Catálogo de publicaciones - libros

This paper deals with the interpretation of biomedical multivariate time series for extracting typical scenarios. This task is known to be difficult, due to the temporal nature of the data at hand, and to the context-sensitive aspect of data interpretation, which hamper the formulation of knowledge about the kind of patterns to detect and their interrelations. A new way to tackle this problem is proposed, based on a collaborative approach between a human and a machine by means of specific annotations. Two grounding principles, namely autonomy and knowledge discovery, support the co-construction of successive abstraction levels for data interpretation. A multi-agent system is proposed to implement effectively these two principles. Respiratory time series data (Flow, Paw) have been explored with our system for patient/ventilator asynchronies characterization studies.

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 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.

Decision-making is a crucial aspect of emergency response during mass casualty incidents (MCIs). MCIs require rapid decisions to be taken by geographically-dispersed teams in an environment characterized by insufficient information, ineffective collaboration and inadequate resources. Despite the increasing adoption of decision support systems in healthcare, there is limited evidence of their value in large-scale disasters. We conducted focus groups with emergency medical services and emergency department personnel who revealed that one of the main challenges in emergency response during MCIs is information management. Therefore, to alleviate the issues arising from ineffective information management, we propose R-CAST-MED, an intelligent agent architecture built on Recognition-Primed Decision-making (RPD) and Shared Mental Models (SMMs). A simulation of R-CAST-MED showed that this tool enabled efficient information management by identifying relevant information, inferring missing information and sharing information with other agents, which led to effective collaboration and coordination of tasks across teams.

This paper presents the cognitive model of gastroesophageal reflux disease (GERD) developed for the Maryland Virtual Patient simulation and mentoring environment. GERD represents a class of diseases that have a large number of clinical manifestations. Our model at once manages that complexity while offering robust automatic function in response to open-ended user actions. This ontologically grounded model is largely based on script-oriented representations of causal chains reflecting the actual physiological processes in virtual patients. A detailed description of the GERD model is presented along with a high-level description of the environment for which it was developed.

The paper presents a distributed approach for the interpretation of epileptic signals based on a dynamical vectorial analysis method. The approach associates signal processing methods into a situated, reactive, cooperative and decentralized implementation. The objective is to identify and locate the various interictal and ictal epileptiform events (pathological and/or normal) contained in intracerebral EEG signals (one hundred recording channels in general) recorded in patients suffering from partial temporal lobe epilepsy. This approach associates some signal processing methods (spectral analysis, causality measurements, detection, classification) in a multi-agent system.

One of the hardest tasks in any healthcare application is the management of knowledge. Organisational information as well as medical concepts should be represented in an appropriate way in order to improve interoperability among existing systems, to allow the implementation of knowledge-based intelligent systems, or to provide high level support to healthcare professionals. This paper proposes the inclusion of an especially designed ontology into an agent-based medical platform called . The ontology has been constructed as an external resource, allowing agents to coordinate complex activities defined in any clinical guideline.

Many machines used in the modern hospital settings offer real time physiological monitoring. Haemodialysis machines combine a therapeutic treatment system integrated with sophisticated monitoring equipment. A large array of parameters can be collected including cardiovascular measures such as heart rate and blood pressure together with treatment related data including relative blood volume, ultrafiltration rate and small molecule clearance. A small subset of this information is used by clinicians to monitor treatment and plan therapeutic strategies but it is not usually analysed in any detail. The focus of this paper is the analysis of data collected over a number of treatment sessions with a view to predicting patient physiological behaviour whilst on dialysis and correlating this with clinical characteristics of individual patients.

One of the commonest complications experienced by patients on dialysis is symptomatic hypotension. We have taken real time treatment data and outline a program of work which attempts to predict when hypotension is likely to occur, and which patients might be particularly prone to haemodynamic instability. This initial study has investigated: the rate of change of blood pressure versus rate of change of heart rate, rate of fluid removal, and rate of uraemic toxin clearance. We have used a variety of machine learning techniques (including hierarchical clustering, and Bayesian Network analysis algorithms). We have been able to detect from this dataset, 3 distinct groups which appear to be clinically meaningful. Furthermore we have investigated whether it is possible to predict changes in blood pressure in terms of other parameters with some encouraging results that merit further study.

Nowadays, methods for discovering temporal knowledge try to extract more complete and representative patterns. The use of qualitative temporal constraints can be helpful in that aim, but its use should also involve methods for reasoning with them (instead of using them just as a high level representation) when a pattern consists of a constraint network instead of an isolated constraint.

In this paper, we put forward a method for mining temporal patterns that makes use of a formal model for representing and reasoning with qualitative temporal constraints. Three steps should be accomplished in the method: 1) the selection of a model that allows a trade off between efficiency and representation; 2) a preprocessing step for adapting the input to the model; 3) a data mining algorithm able to deal with the properties provided by the model for generating a representative output.

In order to implement this method we propose the use of the Fuzzy Temporal Constraint Network (FTCN) formalism and of a temporal abstraction method for preprocessing. Finally, the ideas of the classic methods for data mining inspire an algorithm that can generate FTCNs as output.

Along this paper, we focus our attention on the data mining algorithm.

The care for patients with chronic and progressive diseases often requires that reliable estimates of their remaining lifetime are made. The predominant method for obtaining such individual prognoses is to analyze historical data using Cox regression, and apply the resulting model to data from new patients. However, the black-box nature of the Cox regression model makes it unattractive for clinical practice. Instead most physicians prefer to relate a new patient to the histories of similar, individual patients that were treated before. This paper presents a prognostic inference method that combines the -nearest neighbor paradigm with Cox regression. It yields survival predictions for individual patients, based on small sets of similar patients from the past, and can be used to implement a prognostic case-retrieval system. To evaluate the method, it was applied to data from patients with idiopathic interstitial pneumonia, a progressive and lethal lung disease. Experiments pointed out that the method competes well with Cox regression. The best predictive performance was obtained with a neighborhood size of 20.

Disease processes in patients are temporal in nature and involve uncertainty. It is necessary to gain insight into these processes when aiming at improving the diagnosis, treatment and prognosis of disease in patients. One way to achieve these aims is by explicitly modelling disease processes; several researchers have advocated the use of dynamic Bayesian networks for this purpose because of the versatility and expressiveness of this time-oriented probabilistic formalism. In the research described in this paper, we investigate the role of context-specific independence information in modelling the evolution of disease. The hypothesis tested was that within similar populations of patients differences in the learnt structure of a dynamic Bayesian network may result, depending on whether or not patients have a particular disease. This is an example of temporal context-specific independence information. We have tested and confirmed this hypothesis using a constraint-based Bayesian network structure learning algorithm which supports incorporating background knowledge into the learning process. Clinical data of mechanically-ventilated ICU patients, some of whom developed ventilator-associated pneumonia, were used for that purpose.