Jimison Iavindrasana

Computation of semantic distance between adverse drug reactions terms may be an efficient way to group related medical conditions in pharmacovigilance case reports. Previous experience with ICD-10 on a semantic distance tool highlighted a bottleneck related to manual description of formal definitions in large terminologies. We propose a method based on acquisition of formal definitions by knowledge extraction from UMLS and morphosemantic analysis. These formal definitions are expressed with SNOMED International terms. We provide formal definitions for 758 WHO-ART terms: 321 terms defined from UMLS, 320 terms defined using morphosemantic analysis and 117 terms defined after expert evaluation. Computation of semantic distance (e.g. k-nearest neighbours) was implemented in J2EE terminology services. Similar WHO-ART terms defined by automated knowledge acquisition and ICD terms defined manually show similar behaviour in the semantic distance tool. Our knowledge acquisition method can help us to generate new formal definitions of medical terms for our semantic distance terminology services.

This paper concerns lung tissue classification using asymmetric-margin support vector machine (ASVM) to handle the imbalance of the positive and negative classes in a one-against-all multiclass classification problem. The hyperparameters of the algorithm are obtained using an optimization of the upper bound of the leave-one-out error of the ASVM. The ASVM is applied on the dataset with its original distribution and oversampled so that the ratio of the examples is equal to the prevalence of patients having the tissue in the database. The two versions of the ASVM models were compared with a model build with a conventional SVM. The ASVM improved the results obtained with a conventional SVM. The incorporation of prior knowledge concerning the prevalence of the patients improved the results obtained with ASVM.

We compare five common classifier families in their ability to categorize six lung tissue patterns in high-resolution computed tomography (HRCT) images of patients affected with interstitial lung diseases (ILD) but also normal tissue. The evaluated classifiers are Naive Bayes, k-Nearest Neighbor (k-NN), J48 decision trees, Multi-Layer Perceptron (MLP) and Support Vector Machines (SVM). The dataset used contains 843 regions of interest (ROI) of healthy and five pathologic lung tissue patterns identified by two radiologists at the University Hospitals of Geneva. Correlation of the feature space composed of 39 texture attributes is studied. A grid search for optimal parameters is carried out for each classifier family. Two complementary metrics are used to characterize the performances of classification. Those are based on McNemar's statistical tests and global accuracy. SVM reached best values for each metric and allowed a mean correct prediction rate of 87.9% with high class-specific precision on testing sets of 423 ROIs.

The vast availability of medical patient data in digital format creates the opportunity to use these data in medical informatics research projects. The objective is to improve future care by providing the medical staff with methods for automated data processing, including textual and visual information analysis and retrieval from medical databases. Many medical institutions do not possess a specific research computing infrastructure or the required budget for such an infrastructure to enable processing of these large amounts of data. Still, many institutions have many desktop PCs that could serve for biomedical research during the time they are little used without the need for expensive investments. The KnowARC project aims at building a middleware for such a simple-to-install Grid network. This article reviews requirements for computing Grids in large hospital environments. We use the computing infrastructure in the University Hospitals of Geneva as an example, and then present the solutions that the European Union-funded KnowARC project plans to undertake to solve the current problems. Methods currently employed in common Grid middleware distributions are also reviewed and assessed in relation to the goals of KnowARC. The computing infrastructure at the University Hospitals of Geneva is described as well as the needs and requirements for computing and storage services within this domain. A list of requirements for a Grid middleware to employ in such a challenging environment is developed. Finally, the proposed solutions and ideas of the KnowARC project are described in detail to present the project to a larger community. First proof of concept implementations and test results are described to illustrate how Grid networks are expected to become an important supplier of computational resources, which are required in several domains in biomedical research. A continuous process will be necessary to feed in the requirements of the biomedical domain to developers of Grid middleware to make the outcome meet the specific needs of the biomedical community.

In this paper, we compare five common classifier families in their ability to categorize six lung tissue patterns in high-resolution computed tomography (HRCT) images of patients affected with interstitial lung diseases (ILD) and with healthy tissue. The evaluated classifiers are naive Bayes, k-nearest neighbor, J48 decision trees, multilayer perceptron, and support vector machines (SVM). The dataset used contains 843 regions of interest (ROI) of healthy and five pathologic lung tissue patterns identified by two radiologists at the University Hospitals of Geneva. Correlation of the feature space composed of 39 texture attributes is studied. A grid search for optimal parameters is carried out for each classifier family. Two complementary metrics are used to characterize the performances of classification. These are based on McNemar’s statistical tests and global accuracy. SVM reached best values for each metric and allowed a mean correct prediction rate of 88.3% with high class-specific precision on testing sets of 423 ROIs.