Free-form Text Research Articles

Condensed Memory Networks for Clinical Diagnostic Inferencing

Diagnosis of a clinical condition is a challenging task, which often requires significant medical investigation. Previous work related to diagnostic inferencing problems mostly consider multivariate observational data (e.g. physiological signals, lab tests etc.). In contrast, we explore the problem using free-text medical notes recorded in an electronic health record (EHR). Complex tasks like these can benefit from structured knowledge bases, but those are not scalable. We instead exploit raw text from Wikipedia as a knowledge source. Memory networks have been demonstrated to be effective in tasks which require comprehension of free-form text. They use the final iteration of the learned representation to predict probable classes. We introduce condensed memory neural networks (C-MemNNs), a novel model with iterative condensation of memory representations that preserves the hierarchy of features in the memory. Experiments on the MIMIC-III dataset show that the proposed model outperforms other variants of memory networks to predict the most probable diagnoses given a complex clinical scenario.

The BioC-BioGRID corpus: full text articles annotated for curation of protein-protein and genetic interactions.

A great deal of information on the molecular genetics and biochemistry of model organisms has been reported in the scientific literature. However, this data is typically described in free text form and is not readily amenable to computational analyses. To this end, the BioGRID database systematically curates the biomedical literature for genetic and protein interaction data. This data is provided in a standardized computationally tractable format and includes structured annotation of experimental evidence. BioGRID curation necessarily involves substantial human effort by expert curators who must read each publication to extract the relevant information. Computational text-mining methods offer the potential to augment and accelerate manual curation. To facilitate the development of practical text-mining strategies, a new challenge was organized in BioCreative V for the BioC task, the collaborative Biocurator Assistant Task. This was a non-competitive, cooperative task in which the participants worked together to build BioC-compatible modules into an integrated pipeline to assist BioGRID curators. As an integral part of this task, a test collection of full text articles was developed that contained both biological entity annotations (gene/protein and organism/species) and molecular interaction annotations (protein–protein and genetic interactions (PPIs and GIs)). This collection, which we call the BioC-BioGRID corpus, was annotated by four BioGRID curators over three rounds of annotation and contains 120 full text articles curated in a dataset representing two major model organisms, namely budding yeast and human. The BioC-BioGRID corpus contains annotations for 6409 mentions of genes and their Entrez Gene IDs, 186 mentions of organism names and their NCBI Taxonomy IDs, 1867 mentions of PPIs and 701 annotations of PPI experimental evidence statements, 856 mentions of GIs and 399 annotations of GI evidence statements. The purpose, characteristics and possible future uses of the BioC-BioGRID corpus are detailed in this report.Database URL: http://bioc.sourceforge.net/BioC-BioGRID.html

Combination of Deep Recurrent Neural Networks and Conditional Random Fields for Extracting Adverse Drug Reactions from User Reviews.

Adverse drug reactions (ADRs) are an essential part of the analysis of drug use, measuring drug use benefits, and making policy decisions. Traditional channels for identifying ADRs are reliable but very slow and only produce a small amount of data. Text reviews, either on specialized web sites or in general-purpose social networks, may lead to a data source of unprecedented size, but identifying ADRs in free-form text is a challenging natural language processing problem. In this work, we propose a novel model for this problem, uniting recurrent neural architectures and conditional random fields. We evaluate our model with a comprehensive experimental study, showing improvements over state-of-the-art methods of ADR extraction.

Organization of User-Generated Information in Image Collections and Impact of Rhetorical Mechanisms

To collect information with crowdsourcing is a popular method for cultural heritage institutions. User comments in free-text format are especially propagated as empowering users and their influence ...

Pertinence of electronic death certificates for real-time surveillance and alert, France, 2012–2014

ObjectivesIn France, the early mortality monitoring, conducted by Santé publique France, the French National Public Health agency (SpFrance) (formerly French Institute for public health surveillance-InVS), is based on the administrative data provided by the National Institute for Statistic and Economic Studies (INSEE) and consequently does not allow analyses on medical causes of death. Since 2007, the physicians can certify deaths electronically. In this electronic system (Electronic Death Registration System; EDRS), the medical causes of death, in free-text format, are directly transmitted to SpFrance. In the future, these data could be used in a real-time surveillance system by medical causes of death. The objective of this study was to evaluate the pertinence of e-death certification using the following assessment criteria: timeliness, representativeness, and completeness of sociodemographic and medical information included in the e-death certificates. Study designThis study consisted of a descriptive analysis of the information collected by e-death certificates recorded between January 1, 2012 and July 31, 2014. MethodsThe study quantified the temporal and geographical evolution of the deployment of the EDRS between 2012 and 2014. The timeliness of the system was estimated by calculating the delay between the dates of death and of data availability for analysis. Sociodemographic and death-related characteristics were described. The frequency of missing data was measured for each variable. The number of completed fields per certificate and the number of words per field and per certificate were calculated for the medical causes of death. ResultsBetween January 2012 and July 2014, 77,776 e-death certificates were collected. A slight increase in the use of the e-death certification was observed during the study period, reaching 6.1% of the total number of deaths in 2014. Good national coverage was noted. Nearly 79% of e-certificates were submitted to SpFrance on the day of the death. We observed a high completeness of the e-certificates. The rate of missing data did not exceed 2.7% for sociodemographic variables. On average, 10 words, distributed in three fields, were used to describe the medical causes of death. ConclusionsE-death certificates constitute a reactive source of information on medical causes of death. The deployment of EDRS is of major public health interest for the development of a real-time warning surveillance system of mortality by cause.

Web-based Real-Time Case Finding for the Population Health Management of Patients With Diabetes Mellitus: A Prospective Validation of the Natural Language Processing-Based Algorithm With Statewide Electronic Medical Records.

BackgroundDiabetes case finding based on structured medical records does not fully identify diabetic patients whose medical histories related to diabetes are available in the form of free text. Manual chart reviews have been used but involve high labor costs and long latency.ObjectiveThis study developed and tested a Web-based diabetes case finding algorithm using both structured and unstructured electronic medical records (EMRs).MethodsThis study was based on the health information exchange (HIE) EMR database that covers almost all health facilities in the state of Maine, United States. Using narrative clinical notes, a Web-based natural language processing (NLP) case finding algorithm was retrospectively (July 1, 2012, to June 30, 2013) developed with a random subset of HIE-associated facilities, which was then blind tested with the remaining facilities. The NLP-based algorithm was subsequently integrated into the HIE database and validated prospectively (July 1, 2013, to June 30, 2014).ResultsOf the 935,891 patients in the prospective cohort, 64,168 diabetes cases were identified using diagnosis codes alone. Our NLP-based case finding algorithm prospectively found an additional 5756 uncodified cases (5756/64,168, 8.97% increase) with a positive predictive value of .90. Of the 21,720 diabetic patients identified by both methods, 6616 patients (6616/21,720, 30.46%) were identified by the NLP-based algorithm before a diabetes diagnosis was noted in the structured EMR (mean time difference = 48 days).ConclusionsThe online NLP algorithm was effective in identifying uncodified diabetes cases in real time, leading to a significant improvement in diabetes case finding. The successful integration of the NLP-based case finding algorithm into the Maine HIE database indicates a strong potential for application of this novel method to achieve a more complete ascertainment of diagnoses of diabetes mellitus.

Indulgent thinking? Ecological momentary assessment of overweight and healthy-weight participants' cognitions and emotions

Cognitions and emotions are considered important determinants of eating behaviour in cognitive behavioural models of obesity. Ecological data on these determinants is still limited. The present study investigated cognitions and emotions of overweight (n = 57) and healthy-weight (n = 43) participants via Ecological Momentary Assessment.It was found that eating-related cognitions mainly focused on desire and taste. Unexpectedly, dysfunctional cognitions (i.e., thoughts that may promote overeating) did not occur more often for overweight participants in almost all cases. So, the present EMA study provides no evidence for a role of dysfunctional cognitions in obesity-promoting eating behaviour when assessing eating-related cognitions immediately prior to eating events using a free-text format assessment.Right before eating events, participants mostly reported feeling calm/relaxed and cheerful/happy. Overweight participants scored higher on negative emotions, both at eating events and non-eating moments, than did healthy-weight participants. In addition, scores on standard questionnaires assessing emotional eating were positively associated with negative emotions reported at both eating and non-eating moments. As such, negative emotions, as assessed in the present study, do not seem to be specific triggers for food consumption.

Impact of an education intervention on nursing diagnoses in free-text format in electronic health records: A pretest–posttest study in a medical department at a university hospital

Discussions on how nursing documentation should be carried out have been ongoing for the last decade. In this study, free-text format for nursing diagnoses was introduced to nursing staff at a university hospital in Norway. The aim of the study was to investigate the impact of an education intervention introducing nursing diagnoses in a free-text format following a problem-etiology-symptom structure. A pretest–posttest design was performed to assess changes in quality and quantity in the nursing documentation using the audit instrument N-Catch II. Several elements in the nursing documentation had statistically significant changes; the largest was found for quantity in nursing diagnoses. Education interventions aimed to improve nurses’ documentation in the electronic health record may have an effect on more complete and accurate nursing documentation. The presented education intervention showed a significant impact of more accurate nursing diagnoses and significant improvements in nursing documentation.

HPC and the Big Data challenge

High performance computing (HPC) and Big Data are technologies vital for advancement in science, business and industry. HPC combines computing power of supercomputers and computer clusters, and parallel and distributed processing techniques for solving complex computational problems. The term Big Data refers to the fact that more data are being produced, consumed and stored than ever before. This is resulting in datasets that are too large, complex, and/or dynamic to be managed and analysed by traditional methods. Access to HPC systems and the ability to model, simulate and manipulate massive and dynamic data, is now critical for research, business and innovation. In this paper an overview of HPC and Big Data technology is presented. The paper outlines the advances in computer technology enabling Peta and Exa scale and energy efficient computing, and Big Data challenges of extracting meaning and new information from the data. As an example of HPC and Big Data synergy in risk analysis, a case study of processing close-call data is conducted using HPC resources at the University of Huddersfield. A parallel program was designed and implemented on the university's Hadoop cluster to speed up processing of unstructured free form text records pertaining to close call railway events, in order to identify potential risks and incidents. This case study demonstrates the benefits of using HPC with parallel programming techniques, and the improvements achieved compared to serial processing on a standard workstation computer system. However, it also highlights the challenges in risk analysis of Big Data that require novel approaches in HPC system and software design.

Cleaning historical maintenance work order data for reliability analysis

Purpose – The purpose of this paper is to identify quality issues with using historical work order (WO) data from computerised maintenance management systems for reliability analysis; and develop an efficient and transparent process to correct these data quality issues to ensure data is fit for purpose in a timely manner. Design/methodology/approach – This paper develops a rule-based approach to data cleansing and demonstrates the process on data for heavy mobile equipment from a number of organisations. Findings – Although historical WO records frequently contain missing or incorrect functional location, failure mode, maintenance action and WO status fields the authors demonstrate it is possible to make these records fit for purpose by using data in the freeform text fields; an understanding of the maintenance tactics and practices at the operation; and knowledge of where the asset is in its life cycle. The authors demonstrate that it is possible to have a repeatable and transparent process to deal with the data cleaning activities. Originality/value – How engineers deal with raw maintenance data and the decisions they make in order to produce a data set for reliability analysis is seldom discussed in detail. Assumptions and actions are often left undocumented. This paper describes typical data cleaning decisions we all have to make as a routine part of the analysis and presents a process to support the data cleaning decisions in a repeatable and transparent fashion.

An Open Library of CT Patient Projection Data.

Lack of access to projection data from patient CT scans is a major limitation for development and validation of new reconstruction algorithms. To meet this critical need, we are building a library of CT patient projection data in an open and vendor-neutral format, DICOM-CT-PD, which is an extended DICOM format that contains sinogram data, acquisition geometry, patient information, and pathology identification. The library consists of scans of various types, including head scans, chest scans, abdomen scans, electrocardiogram (ECG)-gated scans, and dual-energy scans. For each scan, three types of data are provided, including DICOM-CT-PD projection data at various dose levels, reconstructed CT images, and a free-form text file. Several instructional documents are provided to help the users extract information from DICOM-CT-PD files, including a dictionary file for the DICOM-CT-PD format, a DICOM-CT-PD reader, and a user manual. Radiologist detection performance based on the reconstructed CT images is also provided. So far 328 head cases, 228 chest cases, and 228 abdomen cases have been collected for potential inclusion. The final library will include a selection of 50 head, chest, and abdomen scans each from at least two different manufacturers, and a few ECG-gated scans and dual-source, dual-energy scans. It will be freely available to academic researchers, and is expected to greatly facilitate the development and validation of CT reconstruction algorithms.

Zero-Shot Event Detection by Multimodal Distributional Semantic Embedding of Videos

We propose a new zero-shot Event-Detection method by Multi-modal Distributional Semantic embedding of videos. Our model embeds object and action concepts as well as other available modalities from videos into a distributional semantic space. To our knowledge, this is the first Zero-Shot event detection model that is built on top of distributional semantics and extends it in the following directions: (a) semantic embedding of multimodal information in videos (with focus on the visual modalities), (b) semantic embedding of concepts definitions, and (c) retrieve videos by free text event query (e.g., "changing a vehicle tire") based on their content. We first embed the video into the multi-modal semantic space and then measure the similarity between videos with the event query in free text form. We validated our method on the large TRECVID MED (Multimedia Event Detection) challenge. Using only the event title as a query, our method outperformed the state-the-art that uses big descriptions from 12.6\% to 13.5\% with MAP metric and from 0.73 to 0.83 with ROC-AUC metric. It is also an order of magnitude faster.

Dynamic Concept Composition for Zero-Example Event Detection

In this paper, we focus on automatically detecting events in unconstrained videos without the use of any visual training exemplars. In principle, zero-shot learning makes it possible to train an event detection model based on the assumption that events (e.g. birthday party) can be described by multiple mid-level semantic concepts (e.g. ``blowing candle'', ``birthday cake''). Towards this goal, we first pre-train a bundle of concept classifiers using data from other sources. Then we evaluate the semantic correlation of each concept w.r.t. the event of interest and pick up the relevant concept classifiers, which are applied on all test videos to get multiple prediction score vectors. While most existing systems combine the predictions of the concept classifiers with fixed weights, we propose to learn the optimal weights of the concept classifiers for each testing video by exploring a set of online available videos with free-form text descriptions of their content. To validate the effectiveness of the proposed approach, we have conducted extensive experiments on the latest TRECVID MEDTest 2014, MEDTest 2013 and CCV dataset. The experimental results confirm the superiority of the proposed approach.

Standardised reporting of EEG

EEGs are traditionally described and reported in free text format, which contributes to the low inter-observer agreement and makes it impossible to save the extracted features in a database. Under the auspices of the European Chapter of the International Federation of Clinical Neurophysiology (IFCN) and the Commission on European Affairs of the International League Against Epilepsy, a working group of European EEG experts developed a consensus on the structure of the EEG report and the terms describing the clinically relevant EEG features. SCORE is the acronym for Standardised Computer-based Organised Reporting of EEG. Subsequently the consensus template was implemented in the SCORE software. The EEGer describes the findings in the report by choosing from the pre-defined terms in the software. Simultaneously the report is generated and the scored features are saved in the database. Currently an international taskforce under the auspices of IFCN is working on the revised version of SCORE. The standardised EEG reporting will improve the quality of EEGs in the clinical practice and it will be a powerful tool for education and research.

Natural Language Processing Technologies in Radiology Research and Clinical Applications.

The migration of imaging reports to electronic medical record systems holds great potential in terms of advancing radiology research and practice by leveraging the large volume of data continuously being updated, integrated, and shared. However, there are significant challenges as well, largely due to the heterogeneity of how these data are formatted. Indeed, although there is movement toward structured reporting in radiology (ie, hierarchically itemized reporting with use of standardized terminology), the majority of radiology reports remain unstructured and use free-form language. To effectively "mine" these large datasets for hypothesis testing, a robust strategy for extracting the necessary information is needed. Manual extraction of information is a time-consuming and often unmanageable task. "Intelligent" search engines that instead rely on natural language processing (NLP), a computer-based approach to analyzing free-form text or speech, can be used to automate this data mining task. The overall goal of NLP is to translate natural human language into a structured format (ie, a fixed collection of elements), each with a standardized set of choices for its value, that is easily manipulated by computer programs to (among other things) order into subcategories or query for the presence or absence of a finding. The authors review the fundamentals of NLP and describe various techniques that constitute NLP in radiology, along with some key applications.

Semantic representation of reported measurements in radiology.

BackgroundIn radiology, a vast amount of diverse data is generated, and unstructured reporting is standard. Hence, much useful information is trapped in free-text form, and often lost in translation and transmission. One relevant source of free-text data consists of reports covering the assessment of changes in tumor burden, which are needed for the evaluation of cancer treatment success. Any change of lesion size is a critical factor in follow-up examinations. It is difficult to retrieve specific information from unstructured reports and to compare them over time. Therefore, a prototype was implemented that demonstrates the structured representation of findings, allowing selective review in consecutive examinations and thus more efficient comparison over time.MethodsWe developed a semantic Model for Clinical Information (MCI) based on existing ontologies from the Open Biological and Biomedical Ontologies (OBO) library. MCI is used for the integrated representation of measured image findings and medical knowledge about the normal size of anatomical entities. An integrated view of the radiology findings is realized by a prototype implementation of a ReportViewer. Further, RECIST (Response Evaluation Criteria In Solid Tumors) guidelines are implemented by SPARQL queries on MCI. The evaluation is based on two data sets of German radiology reports: An oncologic data set consisting of 2584 reports on 377 lymphoma patients and a mixed data set consisting of 6007 reports on diverse medical and surgical patients. All measurement findings were automatically classified as abnormal/normal using formalized medical background knowledge, i.e., knowledge that has been encoded into an ontology. A radiologist evaluated 813 classifications as correct or incorrect. All unclassified findings were evaluated as incorrect.ResultsThe proposed approach allows the automatic classification of findings with an accuracy of 96.4 % for oncologic reports and 92.9 % for mixed reports. The ReportViewer permits efficient comparison of measured findings from consecutive examinations. The implementation of RECIST guidelines with SPARQL enhances the quality of the selection and comparison of target lesions as well as the corresponding treatment response evaluation.ConclusionsThe developed MCI enables an accurate integrated representation of reported measurements and medical knowledge. Thus, measurements can be automatically classified and integrated in different decision processes. The structured representation is suitable for improved integration of clinical findings during decision-making. The proposed ReportViewer provides a longitudinal overview of the measurements.

ANALYSIS OF SMS FEEDBACK AND ONLINE FEEDBACK USING SENTIMENT-ANALYSIS FOR ASSESSMENT OF TEACHING

In this, the system will collect SMS’s and online Feedback from students in free text format. To get instance response of lecturer feedback by analyzing free text messages using sentiment analysis. It is very low cost and plat form independent. It will show you three different outputs, i.e. Good, Bad and Average/Neutral. The result will be shown in graphical format.

MapAffil: A Bibliographic Tool for Mapping Author Affiliation Strings to Cities and Their Geocodes Worldwide.

Bibliographic records often contain author affiliations as free-form text strings. Ideally one would be able to automatically identify all affiliations referring to any particular country or city such as Saint Petersburg, Russia. That introduces several major linguistic challenges. For example, Saint Petersburg is ambiguous (it refers to multiple cities worldwide and can be part of a street address) and it has spelling variants (e.g., St. Petersburg, Sankt-Peterburg, and Leningrad, USSR). We have designed an algorithm that attempts to solve these types of problems. Key components of the algorithm include a set of 24,000 extracted city, state, and country names (and their variants plus geocodes) for candidate look-up, and a set of 1.1 million extracted word n-grams, each pointing to a unique country (or a US state) for disambiguation. When applied to a collection of 12.7 million affiliation strings listed in PubMed, ambiguity remained unresolved for only 0.1%. For the 4.2 million mappings to the USA, 97.7% were complete (included a city), 1.8% included a state but not a city, and 0.4% did not include a state. A random sample of 300 manually inspected cases yielded six incompletes, none incorrect, and one unresolved ambiguity. The remaining 293 (97.7%) cases were unambiguously mapped to the correct cities, better than all of the existing tools tested: GoPubMed got 279 (93.0%) and GeoMaker got 274 (91.3%) while MediaMeter CLIFF and Google Maps did worse. In summary, we find that incorrect assignments and unresolved ambiguities are rare (< 1%). The incompleteness rate is about 2%, mostly due to a lack of information, e.g. the affiliation simply says "University of Illinois" which can refer to one of five different campuses. A search interface called MapAffil has been developed at the University of Illinois in which the longitude and latitude of the geographical city-center is displayed when a city is identified. This not only helps improve geographic information retrieval but also enables global bibliometric studies of proximity, mobility, and other geo-linked data.

Synthesizing Java expressions from free-form queries

We present a new code assistance tool for integrated development environments. Our system accepts as input free-form queries containing a mixture of English and Java, and produces Java code expressions that take the query into account and respect syntax, types, and scoping rules of Java, as well as statistical usage patterns. In contrast to solutions based on code search, the results returned by our tool need not directly correspond to any previously seen code fragment. As part of our system we have constructed a probabilistic context free grammar for Java constructs and library invocations, as well as an algorithm that uses a customized natural language processing tool chain to extract information from free-form text queries. We present the results on a number of examples showing that our technique (1) often produces the expected code fragments, (2) tolerates much of the flexibility of natural language, and (3) can repair incorrect Java expressions that use, for example, the wrong syntax or missing arguments.

What motivates young physicians? - a qualitative analysis of the learning climate in specialist medical training.

BackgroundNot least the much-invoked shortage of physicians in the current and the next generation has resulted in a wide range of efforts to improve postgraduate medical training. This is also in the focus of the current healthcare policy debate. Furthermore, quality and scope of available postgraduate training are important locational advantages in the competition for medical doctors. This study investigates the preferences and concerns that German house officers (HOs) have about their current postgraduate training. It also highlights how HOs evaluate the quality of their current postgraduate training and the learning environment.MethodsHOs were asked to answer the question: “Which things are of capital importance to you personally in your medical training?”, using a free text format. The survey was conducted web based (Lime survey) and all data was anonymized. Summarizing qualitative analyses were performed using the software tool MaxQDA.ResultsA total of 255 HOs participated in this study (female: n = 129/50.6 %; male: n = 126/49.4 %; age: 32 + 6 years) associated with 17 different German hospitals and from four medical specialties. Ten categories were generated from a total of 366 free text answers: 1. methodology of learning (n = 66), 2. supervision (n = 66), 3. learning structure (n = 61), 4. teaching competence (n = 37), 5. dedication (n = 34), 6. work climate (n = 29), 7. feedback/communication (n = 22), 8. challenge/patient safety (n = 21), 9. time/resources (n = 17), 10. personal security/safety (n = 13).ConclusionsHOs want a reliable and curriculum-guided learning structure. Different studying techniques should be used with sufficient (time) resources available in a trusting and communicative learning environment. Competent and dedicated instructors are expected to give individual and specific feedback to the HOs on individual strengths and deficits. Instructors should develop educational concepts in cooperation with the HOs and at the same time avoid excessive demands on HOs or hazards to patients.