Catálogo de publicaciones - libros

Integration of different information sources has been a problem that has been challenging (or perhaps better: plaguing) Computer Science throughout the decades. As soon as we had two computers, we wanted to exchange information between them, and as soon as we had two databases, we wanted to link them together.

Fortunately, Computer Science has made much progress on different levels:

interoperability between systems has been all but solved: with the advent of hardware standards such as Ethernet, and with protocols such as TCP/IP and HTTP, we can nowadays walk into somebody’s house or office, and successfully plug our computer into the network, giving instant world-wide physical connectivity.

Physical connectivity is not sufficient. We must also agree on the of the messages we will exchange. Again, much progress has been made in recent years, with open standards such HTML and XML.

Today access to computing power and communication has become nearly ubiquitous. Mobile computers and even mobile phones have computing power, storage and graphics capabilities comparable to PCs from a few years ago. With the advent of GPRS, UMTS, WLAN and other networking technologies high speed Internet access is possible nearly anywhere. At the same time an enormous amount of permanently updated information has become available online. In fact one can say that for nearly any situation there is guaranteed to be a piece of useful information somewhere on the network. This includes such trivial everyday things like restaurant menus and transportation delays but also information relevant to a variety of professional applications. The latter include building plans (relevant for rescue personnel), patient record (needed for example by emergency medics) and multimedia manuals (for maintenance and assembly work).

Data driven interpretation of multiple physiological measurements in the domain of intensive care is a key point to provide decision support. The abstraction method presented in this paper provides two levels of symbolic interpretation. The first, at mono parametric level, provides 4 classes (increasing, decreasing, constant and transient) by combination of trends computed at two characteristic spans. The second, at multi parametric level, gives an index of global behavior of the system, that is used to segment the observation. Each segment is therefore described as a sequence of words that combines the results of symbolization. Each step of the abstraction process leads to a visual representation that can be validated by the clinician. Construction of sequences do not need any prior introduction of medical knowledge. Sequences can be introduced in a machine learning process in order to extract temporal patterns related to specific clinical or technical events.

This paper presents a novel algorithm for extracting rules expressing complex patterns from temporal data. Typically, a temporal rule describes a temporal relationship between the antecedent and the consequent, which are often time-stamped events. In this paper we introduce a new method to learn rules with complex temporal patterns in both the antecedent and the consequent, which can be applied in a variety of biomedical domains. Within the proposed approach, the user defines a set of complex interesting patterns that will constitute the basis for the construction of the temporal rules. Such complex patterns are represented with a Temporal Abstraction formalism. An APRIORI-like algorithm then extracts precedence temporal relationships between the complex patterns. The paper presents the results obtained by the rule extraction algorithm in two different biomedical applications. The first domain is the analysis of time series coming from the monitoring of hemodialysis sessions, while the other deals with the biological problem of inferring regulatory networks from gene expression data.

The temporal dimension is a characterizing factor of many diseases, in particular, of the exanthematic diseases. Therefore, the diagnosis of this kind of diseases can be based on the recognition of the typical temporal progression and duration of different symptoms. To this aim, we propose to apply a temporal reasoning system we have developed. The system is able to handle both qualitative and metric temporal knowledge affected by vagueness and uncertainty. In this preliminary work, we show how the fuzzy temporal framework allows us to represent typical temporal structures of different exanthematic diseases (e.g. Scarlet Fever, Measles, Rubella et c.) thus making possible to find matches with data coming from the patient disease.

Several systems have been designed to reason about longitudinal patient data in terms of abstract, clinically meaningful concepts derived from raw time-stamped clinical data. However, current approaches are limited by their treatment of missing data and of the inherent uncertainty that typically underlie clinical raw data. Furthermore, most approaches have generally focused on a single patient. We have designed a new probability-oriented methodology to overcome these conceptual and computational limitations. The new method includes also a practical parallel computational model that is geared specifically for implementing our probabilistic approach in the case of abstraction of a large number of electronic medical records.

Time is an essential element in the clinical management of patients as disease processes develop in time. A typical example of a disease process where time is considered important is the development of ventilator-associated pneumonia (VAP). A Bayesian network was developed previously to support clinicians in the diagnosis and treatment of VAP. In the research described in this paper, we have investigated whether this Bayesian network can also be used to analyse the temporal data collected in the ICU for patterns indicating development of VAP. In addition, it was studied whether the Bayesian network was able to suggest appropriate antimicrobial treatment. A temporal database with over 17700 patient days was used for this purpose.

The blood-glucose-level (BGL) of Intensive Care (IC) patients requires close monitoring and control. In this paper we describe a general data-driven analytical method for studying the effectiveness of BGL management. The method is based on developing and studying a clinical outcome reflecting the effectiveness of treatment in time. Decision trees are induced in order to discover relevant patient and other characteristics for influencing this outcome. By systematically varying the start and duration of time intervals in which the outcome behavior is studied, our approach distinguishes between time-related (e.g. the BGL at admission time) and intrinsic-related characteristics (e.g. the patient being diabetic).

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.

This paper presents a method to choose the threshold for dichotomization of survival outcomes in a structured fashion based on data analysis. The method is illustrated with an application to the prediction problem of the outcome (ICU LOS). Threshold selection is based on comparing the calibration of predictive models for dichotomized outcomes with increasing threshold values. To quantify model calibration a measure insensitive to class unbalance is used. The threshold value for which the associated predictive model has superior calibration is selected, and the corresponding model is used in practice. Using this method to select the threshold for ICU LOS, the best model calibration is found at a threshold of five days.