterative Example-based Visual Query (IEVQuery) framework to query shapes and distributions. More specifically, we extract features from nuclei in pathology databases, such as shape polygon nuclei density distribution, and nuclei growth directions to build search indexes. The user employs visual interactions such as sketching to input queries for interesting patterns. Meanwhile, the user is allowed to iteratively create queries, which are based on previous search results, to finely tune the features more accurately to find preferred results. We build a system to enable users to specify sketch based queries interactively for (1) nuclei shapes, (2) nuclei densities, and (3) nuclei growth directions. To validate our methods, we take a pathology database [11] consisting of hundreds of millions of nuclei, and enable the user to search in the database to find most matching results through our system.

In the health care industry, DICOM (Digital Imaging and Communication in Medicine) standard has become very popular for storage and transmission of digital medical images and reports. The ever-increasing size, high velocity and variety of the DICOM data collections make them more and more inefficient to be stored and queried them using a single data storage technique, e.g., a row store or a column store. In this study, we first highlight challenges in DICOM data management. We then describe HYTORMO, a new model to store and query the DICOM data. HYTORMO uses a hybrid data storage strategy that is aimed not only to leverage the advantage of both row and column stores, but also to attempt to keep a trade-off among reducing disk I/O cost, reducing tuple construction cost and reducing storage space. In addition, Bloom filters are applied to reduce network I/O cost during query processing. We prototyped our model on the top of Spark. Our preliminary experiments validate the proposed model in real DICOM datasets and show the effectiveness of our method.

Systematic analysis of high resolution whole slide images enables more effective diagnosis, prognosis and prediction of cancer and other important diseases. Due to the enormous sizes and dimensions of whole slide images, the analysis requires extensive computing resources which are not commonly available. Images have to be divided into smaller regions for processing due to computer memory limitations, which will lead to inaccurate results due to the ignorance of boundary crossing objects. In this paper, we propose a highly scalable and cost effective MapReduce based image analysis framework for whole slide image processing, and provide a cloud based implementation. The framework takes a grid-based overlapping partitioning scheme, and provides parallelization of image segmentation based on MapReduce. It provides graceful handling of boundary objects with a highly efficient spatial indexing based matching method, thus avoiding loss of accuracy due to partitioning. We demonstrate that the system achieves high scalability and is cost-effective – our experiments demonstrate that it costs less than fifteen cents to analyze one image on average using Amazon Elastic MapReduce.

Cancer is a disease of biological and cell cycle processes, driven by dosage of the limited set of drugs, resistance, mutations, and side effects. The identification of such limited set of drugs and their targets, pathways, and effects based on large scale multi-omics, multi-dimensional datasets is one of key challenging tasks in data-driven cancer genomics. This paper demonstrates the use of public databases associated with Drug-Target(Gene/Protein)-Disease to dissect the in-depth analysis of approved cancer drugs, their genetic associations, their pathways to establish a dosage balancing mechanism. This paper will also help to understand cancer as a disease associated pathways and effect of drug treatment on the cancer cells. We employ the Semantic Web approach to provide an integrated knowledge discovery process and the network of integrated datasets. The approach is employed to sustain the biological questions involving (1) Associated drugs and their omics signature, (2) Identification of gene association with integrated Drug-Target databases (3) Mutations, variants, and alterations from these targets (4) Their PPI Interactions and associated oncogenic pathways (5) Associated biological process aligned with these mutations and pathways to identify IC-50 level of each drug along-with adverse events and alternate indications. In principal this large semantically integrated database of around 30 databases will serve as Semantic Linked Association Prediction in drug discovery to explore and expand the dosage balancing and drug re-purposing.

Medical research use cases are population centric, unlike the clinical use cases which are patient or individual centric. Hence the research use cases require accessing medical archives and data source repositories of heterogeneous nature. Traditionally, in order to query data from these data sources, users manually access and download parts or whole of the data sources. The existing solutions tend to focus on a specific data format or storage, which prevents using them for a more generic research scenario with heterogeneous data sources where the user may not have the knowledge of the schema of the data a priori.

In this paper, we propose and discuss the design, implementation, and evaluation of Data Café, a scalable distributed architecture that aims to address the shortcomings in the existing approaches. Data Café lets the resource providers create biomedical data lakes from various data sources, and lets the research data users consume the data lakes efficiently and quickly without having a priori knowledge of the data schema.

In this study, we describe our preliminary efforts in building an i2b2-based integrated data repository that supports centralized data management for ovarian cancer clinical research, and discuss important lessons learnt that would inspire the evaluation and enhancement for future generic cancer-specific data repository. We collected multiple types of heterogeneous clinical data, including demographic, outcome, chemo-treatment and lab-test information for ovarian cancer. To better integrate different data types, we conducted data normalization procedures through reusing standard codes and creating mappings between local codes and standard vocabularies. We also developed the extract, transform and load (ETL) scripts to load the data into an i2b2 instance. Through further analytic practices, we evaluated major expectations of the systems according to common clinical research needs, including cohort query and identification, clinical data-based hypothesis-testing, and exploratory data-mining. We also identified and discussed outstanding issues we will address through additional enhancement of existing i2b2 system.

Every hospital uses their own format for creating their information system for storing patient’s data. This doesn’t allow hospitals to exchange patient data. Hence, there is requirement for standardized Hospital Information System (HIS). Also the HIS should be able to incorporate semantic interoperability. With technological advancement, clinicians and patients should get themselves involved in using the Electronic Health Records (EHRs). The current research provides roadmap for the introduction of domain specific clinical application following openEHR standards based on Multi-Level Modeling. Standardization will help in reducing cost and medical errors as well enhancing data quality. The current study focuses on (1) advantage of EHRs, (2) the need for standardization to improve quality of health records, thereby establishing interoperability among hospitals, (3) recognizing the use of archetypes for knowledge-based systems, (4) proposing framework for standardization, and (5) comparison of proposed approach with current HIS.