Federating distributed and heterogeneous information sources in neuroimaging: the NeuroBase Project.

The NeuroBase project aims at studying the requirements for federating, through the Internet, information sources in neuroimaging. These sources are distributed in different experimental sites, hospitals or research centers in cognitive neurosciences, and contain heterogeneous data and image processing programs. More precisely, this project consists in creating of a shared ontology, suitable for supporting various neuroimaging applications, and a computer architecture for accessing and sharing relevant distributed information. We briefly describe the semantic model and report in more details the architecture we chose, based on a media-tor/wrapper approach. To give a flavor of the future deployment of our architecture, we de-scribe a demonstrator that implements the comparison of distributed image processing tools applied to distributed neuroimaging data.

A UMLS-based knowledge acquisition tool for rule-based clinical decision support system development.

Decision support systems in the medical field have to be easily modified by medical experts themselves. The authors have designed a knowledge acquisition tool to facilitate the creation and maintenance of a knowledge base by the domain expert and its sharing and reuse by other institutions. The Unified Medical Language System (UMLS) contains the domain entities and constitutes the relations repository from which the expert builds, through a specific browser, the explicit domain ontology. The expert is then guided in creating the knowledge base according to the pre-established domain ontology and condition-action rule templates that are well adapted to several clinical decision-making processes. Corresponding medical logic modules are eventually generated. The application of this knowledge acquisition tool to the construction of a decision support system in blood transfusion demonstrates the value of such a pragmatic methodology for the design of rule-based clinical systems that rely on the highly progressive knowledge embedded in hospital information systems.

Scenario recognition for temporal reasoning in medical domains.

The recognition of high level clinical scenes is fundamental in patient monitoring. In this paper, we propose a technique for recognizing a session, i.e. the clinical process evolution, by comparison against a predetermined set of scenarios, i.e. the possible behaviors for this process. We use temporal constraint networks to represent both scenario and session. Specific operations on networks are then applied to perform the recognition task. An index of temporal proximity is introduced to quantify the degree of matching between two temporal networks in order to select the best scenario fitting a session. We explore the application of our technique, implemented in the Déjà Vu system, to the recognition of typical medical scenarios with both precise and imprecise temporal information.

A cooperative framework for segmentation of MRI brain scans.

Automatic segmentation of MRI brain scans is a complex task for two main reasons: the large variability of the human brain anatomy, which limits the use of general knowledge and, inherent to MRI acquisition, the artifacts present in the images that are difficult to process. To tackle these difficulties, we propose to mix, in a cooperative framework, several types of information and knowledge provided and used by complementary individual systems: presently, a multi-agent system, a deformable model and an edge detector. The outcome is a cooperative segmentation performed by a set of region and edge agents constrained automatically and dynamically by both, the specific gray levels in the considered image, statistical models of the brain structures and general knowledge about MRI brain scans. Interactions between the individual systems follow three modes of cooperation: integrative, augmentative and confrontational cooperation, combined during the three steps of the segmentation process namely, the specialization of the seeded-region-growing agents, the fusion of heterogeneous information and the retroaction over slices. The described cooperative framework allows the dynamic adaptation of the segmentation process to the own characteristics of each MRI brain scan. Its evaluation using realistic brain phantoms is reported.

NéoGanesh: a working system for the automated control of assisted ventilation in ICUs.

Automating the control of therapy administered to a patient requires systems which integrate the knowledge of experienced physicians. This paper describes NéoGanesh, a knowledge-based system which controls, in closed-loop, the mechanical assistance provided to patients hospitalized in intensive care units. We report on how new advances in knowledge representation techniques have been used to model medical expertise. The clinical evaluation shows that such a system relieves the medical staff of routine tasks, improves patient care, and efficiently supports medical decisions regarding weaning. To be able to work in closed-loop and to be tested in real medical situations, NéoGanesh deals with a voluntarily limited problem. However, embedded in a powerful distributed environment, it is intended to support future extensions and refinements and to support reuse of knowledge bases.

Using a general theory of time and change in patient monitoring: experiment and evaluation.

In this paper, we propose to use one of the well-known general theories of time and change, namely the Event Calculus (Kowalski and Sergot, New Generation Computing 4, 67-95, 1986), to represent temporal aspects in intelligent medical monitoring systems. In particular, we explore the application of CEC (Chittaro and Montanari, Computational Intelligence 12, 359-382, 1996) (an efficient implementation of the Event Calculus) to the management of mechanical ventilation. First, we present the prototype we have built, which has been extensively tested on patient's data from real clinical cases. Then, we provide a thorough evaluation of the obtained results, pointing out both strengths and weaknesses of the approach, and identifying a number of extensions which can be extremely useful to scale up the medical application of the approach.

Knowledge-based systems for automatic ventilatory management.

In intensive care and anesthesia, the demand for computerized medical assistants is potentially considerable, to filter and synthesize the growing mass of clinical parameters and information available. The authors detail how knowledge-based computerized assistants can be constructed for automatic ventilation management and report clinical results obtained with such a system for closed-loop control of pressure support ventilation and decision for extubation.

Local landmark alignment for high-resolution fMRI group studies: toward a fine cortical investigation of hand movements in human.

Converging evidence conclusively demonstrates the robust relationship between anatomical landmarks and underlying functional organization in primary cortical regions. In consequence, a precise alignment across subjects of such specific individual landmarks should improve the overlap of the corresponding functional areas and thus the detection of active clusters at the group level. In an effort to define a dedicated processing pipeline for a fine non-invasive exploration of the motor cortex in human, we evaluated four recent non-linear registration methods based on anatomical and functional indexes. We used high-resolution functional MRI data to finely reveal the impact of the registration on the cortical assignment of the detected clusters. Our results first demonstrate that the quality of registration strongly affects the statistical significance and the assignment of activated clusters to specific anatomical regions, here in the primary motor area. Our results also illustrate the bias induced by the chosen reference template on the detected clusters. The analysis of the Jacobian of the deformation field informs us about how each method deforms the anatomical structures and functional maps. The methodology we propose, combining high resolution fMRI and non-linear registration method, allows a robust non-invasive exploration of the motor cortex.

Fully Bayesian joint model for MR brain scan tissue and structure segmentation.

In most approaches, tissue and subcortical structure segmentations of MR brain scans are handled globally over the entire brain volume through two relatively independent sequential steps. We propose a fully Bayesian joint model that integrates local tissue and structure segmentations and local intensity distributions. It is based on the specification of three conditional Markov Random Field (MRF) models. The first two encode cooperations between tissue and structure segmentations and integrate a priori anatomical knowledge. The third model specifies a Markovian spatial prior over the model parameters that enables local estimations while ensuring their consistency, handling this way nonuniformity of intensity without any bias field modelization. The complete joint model provides a sound theoretical framework for carrying out tissue and structure segmentation by distributing a set of local and cooperative MRF models. The evaluation, using a previously affine-registred atlas of 17 structures and performed on both phantoms and real 3T brain scans, shows good results.

LOCUS: local cooperative unified segmentation of MRI brain scans.

We propose to carry out cooperatively both tissue and structure segmentations by distributing a set of local and cooperative models in a unified MRF framework. Tissue segmentation is performed by partitionning the volume into subvolumes where local MRFs are estimated in cooperation with their neighbors to ensure consistency. Local estimation fits precisely to the local intensity distribution and thus handles nonuniformity of intensity without any bias field modelization. Structure segmentation is performed via local MRFs that integrate localization constraints provided by a priori general fuzzy description of brain anatomy. Structure segmentation is not reduced to a postprocessing step but cooperates with tissue segmentation to gradually and conjointly improve models accuracy. The evaluation was performed using phantoms and real 3T brain scans. It shows good results and in particular robustness to nonuniformity and noise with a low computational cost.

Multimodal MRI segmentation of ischemic stroke lesions.

The problem addressed in this paper is the automatic segmentation of stroke lesions on MR multi-sequences. Lesions enhance differently depending on the MR modality and there is an obvious gain in trying to account for various sources of information in a single procedure. To this aim, we propose a multimodal Markov random field model which includes all MR modalities simultaneously. The results of the multimodal method proposed are compared with those obtained with a mono-dimensional segmentation applied on each MRI sequence separately. We constructed an Atlas of blood supply territories to help clinicians in the determination of stroke subtypes and potential functional deficit.

Effective domain-dependent reuse in medical knowledge bases.

Knowledge reuse is now a critical issue for most developers of medical knowledge-based systems. As a rule, reuse is addressed from an ambitious, knowledge-engineering perspective that focuses on reusable general purpose knowledge modules, concepts, and methods. However, such a general goal fails to take into account the specific aspects of medical practice. From the point of view of the knowledge engineer, whose goal is to capture the specific features and intricacies of a given domain, this approach addresses the wrong level of generality. In this paper, we adopt a more pragmatic viewpoint, introducing the less ambitious goal of "domain-dependent limited reuse" and suggesting effective means of achieving it in practice. In a knowledge representation framework combining objects and production rules, we propose three mechanisms emerging from the combination of object-oriented programming and rule-based programming. We show these mechanisms contribute to achieve limited reuse and to introduce useful limited variations in medical expertise.

A knowledge-based system for assisted ventilation of patients in intensive care units.

The procedure for weaning a patient with respiratory insufficiency from mechanical ventilation may be complex and requires expertise obtained by long clinical practice. We designed a knowledge-based system for the management of patients receiving respiratory support and implemented a weaning procedure. The system is intended for patients whose spontaneous respiratory activity is assisted by a Hamilton Veolar ventilator delivering a positive pressure plateau during inspiration (Pressure Support Ventilation mode). Our closed-loop real-time system running on a Personal Computer continuously adapts the assistance provided by the ventilator to the patient's evolution, and indicates when the patient can be withdrawn from the ventilator. Three parameters are used to appreciate the 'respiratory comfort' of the patient: breathing frequency, which we consider the most informative index, tidal volume and end-tidal CO2 pressure. A preliminary study of 19 patients was performed to evaluate the ability of our system to adapt the assistance to the patient's needs, with the main objective of facilitating weaning by gradually lowering the level of assistance. In 10 of these patients, considered as good candidates for weaning on the strength of objective criteria, the system maintained the breathing pattern in a zone of comfort for 95% of the period of assisted ventilation and stated that they were 'weanable'. This was consistent with the clinical evolution of all 10 patients. These results show that such a system can provide effective management for mechanically ventilated patients.

High impedance mechanical ventilator for small animals: use of programmable controller.

We built a simple high impedance ventilator, which generates a pattern of flow largely independent of respiratory mechanics, to mechanically ventilate anaesthetized small animals. The system includes a source of compressed gas with an electronic valve and a flow controller on the inspiratory side and a second valve on the expiratory side. The two valves are driven by a programmable controller. To assess the performance of this ventilator we measured the delivered tidal volume while the ventilator was connected to an external, gradually varying resistance. This resistance was progressively increased to simulate bronchoconstriction of the respiratory system. Comparison with a volume-controlled ventilator was made. The use of a programmable controller also allows control of different patterns of mechanical ventilation, such as end-inspiratory pause or the static pressure-volume relationship, which can be used to perform lung function tests. The system is a simple, versatile device allowing both reliable mechanical ventilation and lung function assessment in small rodents and is suitable for routine use in laboratories.

Sequence of pattern onset responses in the human visual areas: an fMRI constrained VEP source analysis.

We measured the timing of activity in distinct functional areas of the human visual cortex after onset of a visual pattern. This is not possible with visual evoked potentials (VEPs) or magnetic fields alone, and direct combination of functional magnetic resonance imaging (fMRI) with electromagnetic data has turned out to be difficult. We tested a relatively new approach, where both position and orientation of the active cortex was given to the VEP source model. Subjects saw the same visual patterns flashed ON and OFF, both when recording VEPs and fMRI responses. We identified the positions and orientations of the activated cortex in four retinotopic areas in each individual, and the corresponding dipoles were seeded to model the individual evoked potential data. Unexplained variance, comprising signals from other areas, was inversely modeled. Despite the partially a priori fixed model and optimized signal-to-noise ratio of VEP data, full separation of retinotopic areas was only seldom possible due to crosstalk between the adjacent sources, but separation was usually possible between areas V1 and V3/V3a. Whereas the latencies generally followed the hierarchical organization of cortical areas (V1-V2-V3), with around 25 ms between the strongest responses, an early activation emerged 10-20 ms after V1, close to the temporo-occipital junction (LO/V5) and with an additional 20-ms latency in the corresponding region of the opposite hemisphere. Our approach shows that it is feasible to directly seed information from fMRI to electromagnetic source models and to identify the components and dynamics of VEPs in different retinotopic areas of a human individual.

Timing of interactions across the visual field in the human cortex.

While it is generally believed that interactions across long distances in the visual field occur only in the higher-order cortical areas, other results suggest that such interactions are processed very early. In the preceding paper, we identified the latencies within a subset of cortical areas in the human visual system. In the present study, we test in which areas and at which latencies the responses to two visual patterns start interacting. We used functional magnetic resonance imaging directly combined with visual-evoked potential source analysis. Interactions appeared first anterolaterally to the retinotopic areas, at 80 ms for two stimuli presented in the left lower visual quadrant and at 100 ms for symmetrical stimulation of both lower quadrants. In the lateral occipital-V5 region (LOV5), two patterns presented simultaneously in one quadrant elicited a response with shorter latency and infra-linear addition of the amplitudes compared with the patterns presented separately. For bilateral stimulation, the timing of the LOV5 response coincided with the response to contralateral stimulation alone. Other visual areas showed interactions appearing later than within LOV5: starting at 150 ms in V1, at 120 ms in V3-V3a for the left visual hemifield stimulation and at 160 ms for both visual hemifields stimulation. Our data show that distinct patterns in the visual field interact first in LOV5, suggesting that this region must be the first to pool spatial information across the whole visual field.

Clinical evaluation of a computer-controlled pressure support mode.

We have designed a computerized system providing closed-loop control of the level of pressure support ventilation (PSV). The system sets itself at the lowest level of PSV that maintains respiratory rate (RR), tidal volume (VT), and end-tidal CO(2) pressure (PET(CO(2))) within predetermined ranges defining acceptable ventilation (i.e., 12 < RR < 28 cycles/min, VT > 300 ml [> 250 if weight < 55 kg], and PET(CO(2)) < 55 mm Hg [< 65 mm Hg if chronic CO(2) retention]). Ten patients received computer-controlled (automatic) PSV and physician-controlled (standard) PSV, in random order, during 24 h for each mode. An estimation of occlusion pressure (P(0.1)) was recorded continuously. The average time spent with acceptable ventilation as previously defined was 66 +/- 24% of the total ventilation time with standard PSV versus 93 +/- 8% with automatic PSV (p < 0.05), whereas the level of PSV was similar during the two periods (17 +/- 4 cm H(2)O versus 19 +/- 6 cm H(2)O). The time spent with an estimated P(0.1) above 4 cm H(2)O was 34 +/- 35% of the standard PSV time versus only 11 +/- 17% of the automatic PSV time (p < 0.01). Automatic PSV increased the time spent within desired ventilation parameter ranges and apparently reduced periods of excessive workload.

Evaluation of a knowledge-based system providing ventilatory management and decision for extubation.

We evaluated whether a knowledge-based system (KBS) connected to a ventilator in pressure support mode could correctly predict the ability of patients to tolerate total withdrawal from ventilatory support. The KBS was designed to continuously adapt ventilatory assistance to the needs of the patient, to manage a strategy of gradually decreasing ventilatory assistance, and to indicate when the patient was able to breathe without assistance. Thirty-eight patients for whom weaning was being considered were evaluated using a conventional battery of parameters, including weaning criteria, tolerance of a T-piece trial, and outcome 48h after permanent withdrawal of ventilation. The results of this evaluation were compared with the suggestions made by the KBS at the end of a period of KBS-driven mechanical ventilation inserted in the conventional weaning procedure. The positive predictive value of the KBS was 89%, versus 77% for the conventional procedure and 81% for the rapid shallow breathing index alone. The KBS correctly predicted the course of five patients who tolerated a T-piece trial but required ventilation within 48 h. We conclude that our KBS ensured appropriate patient management during the weaning period and improved our ability to predict responses to weaning.

Knowledge acquisition environment for the design of a decision support system: application in blood transfusion.

Blood transfusion is a medical domain where decision support systems (DSSs) could be very helpful to the physicians but must easily and continuously be maintained. We have developed a knowledge acquisition tool that allows the construction and the maintenance of such a system by the domain expert. The methodology used could be applied to another highly evolutive medical domain. In this paper, we detail our knowledge acquisition tool, its use and the final DSS obtained, which is fully integrated into our hospital information network.

Moving illusory contours activate primary visual cortex: an fMRI study.

Identifying the cortical areas activated by illusory contours provides valuable information on the mechanisms of object perception. We applied functional magnetic resonance imaging to identify the visual areas of the human brain involved in the perception of a moving Kanizsa-type illusory contour. Our results indicate that, in addition to other cortical regions, areas V5 and V1 are activated. Activity in area V1 was particularly prominent.