Aime

The Markov Random Field (MRF) probabilistic framework is classically introduced for a robust segmentation of Magnetic Resonance Imaging (MRI) brain scans. Most MRF approaches handle tissues segmentation via global model estimation. Structure segmentation is then carried out as a separate task. We propose in this paper to consider MRF segmentation of tissues and structures as two local and cooperative procedures immersed in a multiagent framework. Tissue segmentation is performed by partitionning the volume in subvolumes where agents estimate local MRF models in cooperation with their neighbours to ensure consistency of local models. These models better reflect local intensity distributions. Structure segmentation is performed via dynamically localized agents that integrate anatomical spatial constraints provided by an a priori fuzzy description of brain anatomy. Structure segmentation is not reduced to a postprocessing step: rather, structure agents cooperate with tissue agents to render models gradually more accurate. We report several experiments that illustrate the working of our multiagent framework. The evaluation was performed using both phantoms and real 3T brain scans and showed a robustness to nonuniformity and noise together with a low computational time. This MRF agent based approach appears as a very promising new tool for complex image segmentation.

There is much empirical evidence about the success of naive Bayesian classification (NBC) in medical applications of attribute-based machine learning. NBC assumes conditional independence between attributes. In classification, such classifiers sum up the pieces of classrelated evidence from individual attributes, independently of other attributes. The performance, however, deteriorates significantly when the "interactions" between attributes become critical. We propose an approach to handling attribute interactions within the framework of "voting" classifiers, such as NBC. We propose an operational test for detecting interactions in learning data and a procedure that takes the detected interactions into account while learning. This approach induces a structuring of the domain of attributes, it may lead to improved classifier's performance and may provide useful novel information for the domain expert when interpreting the results of learning. We report on its application in data analysis and model construction for the prediction of clinical outcome in hip arthroplasty.

The paper proposes a methodology based on a neural network to process the signals related to the hands movements in response to the Transcranial Magnetic Stimulation (TMS) in order to diagnose the pathology and evaluate the treatment of the patients affected by demency diseases. First a time-frequency analysis of such signals is carried out to identify the main signal variables that characterize the demency diseases. Then these variables are processed by a neural network in order to classify the responses into four classes: healthy subjects, people affected by Subcortical Ischemic Vascular Dementia (SIVD) and/or Alzheimer. A comparison between the proposed method and a fuzzy approach previously developed by the authors is presented.

The National Cancer Institute's Thesaurus (NCIT) has been created with the goal of providing a controlled vocabulary which can be used by specialists in the various sub-domains of oncology. It is intended to be used for purposes of annotation in ways designed to ensure the integration of data and information deriving from these various sub-domains, and thus to support more powerful cross-domain inferences. In order to evaluate its suitability for this purpose, we examined the NCIT's treatment of the kinds of entities which ...

The Internet has revolutionized the way knowledge can be accessed and presented. However, the explosion of web content that has followed is now producing major difficulties for effective selection and retrieval of information that is relevant for the task in hand. In disseminating clinical guidelines and other knowledge sources in healthcare, for example, it may be desirable to provide a presentation of current knowledge about best practice which is limited to material that is appropriate for the current patient context.. A promising solution to this problem is to augment conventional guideline documents with decision-making and other "intelligent" services tailored to specific needs at the point of care. In this paper we describe how BMJ's Clinical Evidence, a well-known medical reference on the web, was enhanced with patient data acquisition and decision support services implemented in PROforma.

The aim of this study is to develop the Bayesian Least Squares Support Vector Machine (LS-SVM) classifiers, for preoperatively predicting the malignancy of ovarian tumors. We describe how to perform parameter estimation, input variable selection for LS-SVM within the evidence framework. The issue of computing the posterior class probability for risk minimization decision making is addressed. The relation between the LS-SVM model and kernel principal component analysis is also indicated and used for interpretation of the LS-SVM classifiers.

We have developed a prototype system to support decision making in Breast Cancer, wherein the varied nature of expertise is modelled by multiple ontologies that provide domain-specific grounding to concepts and relationships used. While the different medical experts need to be co-present at a meeting, our system employs a distributed architecture for handling data and invoking services appropriate for the requirements of this decision-making process. This distributed system is built upon Semantic Web technology, which ...

Ambulatory care supporting long-term treatment of type I diabetes mellitus (DM) is based on the analysis of daily notes of serum glucose measurements, carbohydrate intake, and insulin dosage. In order to improve glycaemic control, telemedicine support aims at improving communication between patients and diabetologists. Patient data are collected using mobile phone services. Weekly responses from the diabetes care centre should help the patient to optimize glycaemic control. The telemedical support system VIE-DIAB integrates data collection, visualization, and recommendation handling by using mobile phone and internet services. Its core is a module which visualizes a summary of the patient's diary data of the last 4 weeks. Traditional methods for displaying these data mainly use (coloured) scatter plots, line graphs or summary histograms to give a graphical overview of the data. VIE-DIAB supports a more intuitive view by presenting 4x7 multiples. Each multiple represents the serum glucose values of one day. First results indicate that this form of data presentation is very useful to physicians.

The goal of evidence-based medicine is to uniformly apply evidence gained from scientific research to aspects of clinical practice. In order to achieve this goal, new applications that integrate increasingly disparate health care information resources are required. Access to and provision of evidence must be seamlessly integrated with existing clinical workflow and evidence should be made available where it is most often required-at the point of care. In this paper we address these requirements and outline a concept-based framework that captures the context of a current patient-physician encounter by combining disease and patientspecific information into a logical query mechanism for retrieving relevant evidence from the Cochrane Library. Returned documents are organized by automatically extracting concepts from the evidence-based query to create meaningful clusters of documents which are presented in a manner appropriate for point of care support. The framework is currently being implemented as a prototype software agent that operates within the larger context of a multi-agent application for supporting workflow management of emergency pediatric asthma exacerbations.

In the context of high dependent environments such as intensive care or operating room we would like to predict, as soon as possible, from past and present states, the evolution of the patient's condition. Based on the set of physiological dat, clinicians have difficulties to formalize typical high level scenes that are representative of specific patient's state sequences to recognize. On the other hand, signal processing algorithms are limited to low level pattern detection. We propose an interactive environment for an in-depth exploration by the clinician of data. We hypothesize that such human-computer collaboration could help with the definition of signatures representative of specific situations. To test our hypothesis, we have defined a multi-agent system (MAS ) with the capacities of 1) segmenting, 2) classifying and 3) learning. These three tasks are continuously adapted based on interactions with the clinician. Preliminary results are presented to support our assumption.

Translating clinical guidelines into formal models is beneficial in many ways, but expensive. The progress in medical knowledge requires clinical guidelines to be updated at relatively short intervals, leading to the term living guideline. This causes potentially expensive, frequent updates of the corresponding formal models. When performing these updates, there are two goals: The modelling effort must be minimised and the links between the original document and the formal model must be maintained. In this paper, we describe our solution, using tools and techniques developed during the Protocure II project.

Evidence-based clinical guidelines require frequent updates due to research and technology advances. The quality of guideline updates can be improved if the knowledge underlying the guideline text is explicitly modelled using the so-called guideline patterns (GPs), mappings between a text fragment and a formal representation of its corresponding medical knowledge. Ontology-driven extraction of linguistic patterns is a method to automatically reconstruct the control knowledge captured in guidelines, which facilitates a more effective modelling and authoring of clinical guidelines. We illustrate by examples the use of a method for generating and searching for linguistic guideline patterns in the text of a guideline for treatment of breast cancer, and provide a general evaluation of usefulness of these patterns in the modelling of the guideline analyzed.

Currently over 95% of children who are diagnosed with Acute Lymphoblastic Leukaemia in the UK are enrolled into Medical Research Council trials. The trial protocol specifies that following initial treatment there is a 2-3 year maintenance period during which drug dosage decisions are made weekly according to a set of pre-defined rules. These rules are complex, and there is a significant frequency of error in clinical practice, which can lead to patient harm. We have built a web-based decision support system (called LISA) to address this problem. The dose alteration rules from the MRC protocol were formalised in the PROforma guideline modeling language as a state transition problem, and dose adjustment recommendations are provided into the clinical setting by a PROforma enactment engine. The design and implementation of the decision support module, the safety issues raised and the strategy adopted for resolving them are discussed. System safety is very likely to become a major professional challenge for the medical AI community and it can be addressed, in this case, with relatively straightforward techniques.

The aim of this study is to develop the Bayesian Least Squares Support Vector Machine (LS-SVM) classifiers, for preoperatively predicting the malignancy of ovarian tumors. We describe how to perform parameter estimation, input variable selection for LS-SVM within the evidence framework. The issue of computing the posterior class probability for risk minimization decision making is addressed. The relation between the LS-SVM model and kernel principal component analysis is also indicated and used for interpretation of the LS-SVM classifiers.

Modern medical vocabularies can contain up to hundreds of thousands of concepts. In any particular use-case only a small fraction of these will be needed. In this paper we first define two notions of a diseasecentric subdomain of a large ontology. We then explore two methods for identifying disease-centric subdomains of such large medical vocabularies. The first method is based on lexically querying the ontology with an iteratively extended set of seed queries. The second method is based on manual mapping between concepts from a medical guideline document and ontology concepts. Both methods include concept-expansion over subsumption and equality relations. We use both methods to determine a breast-cancer-centric subdomain of the SNOMED CT ontology. Our experiments show that the two methods produce a considerable overlap, but they also yield a large degree of complementarity, with interesting differences between the sets of concepts that they return. Analysis of the results reveals strengths and weaknesses of the different methods.

The degree of fulfillment of clinical guidelines is considered a key factor when evaluating the quality of a clinical service. Guidelines can be seen as processes describing the sequence of activities to be done. Consequently, workflow formalisms seem to be a valid approach to model the flow of actions in the guideline and their temporal aspects. The application of a guideline to a specific patient (guideline instance) can be modeled by means of a workflow case. The best (worst) application of a guideline, represented as a reference workflow case, can be used to evaluate the quality of the service, by comparing the optimal case with specific patient instances. On the other hand, the correct application of a guideline to a patient involves the fulfillment of the guideline temporal constraints. Thus, the evaluation of the temporal similarity degree between different workflow cases is a key aspect in evaluating health care quality. In this work, we represent a portion of the stroke guideline using a temporal workflow schema and we propose a method to evaluate the temporal similarity between workflow cases. Our proposal, based on temporal constraint networks, consists of a linear combination of functions to differentiate intra-task and inter-task temporal distances.

Clinical guidelines (GLs) play an important role to standardize and organize clinical processes according to evidence-based medicine. Several computer-based GL representation languages have been defined, usually focusing on expressiveness and/or on user-friendliness. In many cases, the interpretation of some constructs in such languages is quite unclear. Only recently researchers have started to provide a formal semantics for some of such languages, thus providing an unambiguous specification for implementers, and a formal ground in which different approaches can be compared, and verification techniques can be applied. Petri Nets are a natural candidate formalism to cope with GL semantics, since they are explicitly geared towards the representation of processes, and are paired with powerful verification mechanisms. We show how Petri Nets can cope with the semantics of GLs in a clear way, taking the system GLARE formalism as a case study.

Image representation is an important issue for medical image analysis, classification and retrieval. Recently, the bag of features approach has been proposed to classify natural scenes, using an analogy in which visual features are to images as words are to text documents. This process involves feature detection and description, construction of a visual vocabulary and image representation building through visual-word occurrence analysis. This paper presents an evaluation of different representations obtained from the bag of features approach to classify histopathology images. The obtained image descriptors are processed using appropriate kernel functions for Support Vector Machines classifiers. This evaluation includes extensive experimentation of different strategies, and analyses the impact of each configuration in the classification result.

Several computer-based approaches to Clinical Guidelines have been developed in the last two decades. However, only recently the community has started to cope with the fact that Clinical Guidelines are just a part of the medical knowledge that physicians have to take into account when treating patients. The procedural knowledge in the guidelines have to be complemented by additional declarative medical knowledge. In this paper, we analyse such an interaction, by studying the conformance problem, defined as evaluating the adherence of a set of performed clinical actions w.r.t. the behaviour recommended by the guideline and by the medical knowledge.

In the area of Medical Informatics, there is an increasing realization that temporal information plays a crucial role, so that suitable database models and query languages are needed to store and support it. In this paper we show that current approaches developed within the database field have some limitations even from the point of view of the data model, so that an important class of temporal medical data cannot be properly represented. We propose a new three-sorted model and a query language that overcome such limitations.

Clinical guidelines (GLs) play an important role to standardize and organize clinical processes according to evidence-based medicine. Several computer-based GL representation languages have been defined, usually focusing on expressiveness and/or on user-friendliness. In many cases, the interpretation of some constructs in such languages is quite unclear. Only recently researchers have started to provide a formal semantics for some of such languages, thus providing an unambiguous specification for implementers, and a formal ground in which different approaches can be compared, and verification techniques can be applied. Petri Nets are a natural candidate formalism to cope with GL semantics, since they are explicitly geared towards the representation of processes, and are paired with powerful verification mechanisms. We show how Petri Nets can cope with the semantics of GLs in a clear way, taking the system GLARE formalism as a case study.

In this paper, we report on the fusion of simple retrieval strategies with thesaural resources in order to perform large-scale text categorization tasks. Unlike most related systems, which rely on training data in order to infer text-to-concept relationships, our approach can be applied with any controlled vocabulary and does not use any training data. The first classification module uses a traditional vector-space retrieval engine, which has been fine-tuned for the task, while the second classifier is based on regular variations of the concept list. For evaluation purposes, the system uses a sample of MedLine and the Medical Subject Headings (MeSH) terminology as collection of concepts. Preliminary results show that performances of the hybrid system are significantly improved as compared to each single system. For top returned concepts, the system reaches performances comparable to machine learning systems, while genericity and scalability issues are clearly in favor of the learningfree approach. We draw conclusion on the importance of hybrids strategies combining data-poor classifiers and knowledge-based terminological resources for general text mapping tasks.

Mammographic analysis is a difficult task due to the complexity of image interpretation. This results in diagnostic uncertainty, thus provoking the need for assistance by computer decision-making tools. Probabilistic modelling based on Bayesian networks is among the suitable tools, as it allows for the formalization of the uncertainty about parameters, models, and predictions in a statistical manner, yet such that available background knowledge about characteristics of the domain can be taken into account. In this paper, we investigate a specific class of Bayesian networks—causal independence models—for exploring the dependencies between two breast image views. The proposed method is based on a multi-stage scheme incorporating domain knowledge and information obtained from two computer-aided detection systems. The experiments with actual mammographic data demonstrate the potential of the proposed two-view probabilistic system for supporting radiologists in detecting breast cancer, both at a location and a patient level.

Case-based reasoning systems applied to safety-critical environments justify specific measures to ensure that the assistance provided is not dangerous to human life. This article presents a case-based reasoning system developed for medical decision-support in a safety-critical environment, the CARE- PARTNER system. Based on the evaluation of the reliability of the system, it stresses the importance to differentiate between reliability and safety, and how case-based reasoning can ensure safety. A knowledge-based approach has enabled to model what types of information are safety-critical, and to adapt the decision-support system results, by introspective reasoning, based on the knowledge type, the user, and the task.

We describe an approach for developing knowledge-based medical decision support systems based on the rather new technology of case-based reasoning. This work is based on the results of the Inreca European project and preliminary results from the Inreca+ project which particularly deals with medical applications. One goal was to start from case-based reasoning technology for technical diagnosis, as it was available among the partners, and 'scale-up' to more general non-technical decision support tasks as typically given in medical domains. Inreca technology is used to build an initial decision support system at the Russian Toxicology Information and Advisory Center in Moscow for diagnosing poison cases that are caused by psychotropes.

Reification of parthood relations according to the SEP-triplet encoding pattern has been employed in the clinical terminology SNOMED CT to simulate transitivity of the part-of relation via transitivity of the is-a relation and to inherit properties along part-of links. In this paper we argue that using a more expressive representation language, which allows for a direct representation of the relevant properties of the part-of relation, makes modelling less error prone while having no adverse effect on the efficiency of reasoning.

Modern medical vocabularies can contain up to hundreds of thousands of concepts. In any particular use-case only a small fraction of these will be needed. In this paper we first define two notions of a diseasecentric subdomain of a large ontology. We then explore two methods for identifying disease-centric subdomains of such large medical vocabularies. The first method is based on lexically querying the ontology with an iteratively extended set of seed queries. The second method is based on manual mapping between concepts from a medical guideline document and ontology concepts. Both methods include concept-expansion over subsumption and equality relations. We use both methods to determine a breast-cancer-centric subdomain of the SNOMED CT ontology. Our experiments show that the two methods produce a considerable overlap, but they also yield a large degree of complementarity, with interesting differences between the sets of concepts that they return. Analysis of the results reveals strengths and weaknesses of the different methods.