11 research outputs found
Electroencéphalographie et interfaces cerveau-machine : nouvelles méthodes pour étudier les états mentaux
No full textAvec les avancées technologiques dans le domaine de l'imagerie cérébrale fonctionnelle et les progrès théoriques dans la connaissance des différents éléments neurophysiologiques liés à la cognition, les deux dernières décennies ont vu l'apparition d'interfaces cerveau-machine (ICM) permettant à une personne d'observer en temps réel, ou avec un décalage qui se limite à quelques secondes, sa propre activité cérébrale. Le domaine clinique en général, et plus particulièrement celui de la neuropsychologie et des pathologies conduisant à un handicap moteur lourd, pour lesquels les applications potentielles sont nombreuses qu'elles soient thérapeutiques ou en vue d'une réhabilitation fonctionnelle, a constitué un moteur important de la recherche sur ce nouveau domaine des neurosciences temps réel. Parmi ces applications, le neurofeedback, ou neurothérapie, qui vise l'acquisition par le sujet du contrôle volontaire de certains aspects de son activité cérébrale en vue de les amplifier ou au contraire les diminuer dans un but thérapeutique, voire d'optimisation cognitive, représente une technique prometteuse, alternative aux thérapies et traitements médicamenteux. Cependant, la validation de ce type d'intervention et la compréhension des mécanismes mis en jeux en sont encore à leurs balbutiements. L'entraînement par neurofeedback est souvent long, pouvant s'étaler sur plusieurs semaines. Il est donc très probable que ce type de rééducation cérébrale sollicite des phénomènes de plasticité qui s'inscrivent dans une dynamique lente, et de ce fait, requiert une durée relativement longue d'entraînement pour atteindre les effets à long terme recherchés. Cependant, à cela peuvent s'ajouter de nombreux éléments perturbateurs qui pourraient être à l'origine de la difficulté de l'apprentissage et des longs entraînements nécessaires pour obtenir les résultats attendus. Parmi eux, les perturbations qui viennent déformer le signal enregistré, ou les éléments artefactuels qui ne font pas partie du signal d'intérêt, sont une première cause potentielle. Le manque de spécificité fonctionnelle du signal retourné au sujet pourrait en constituer une deuxième. Nous avons d'une part développé des outils méthodologiques de traitement du signal en vue d'améliorer la robustesse des analyses des signaux EEG, principalement utilisés jusqu'à maintenant dans le domaine du neurofeedback et des ICM, face aux artefacts et au bruit électromagnétique. D'autre part, si l'on s'intéresse au problème de la spécificité fonctionnelle du signal présenté au sujet, des études utilisant l'IRM fonctionnelle ou des techniques de reconstruction de sources à partir du signal EEG, qui fournissent des signaux ayant une meilleure spécificité spatiale, laissent entrevoir de possibles améliorations de la vitesse d'apprentissage. Afin d'augmenter la spécificité spatiale et la contingence fonctionnelle du feedback présenté au sujet, nous avons étudié la stabilité de la décomposition de l'EEG en différentes sources d'activité électrique cérébrale par Analyse en Composantes Indépendantes à travers différentes séances d'enregistrement effectuées sur un même sujet. Nous montrons que ces décompositions sont stables et pourraient permettre d'augmenter la spécificité fonctionnelle de l'entraînement au contrôle de l'activité cérébrale pour l'utilisation d'une ICM. Nous avons également travaillé à l'implémentation d'un outil logiciel permettant l'optimisation des protocoles expérimentaux basés sur le neurofeedback afin d'utiliser ces composantes indépendantes pour rejeter les artefacts en temps réel ou extraire l'activité cérébrale à entraîner. Ces outils sont utiles dans le cadre de l'analyse et de la caractérisation des signaux EEG enregistrés, ainsi que dans l'exploitation de leurs résultats dans le cadre d'un entraînement de neurofeedback. La deuxième partie de ce travail s'intéresse à la mise en place de protocoles de neurofeedback et à l'impact de l'apprentissage. Nous décrivons tout d'abord des résultats obtenus sur une étude pilote qui cherche à évaluer chez des sujets sains l'impact d'un protocole de neurofeedback basé sur le contrôle du rythme Mu. Les changements comportementaux ont été étudiés à l'aide d'un paradigme de signal stop qui permet d'indexer les capacités attentionnelles et d'inhibition de réponse motrice sur lesquelles on s'attend à ce que l'entraînement ICM ait une influence. Pour clore cette partie, nous présentons un nouvel outil interactif immersif pour l'entraînement cérébral, l'enseignement, l'art et le divertissement pouvant servir à évaluer l'impact de l'immersion sur l'apprentissage au cours d'un protocole de neurofeedback. Enfin, les perspectives de l'apport des méthodes et résultats présentés sont discutées dans le contexte du développement des ICMs de nouvelle génération qui prennent en compte la complexité de l'activité cérébrale. Nous présentons les dernières avancées dans l'étude de certains aspects des corrélats neuronaux liés à deux états mentaux ou classes d'états mentaux que l'on pourrait qualifier d'antagonistes par rapport au contrôle de l'attention : la méditation et la dérive attentionnelle, en vue de leur intégration à plus long terme dans un entraînement ICM par neurofeedback.With new technological advances in functional brain imaging and theoretical progress in the knowledge of the different neurophysiologic processes linked to cognition, the last two decades have seen the emergence of Brain-Machine Interfaces (BCIs) allowing a person to observe in real-time, or with a few seconds delay, his own cerebral activity. Clinical domain in general, and more particularly neuropsychology and pathologies leading to heavy motor handicaps, for which potential applications are numerous, whether therapeutic or for functional rehabilitation, has been a major driver of research on this new field of real-time neurosciences. Among these applications, neurofeedback, or neurotherapy, which aims the subject to voluntary control some aspects of his own cerebral activity in order to amplify or reduce them in a therapeutic goal, or for cognitive optimization, represents a promising technique, and an alternative to drug treatments. However, validation of this type of intervention and understanding of involved mechanisms are still in their infancy. Neurofeedback training is often long, up to several weeks. It is therefore very likely that this type of rehabilitation is seeking brain plasticity phenomena that are part of slow dynamics, and thus require a relatively long drive to achieve the desired long-term effects. However, other disturbing elements that could add up to the cause of the difficulty of learning and long training sessions required to achieve the expected results. Among them, the disturbances that come from recorded signal distortions, or artifactual elements that are not part of the signal of interest, are a first potential cause. The lack of functional specificity of the signal returned to the subject could be a second one. We have developed signal processing methodological tools to improve the robustness to artifacts and electromagnetic noise of EEG signals analysis, the main brain imaging technique used so far in the field of neurofeedback and BCIs. On the other hand, if one looks at the issue of functional specificity of the signal presented to the subject, studies using functional MRI or source reconstruction methods from the EEG signal, which both provide signals having a better spatial specificity, suggest improvements to the speed of learning. Seeing Independent Component Analysis as a potential tool to increase the spatial specificity and functional contingency of the feedback signal presented to the subject, we studied the stability of Independent Component Analysis decomposition of the EEG across different recording sessions conducted on the same subjects. We show that these decompositions are stable and could help to increase the functional specificity of BCI training. We also worked on the implementation of a software tool that allows the optimization of experimental protocols based on neurofeedback to use these independent components to reject artifacts or to extract brain activity in real-time. These tools are useful in the analysis and characterization of EEG signals recorded, and in the exploitation of their results as part of a neurofeedback training. The second part focuses on the development of neurofeedback protocols and the impact of learning. We first describe the results of a pilot study which seeks to evaluate the impact of a neurofeedback protocol based on the Mu rhythm control on healthy subjects. The behavioral changes were studied using a stop signal paradigm that indexes the attentional abilities and inhibition of motor responses on which the BCI training can possibly have influence. To conclude this section, we present a new tool for immersive interactive brain training, education, art and entertainment that can be used to assess the impact of immersion on learning during a neurofeedback protocol. Finally, prospects for methods and results presented are discussed in the context of next-generation BCI development which could take brain activity complexity into account. We present the latest advances in the study of certain aspects of the neural correlates associated with two mental states or classes of mental states that could be described as antagonistic with respect to the control of attention: meditation and mind wandering, for their integration in the longer term in an BCI training using neurofeedback
Emergence of Clonally-Related South Asian Clade I Clinical Isolates of Candida auris in a Greek COVID-19 Intensive Care Unit
No full textCandida auris has recently emerged as a multidrug-resistant yeast implicated in various healthcare-associated invasive infections and hospital outbreaks. In the current study, we report the first five intensive care unit (ICU) cases affected by C. auris isolates in Greece, during October 2020–January 2022. The ICU of the hospital was converted to a COVID-19 unit on 25 February 2021, during the third wave of COVID-19 in Greece. Identification of the isolates was confirmed by Matrix Assisted Laser Desorption Ionization Time of Flight mass spectroscopy (MALDI-TOF]. Antifungal susceptibility testing was performed by the EUCAST broth microdilution method. Based on the tentative CDC MIC breakpoints, all five C. auris isolates were resistant to fluconazole (≥32 μg/mL), while three of them exhibited resistance to amphotericin B (≥2 μg/mL). The environmental screening also revealed the dissemination of C. auris in the ICU. Molecular characterization of C. auris clinical and environmental isolates was performed by MultiLocus Sequence Typing (MLST) of a set of four genetic loci, namely ITS, D1/D2, RPB1 and RPB2, encoding for the internal transcribed spacer region (ITS) of the ribosomal subunit, the large ribosomal subunit region and the RNA polymerase II largest subunit, respectively. MLST analysis showed that all isolates possessed identical sequences in the four genetic loci and clustered with the South Asian clade I strains. Additionally, PCR amplification and sequencing of the CJJ09_001802 genetic locus, encoding for the “nucleolar protein 58” that contains clade-specific repeats was performed. Sanger sequence analysis of the TCCTTCTTC repeats within CJJ09_001802 locus also assigned the C. auris isolates to the South Asian clade I. Our study confirms that C. auris is an emerging yeast pathogen in our region, especially in the setting of the ongoing COVID-19 worldwide pandemic. Adherence to strict infection control is needed to restrain further spread of the pathogen
The role of centre and country factors on process and outcome indicators in critically ill patients with hospital-acquired bloodstream infections
No full textPurpose: The primary objective of this study was to evaluate the associations between centre/country-based factors and two important process and outcome indicators in patients with hospital-acquired bloodstream infections (HABSI). Methods: We used data on HABSI from the prospective EUROBACT-2 study to evaluate the associations between centre/country factors on a process or an outcome indicator: adequacy of antimicrobial therapy within the first 24 h or 28-day mortality, respectively. Mixed logistical models with clustering by centre identified factors associated with both indicators. Results: Two thousand two hundred nine patients from two hundred one intensive care units (ICUs) were included in forty-seven countries. Overall, 51% (n = 1128) of patients received an adequate antimicrobial therapy and the 28-day mortality was 38% (n = 839). The availability of therapeutic drug monitoring (TDM) for aminoglycosides everyday [odds ratio (OR) 1.48, 95% confidence interval (CI) 1.03-2.14] or within a few hours (OR 1.79, 95% CI 1.34-2.38), surveillance cultures for multidrug-resistant organism carriage performed weekly (OR 1.45, 95% CI 1.09-1.93), and increasing Human Development Index (HDI) values were associated with adequate antimicrobial therapy. The presence of intermediate care beds (OR 0.63, 95% CI 0.47-0.84), TDM for aminoglycoside available everyday (OR 0.66, 95% CI 0.44-1.00) or within a few hours (OR 0.51, 95% CI 0.37-0.70), 24/7 consultation of clinical pharmacists (OR 0.67, 95% CI 0.47-0.95), percentage of vancomycin-resistant enterococci (VRE) between 10% and 25% in the ICU (OR 1.67, 95% CI 1.00-2.80), and decreasing HDI values were associated with 28-day mortality. Conclusion: Centre/country factors should be targeted for future interventions to improve management strategies and outcome of HABSI in ICU patients
Effect of adequacy of empirical antibiotic therapy for hospital-acquired bloodstream infections on ICU patient prognosis: a causal inference approach using data from the Eurobact2 study
No full textObjectives: Hospital-acquired bloodstream infections (HA-BSI) in the intensive care unit (ICU) are common life-threatening events. We wanted to investigate the association between early adequate antibiotic therapy and 28-day mortality in ICU patients surviving for at least 1 day after the onset of HA-BSI. Methods: We used individual data from a prospective, observational, multicenter, intercontinental cohort study (Eurobact2). We included patients followed for ≥1 day for whom time-to-appropriate treatment was available. We used an adjusted frailty-Cox proportional hazard model to assess the effect of time-to-treatment-adequacy on 28-day mortality. Infection- and patient-related variables identified as confounders by the Directed Acyclic Graph were used for adjustment. Adequate therapy within 24 hours was used for primary analysis. Secondary analyses were performed for adequate therapy within 48 and 72 hours and for identified patient subgroups. Results: Among the 2,418 patients included in 330 centers worldwide, 28-day mortality was 32.8% (n=402/1226) in patients who were adequately treated within 24 hours after HA-BSI onset and 40% (n=477/1192) in inadequately treated patients (p<0.01). Adequacy within 24 hours was more common in young, immunosuppressed patients, and with HA-BSI due to Gram-negative pathogens. Antimicrobial adequacy was significantly associated with 28-day survival (aHR 0.83, 95% CI 0.72-0.96, p=0.01). The estimated population attributable fraction (PAF) of 28-day mortality of inadequate therapy was 9.15% (95% CI 1.9%-16.2%). Conclusions: In patients with HA-BSI admitted in ICU, the PAF of 28-day mortality of inadequate therapy within 24 hours was 9.15%. This estimate should be used when hypothesizing the possible benefit of any intervention aiming at reducing the time-to-appropriate antimicrobial therapy in HA-BSI
Correction: Shortening antibiotic therapy duration for hospital-acquired bloodstream infections in critically ill patients: a causal inference model from the international EUROBACT-2 database
No full textPresentation, management, and outcomes of older compared to younger adults with hospital-acquired bloodstream infections in the intensive care unit: a multicenter cohort study
No full textPurpose: Older adults admitted to the intensive care unit (ICU) usually have fair baseline functional capacity, yet their age and frailty may compromise their management. We compared the characteristics and management of older (≥ 75 years) versus younger adults hospitalized in ICU with hospital-acquired bloodstream infection (HA-BSI). Methods: Nested cohort study within the EUROBACT-2 database, a multinational prospective cohort study including adults (≥ 18 years) hospitalized in the ICU during 2019-2021. We compared older versus younger adults in terms of infection characteristics (clinical signs and symptoms, source, and microbiological data), management (imaging, source control, antimicrobial therapy), and outcomes (28-day mortality and hospital discharge). Results: Among 2111 individuals hospitalized in 219 ICUs with HA-BSI, 563 (27%) were ≥ 75 years old. Compared to younger patients, these individuals had higher comorbidity score and lower functional capacity; presented more often with a pulmonary, urinary, or unknown HA-BSI source; and had lower heart rate, blood pressure and temperature at presentation. Pathogens and resistance rates were similar in both groups. Differences in management included mainly lower rates of effective source control achievement among aged individuals. Older adults also had significantly higher day-28 mortality (50% versus 34%, p < 0.001), and lower rates of discharge from hospital (12% versus 20%, p < 0.001) by this time. Conclusions: Older adults with HA-BSI hospitalized in ICU have different baseline characteristics and source of infection compared to younger patients. Management of older adults differs mainly by lower probability to achieve source control. This should be targeted to improve outcomes among older ICU patients
Positive communication for decreasing burnout in intensive-care-unit staff: a cluster-randomized trial.
No full textPurpose: Occupational burnout is common among intensive-care-unit (ICU) staff and adversely affects staff well-being and patient care. We hypothesized that a multicomponent intervention based on organizational support and workplace climate improvement would reduce burnout. Methods: The 1:1 cluster-randomized Hello trial involved 370 ICUs from sixty countries allocated to either the intervention or usual care. The four-week intervention designed to promote a positive workplace culture and within-team support used posters, email nudges, greetings during morning meetings, role modeling, and positive messages in boxes and on noticeboards. The primary endpoint was burnout prevalence, measured using the Maslach Burnout Inventory. Secondary outcomes included MBI subscale scores, well-being, job satisfaction, ethical climate, intention to leave, work safety, and professional conflicts. Results: Before the intervention, burnout prevalence was 59.4% (95% CI, 58.6-60.5), with no difference between arms. After the intervention, 4966 intervention-arm and 4602 control-arm healthcare professionals completed the MBI. Burnout prevalence was significantly lower in the intervention arm relative to controls (52.2% vs. 63.3%; adjusted odds ratio, 0.56; 95%CI 0.46-0.68; P < 0.001). Among MBI sub-scales scores, emotional exhaustion and depersonalization were lower, and personal accomplishment was higher in the intervention arm. Staff in the intervention arm reported better job satisfaction, workplace safety, ethical climate, and patient- and family-centered care; they were less often considering a job change. Conclusions: The Hello intervention reduced burnout and improved workplace culture among ICU staff. Given the pragmatic design, the intervention tested may have broad applicability. Trial registration: The trial was registered on ClinicalTrials.gov on June 18, 2024 (NCT06453616)
An information theoretic learning framework based on Renyi’s α entropy for brain effective connectivity estimation
No full textThe interactions among neural populations distributed across different brain regions are at the core of cognitive and perceptual processing. Therefore, the ability of studying the flow of information within networks of connected neural assemblies is of fundamental importance to understand such processes. In that regard, brain connectivity measures constitute a valuable tool in neuroscience. They allow assessing functional interactions among brain regions through directed or non-directed statistical dependencies estimated from neural time series. Transfer entropy (TE) is one such measure. It is an effective connectivity estimation approach based on information theory concepts and statistical causality premises. It has gained increasing attention in the literature because it can capture purely nonlinear directed interactions, and is model free. That is to say, it does not require an initial hypothesis about the interactions present in the data. These properties make it an especially convenient tool in exploratory analyses. However, like any information-theoretic quantity, TE is defined in terms of probability distributions that in practice need to be estimated from data. A challenging task, whose outcome can significantly affect the results of TE. Also, it lacks a standard spectral representation, so it cannot reveal the local frequency band characteristics of the interactions it detects.Las interacciones entre poblaciones neuronales distribuidas en diferentes regiones del cerebro son el núcleo del procesamiento cognitivo y perceptivo. Por lo tanto, la capacidad de estudiar el flujo de información dentro de redes de conjuntos neuronales conectados es de fundamental importancia para comprender dichos procesos. En ese sentido, las medidas de conectividad cerebral constituyen una valiosa herramienta en neurociencia. Permiten evaluar interacciones funcionales entre regiones cerebrales a través de dependencias estadísticas dirigidas o no dirigidas estimadas a partir de series de tiempo. La transferencia de entropía (TE) es una de esas medidas. Es un enfoque de estimación de conectividad efectiva basada en conceptos de teoría de la información y premisas de causalidad estadística. Ha ganado una atención cada vez mayor en la literatura porque puede capturar interacciones dirigidas puramente no lineales y no depende de un modelo. Es decir, no requiere de una hipótesis inicial sobre las interacciones presentes en los datos. Estas propiedades la convierten en una herramienta especialmente conveniente en análisis exploratorios. Sin embargo, como cualquier concepto basado en teoría de la información, la TE se define en términos de distribuciones de probabilidad que en la práctica deben estimarse a partir de datos. Una tarea desafiante, cuyo resultado puede afectar significativamente los resultados de la TE. Además, carece de una representación espectral estándar, por lo que no puede revelar las características de banda de frecuencia local de las interacciones que detecta.DoctoradoDoctor(a) en IngenieríaContents
List of Figures xi
List of Tables xv
Notation xvi
1 Preliminaries 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Probability distribution estimation as an intermediate step in TE
computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 The lack of a spectral representation for TE . . . . . . . . . . . . 7
1.3 Theoretical background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Transfer entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Granger causality . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.3 Information theoretic learning from kernel matrices . . . . . . . . 12
1.4 Literature review on transfer entropy estimation . . . . . . . . . . . . . . 14
1.4.1 Transfer entropy in the frequency domain . . . . . . . . . . . . . . 17
1.5 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.1 General aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.2 Specific aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6 Outline and contributions . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.6.1 Kernel-based Renyi’s transfer entropy . . . . . . . . . . . . . . . . 24
1.6.2 Kernel-based Renyi’s phase transfer entropy . . . . . . . . . . . . 24
1.6.3 Kernel-based Renyi’s phase transfer entropy for the estimation of
directed phase-amplitude interactions . . . . . . . . . . . . . . . . 25
1.7 EEG databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Contents ix
1.7.1 Motor imagery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.7.2 Working memory . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.8 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2 Kernel-based Renyi’s transfer entropy 34
2.1 Kernel-based Renyi’s transfer entropy . . . . . . . . . . . . . . . . . . . . 35
2.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.1 VAR model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.2 Modified linear Kus model . . . . . . . . . . . . . . . . . . . . . . 38
2.2.3 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2.4 Parameter selection . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.1 VAR model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.2 Modified linear Kus model . . . . . . . . . . . . . . . . . . . . . . 46
2.3.3 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3.4 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3 Kernel-based Renyi’s phase transfer entropy 60
3.1 Kernel-based Renyi’s phase transfer entropy . . . . . . . . . . . . . . . . 61
3.1.1 Phase-based effective connectivity estimation approaches considered
in this chapter . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.1 Neural mass models . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.2.3 Parameter selection . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.1 Neural mass models . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3.3 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4 Kernel-based Renyi’s phase transfer entropy for the estimation of directed
phase-amplitude interactions 84
4.1 Kernel-based Renyi’s phase transfer entropy for the estimation of directed
phase-amplitude interactions . . . . . . . . . . . . . . . . . . . . . . . . . 85
x Contents
4.1.1 Transfer entropy for directed phase-amplitude interactions . . . . 85
4.1.2 Cross-frequency directionality . . . . . . . . . . . . . . . . . . . . 85
4.1.3 Phase transfer entropy and directed phase-amplitude interactions 86
4.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.2.1 Simulated phase-amplitude interactions . . . . . . . . . . . . . . . 88
4.2.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.2.3 Parameter selection . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.3.1 Simulated phase-amplitude interactions . . . . . . . . . . . . . . . 92
4.3.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.3.3 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5 Final Remarks 100
5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.3 Academic products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.1 Journal papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.2 Conference papers . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.3.3 Conference presentations . . . . . . . . . . . . . . . . . . . . . . . 105
Appendix A Kernel methods and Renyi’s entropy estimation 106
A.1 Reproducing kernel Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . 106
A.1.1 Reproducing kernels . . . . . . . . . . . . . . . . . . . . . . . . . 106
A.1.2 Kernel-based learning . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.2 Kernel-based estimation of Renyi’s entropy . . . . . . . . . . . . . . . . . 109
Appendix B Surface Laplacian 113
Appendix C Permutation testing 115
Appendix D Kernel-based relevance analysis 117
Appendix E Cao’s criterion 120
Appendix F Neural mass model equations 122
References 12
An information theoretic learning framework based on Renyi’s α entropy for brain effective connectivity estimation
No full textThe interactions among neural populations distributed across different brain regions are at the core of cognitive and perceptual processing. Therefore, the ability of studying the flow of information within networks of connected neural assemblies is of fundamental importance to understand such processes. In that regard, brain connectivity measures constitute a valuable tool in neuroscience. They allow assessing functional interactions among brain regions through directed or non-directed statistical dependencies estimated from neural time series. Transfer entropy (TE) is one such measure. It is an effective connectivity estimation approach based on information theory concepts and statistical causality premises. It has gained increasing attention in the literature because it can capture purely nonlinear directed interactions, and is model free. That is to say, it does not require an initial hypothesis about the interactions present in the data. These properties make it an especially convenient tool in exploratory analyses. However, like any information-theoretic quantity, TE is defined in terms of probability distributions that in practice need to be estimated from data. A challenging task, whose outcome can significantly affect the results of TE. Also, it lacks a standard spectral representation, so it cannot reveal the local frequency band characteristics of the interactions it detects.Las interacciones entre poblaciones neuronales distribuidas en diferentes regiones del cerebro son el núcleo del procesamiento cognitivo y perceptivo. Por lo tanto, la capacidad de estudiar el flujo de información dentro de redes de conjuntos neuronales conectados es de fundamental importancia para comprender dichos procesos. En ese sentido, las medidas de conectividad cerebral constituyen una valiosa herramienta en neurociencia. Permiten evaluar interacciones funcionales entre regiones cerebrales a través de dependencias estadísticas dirigidas o no dirigidas estimadas a partir de series de tiempo. La transferencia de entropía (TE) es una de esas medidas. Es un enfoque de estimación de conectividad efectiva basada en conceptos de teoría de la información y premisas de causalidad estadística. Ha ganado una atención cada vez mayor en la literatura porque puede capturar interacciones dirigidas puramente no lineales y no depende de un modelo. Es decir, no requiere de una hipótesis inicial sobre las interacciones presentes en los datos. Estas propiedades la convierten en una herramienta especialmente conveniente en análisis exploratorios. Sin embargo, como cualquier concepto basado en teoría de la información, la TE se define en términos de distribuciones de probabilidad que en la práctica deben estimarse a partir de datos. Una tarea desafiante, cuyo resultado puede afectar significativamente los resultados de la TE. Además, carece de una representación espectral estándar, por lo que no puede revelar las características de banda de frecuencia local de las interacciones que detecta.Contents
List of Figures xi
List of Tables xv
Notation xvi
1 Preliminaries 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Probability distribution estimation as an intermediate step in TE
computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.2 The lack of a spectral representation for TE . . . . . . . . . . . . 7
1.3 Theoretical background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Transfer entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Granger causality . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.3 Information theoretic learning from kernel matrices . . . . . . . . 12
1.4 Literature review on transfer entropy estimation . . . . . . . . . . . . . . 14
1.4.1 Transfer entropy in the frequency domain . . . . . . . . . . . . . . 17
1.5 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.1 General aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.2 Specific aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.6 Outline and contributions . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.6.1 Kernel-based Renyi’s transfer entropy . . . . . . . . . . . . . . . . 24
1.6.2 Kernel-based Renyi’s phase transfer entropy . . . . . . . . . . . . 24
1.6.3 Kernel-based Renyi’s phase transfer entropy for the estimation of
directed phase-amplitude interactions . . . . . . . . . . . . . . . . 25
1.7 EEG databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Contents ix
1.7.1 Motor imagery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.7.2 Working memory . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.8 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2 Kernel-based Renyi’s transfer entropy 34
2.1 Kernel-based Renyi’s transfer entropy . . . . . . . . . . . . . . . . . . . . 35
2.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.1 VAR model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.2 Modified linear Kus model . . . . . . . . . . . . . . . . . . . . . . 38
2.2.3 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2.4 Parameter selection . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.1 VAR model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.2 Modified linear Kus model . . . . . . . . . . . . . . . . . . . . . . 46
2.3.3 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.3.4 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3 Kernel-based Renyi’s phase transfer entropy 60
3.1 Kernel-based Renyi’s phase transfer entropy . . . . . . . . . . . . . . . . 61
3.1.1 Phase-based effective connectivity estimation approaches considered
in this chapter . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.1 Neural mass models . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.2.3 Parameter selection . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.1 Neural mass models . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3.3 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4 Kernel-based Renyi’s phase transfer entropy for the estimation of directed
phase-amplitude interactions 84
4.1 Kernel-based Renyi’s phase transfer entropy for the estimation of directed
phase-amplitude interactions . . . . . . . . . . . . . . . . . . . . . . . . . 85
x Contents
4.1.1 Transfer entropy for directed phase-amplitude interactions . . . . 85
4.1.2 Cross-frequency directionality . . . . . . . . . . . . . . . . . . . . 85
4.1.3 Phase transfer entropy and directed phase-amplitude interactions 86
4.2 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.2.1 Simulated phase-amplitude interactions . . . . . . . . . . . . . . . 88
4.2.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.2.3 Parameter selection . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.3.1 Simulated phase-amplitude interactions . . . . . . . . . . . . . . . 92
4.3.2 EEG data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.3.3 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5 Final Remarks 100
5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.3 Academic products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.1 Journal papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.2 Conference papers . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.3.3 Conference presentations . . . . . . . . . . . . . . . . . . . . . . . 105
Appendix A Kernel methods and Renyi’s entropy estimation 106
A.1 Reproducing kernel Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . 106
A.1.1 Reproducing kernels . . . . . . . . . . . . . . . . . . . . . . . . . 106
A.1.2 Kernel-based learning . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.2 Kernel-based estimation of Renyi’s entropy . . . . . . . . . . . . . . . . . 109
Appendix B Surface Laplacian 113
Appendix C Permutation testing 115
Appendix D Kernel-based relevance analysis 117
Appendix E Cao’s criterion 120
Appendix F Neural mass model equations 122
References 125DoctoradoDoctor(a) en Ingenierí
Shortening antibiotic therapy duration for hospital-acquired bloodstream infections in critically ill patients: a causal inference model from the international EUROBACT-2 database
No full textIntroduction: Hospital-acquired bloodstream infections (HA-BSIs) are severe and require antibiotic therapy. In non-complicated BSIs, shortened therapy reduces side effects without compromising efficacy. The impact of shortened antibiotic therapy in HA-BSI critically ill patients without indication of prolonged therapy requires further evaluation. Methods: Using the international prospective EUROBACT-2 cohort, we compared shortened (7-10 days) versus long (14-21 days) treatment durations in ICU patients eligible for shortened therapy. Patients without antibiotic therapy within 3 days after HA-BSI occurrence or requiring prolonged therapy (due to infection source, microorganism, or clinical deterioration) were excluded. Treatment failure, defined as death, persistent infection, or subsequent infectious complications by Day 28, was assessed using an inverse-probability of treatment weighted (IPTW) logistic regression. Results: Among 2600 patients, 550 were eligible for shortened treatment, 213 received short, and 337 received long treatment. The most common infection source was intravascular catheters (33%), most common microorganisms were Enterobacterales (39%). Patients with long treatment were more frequently infected with Staphylococcus aureus (11% vs. 5.6%, p = 0.025) or difficult-to-treat microorganisms (23% vs. 7%, p < 0.001), and received more commonly combination therapy (46% vs. 30%, p < 0.001). Short treatment was associated with reduced 28-day treatment failure (OR 0.64, 95% CI 0.44-0.93, p = 0.019), mainly due to reduction in subsequent infectious complications (OR 0.58, 95% CI 0.37-0.91, p = 0.018). Mortality (OR 0.92 [95% CI 0.59, 1.43], p = 0.7) and persistent infection rates (OR 0.47 [95% CI 0.17, 1.14], p = 0.12) were similar. Conclusions: In selected ICU patients with HA-BSI, shortened antibiotic treatment might be considered. Eurobact2 was a prospective international cohort study, registered in ClinicalTrials.org (NCT03937245)
