Pawel Matykiewicz

A shared task involving multi-label classification of clinical free text

This paper reports on a shared task involving the assignment of ICD-9-CM codes to radiology reports. Two features distinguished this task from previous shared tasks in the biomedical domain. One is that it resulted in the first freely distributable corpus of fully anonymized clinical text. This resource is permanently available and will (we hope) facilitate future research. The other key feature of the task is that it required categorization with respect to a large and commercially significant set of labels. The number of participants was larger than in any previous biomedical challenge task. We describe the data production process and the evaluation measures, and give a preliminary analysis of the results. Many systems performed at levels approaching the inter-coder agreement, suggesting that human-like performance on this task is within the reach of currently available technologies.

Sentiment Analysis of Suicide Notes: A Shared Task.

This paper reports on a shared task involving the assignment of emotions to suicide notes. Two features distinguished this task from previous shared tasks in the biomedical domain. One is that it resulted in the corpus of fully anonymized clinical text and annotated suicide notes. This resource is permanently available and will (we hope) facilitate future research. The other key feature of the task is that it required categorization with respect to a large set of labels. The number of participants was larger than in any previous biomedical challenge task. We describe the data production process and the evaluation measures, and give a preliminary analysis of the results. Many systems performed at levels approaching the inter-coder agreement, suggesting that human-like performance on this task is within the reach of currently available technologies.

Clustering Semantic Spaces of Suicide Notes and Newsgroups Articles.

Gas contaminants generated from the grease of bearings operating in a magnetic disk device adhere to the magnetic head and causing stiction and magnetic head crash. Such bearings are typically spindle bearings and pivot bearings. In operation, heat is generated that causes the bearing grease to generate the gases. By heat treating the grease, slow evaporating substances of hydrocarbon compounds with carbons of C9 to C13 in the grease are reduced to not more than 0.001 wt. % of the grease. The grease can then be used to lubricate the bearings. The heat treatment is performed at a temperature of 60 DEG to 100 DEG C. at a pressure of 10-3 to 10-8 torr. The assembled, lubricated bearings can be heat treated separately from the magnetic disk device or the heat treatment can take place after the bearings have been assembled in the magnetic disk device.

Using Natural Language Processing to Classify Suicide Notes

We hypothesize that machine-learning algorithms (MLA) could classify completer and ideator suicide notes as well a mental health professionals (MHP). Five MHPs classified 66 notes as either ideator or completer; machine learning algorithms (MLA) were used for the same task. Results: MHPs were accurate 71% of the time; the SMO algorithm was accurate 79% of the time. This is an important first step in developing an evidence based suicide predictor for emergency department use. Language: en

Referral to the Emergency Department by a Primary Care Provider Predicts Severity of Illness

OBJECTIVE: The purpose of this study was to assess whether referral to a pediatric emergency department (PED) by a primary care provider was associated with greater severity of illness, as determined on the basis of clinical measures and increased resource utilization. METHODS: A retrospective study of data for 121 088 children who presented to a PED with abdominal pain, fever, or respiratory complaints during a 5-year period was performed. Demographic data, referral status, and proxy markers of illness severity were collected from the medical records and analyzed. RESULTS: A total of 26.3% of all patients seen in the PED presented with these 3 complaint categories. With adjustment for age, gender, race, and insurance class, referred patients were significantly more likely to have high triage acuity designations, higher rates of very abnormal vital signs, and higher admission rates, compared with patients who were self-referred. Referred patients were more likely to undergo testing (laboratory or radiologic), to receive intravenous fluid therapy and pain medications, and to be assigned higher-severity discharge diagnoses, such as appendicitis, septic shock, or status asthmaticus. CONCLUSIONS: Referral by a primary care provider to a PED was significantly and independently associated with greater severity of illness and resource utilization. Referral status should be considered in algorithms used to triage cases for evaluation in the PED.

Nonambiguous concept mapping in medical domain

Automatic annotation of medical texts for various natural language processing tasks is a very important goal that is still far from being accomplished. Semantic annotation of a free text is one of the necessary steps in this process. Disambiguation is frequently attempted using either rule-based or statistical approaches to semantical analysis. A neurocognitive approach for a nonambiguous concept mapping is proposed here. Concepts are taken from the Unified Medical Language System (UMLS) collection of ontologies. An active part of the whole semantic memory based on these concepts forms a graph of consistent concepts (GCC). The text is analyzed by spreading activation in the network that consist of GCC and related concepts in the semantic network. A scoring function is used for choosing the meaning of the concepts that fit in the best way to the current interpretation of the text. ULMS knowledge sources are not sufficient to fully characterize concepts and their relations. Annotated texts are used to learn new relations useful for disambiguation of word meanings.

Towards understanding of natural language: neurocognitive inspirations

Neurocognitive processes responsible for representation of meaning and understanding of words are investigated. First a review of current knowledge about word representation, recent experiments linking it to associative memory and to right hemisphere synchronous activity is presented. Various conjectures on how meaning arises and how reasoning and problem solving is done are presented. These inspirations are used to make systematic approximation to spreading activation in semantic memory networks. Using hierarchical ontologies representations of short texts are enhanced and it is shown that highdimensional vector models may be treated as a snapshot approximation of the neural activity. Clustering short medical texts into different categories is greatly enhanced by this process, thus facilitating understanding of the text.

Assessing the similarity of surface linguistic features related to epilepsy across pediatric hospitals

Objective The constant progress in computational linguistic methods provides amazing opportunities for discovering information in clinical text and enables the clinical scientist to explore novel approaches to care. However, these new approaches need evaluation. We describe an automated system to compare descriptions of epilepsy patients at three different organizations: Cincinnati Children’s Hospital, the Children’s Hospital Colorado, and the Children’s Hospital of Philadelphia. To our knowledge, there have been no similar previous studies. Materials and methods In this work, a support vector machine (SVM)-based natural language processing (NLP) algorithm is trained to classify epilepsy progress notes as belonging to a patient with a specific type of epilepsy from a particular hospital. The same SVM is then used to classify notes from another hospital. Our null hypothesis is that an NLP algorithm cannot be trained using epilepsy-specific notes from one hospital and subsequently used to classify notes from another hospital better than a random baseline classifier. The hypothesis is tested using epilepsy progress notes from the three hospitals. Results We are able to reject the null hypothesis at the 95% level. It is also found that classification was improved by including notes from a second hospital in the SVM training sample. Discussion and conclusion With a reasonably uniform epilepsy vocabulary and an NLP-based algorithm able to use this uniformity to classify epilepsy progress notes across different hospitals, we can pursue automated comparisons of patient conditions, treatments, and diagnoses across different healthcare settings.

Selecting anti-epileptic drugs: a pediatric epileptologist's view, a computer's view

To identify which clinical characteristics are important to include in clinical decision support systems developed for Antiepileptic Drug (AEDs) selection.

Multiple Inheritance Problem in Semantic Spreading Activation Networks

Semantic networks inspired by semantic information processing by the brain frequently do not improve the results of text classification. This counterintuitive fact is explained here by the multiple inheritance problem, which corrupts real-world knowledge representation attempts. After a review of early work on the use of semantic networks in text classification, our own heuristic solution to the problem is presented. Significance testing is used to contrast results obtained with pruned and entire semantic networks applied to medical text classification problems. The algorithm has been motivated by the process of spreading neural activation in the brain. The semantic network activation is propagated throughout the network until no more changes to the text representation are detected. Solving the multiple inheritance problem for the purpose of text classification is similar to embedding inhibition in the spreading activation process – a crucial mechanism for a healthy brain.