Exploring time-varying connectivity networks in neurodegenerative disorders is a recent field of research in functional MRI. Dementia with Lewy bodies (DLB) represents 20% of the neurodegenerative forms of dementia. Fluctuations of cognition and vigilance are the key symptoms of DLB. To date, no dynamic functional connectivity (DFC) investigations of this disorder have been performed. In this paper, we refer to the concept of connectivity state as a piecewise stationary configuration of functional connectivity between brain networks. From this concept, we propose a new method for group-level as well as for subject-level studies to compare and characterize connectivity state changes between a set of resting-state networks (RSNs). Dynamic Bayesian networks, statistical and graph theory-based models, enable one to learn dependencies between interacting state-based processes. Product hidden Markov models (PHMM), an instance of dynamic Bayesian networks, are introduced here to capture both statistical and temporal aspects of DFC of a set of RSNs. This analysis was based on sliding-window cross-correlations between seven RSNs extracted from a group independent component analysis performed on 20 healthy elderly subjects and 16 patients with DLB. Statistical models of DFC differed in patients compared to healthy subjects for the occipito-parieto-frontal network, the medial occipital network and the right fronto-parietal network. In addition, pairwise comparisons of DFC of RSNs revealed a decrease of dependency between these two visual networks (occipito-parieto-frontal and medial occipital networks) and the right fronto-parietal control network. The analysis of DFC state changes thus pointed out networks related to the cognitive functions that are known to be impaired in DLB: visual processing as well as attentional and executive functions. Besides this context, product HMM applied to RSNs cross-correlations offers a promising new approach to investigate structural and temporal aspects of brain DFC.

Information management for critical care monitoring is still a very difficult task. Medical staff is often overwhelmed by the amount of data provided by the increased number of specific monitoring devices and instrumentation, and the lack of an effective automated system. Specifically, a basic task such as arrhythmia detection still produce an important amount of undesirable alarms, due in part to the mechanistic approach of current monitoring systems. In this work, multisensor and multisource data fusion schemes to improve atrial and ventricular activity detection in critical care environments are presented. Applications of these schemes are quantitatively evaluated and compared with current methods, showing the potential advantages of data fusion techniques for event detection in noise corrupted signals.

Abstract:Wave recognition in ECG signals by Hidden Markov Models (HMMs) relies on the stationary assumption for the set of parameters used to describe ECG waves. This approach seems unnatural and consequently generates severe errors in practice. A new class of HMMs called Modified Continuous Variable Duration HMMs is proposed to account for the specific properties of the ECG signal. An application of the latter, coupled with a multiresolution front-end analysis of the ECG is presented. Results show these methods can increase the perfomance of ECG recognition compared to classical HMMs.