Tesi di dottorato >
Scienze mediche >
Please use this identifier to cite or link to this item:
|Title: ||Biostatistical tools in neurosciences|
|Authors: ||Borelli, Massimo|
|Supervisor/Tutor: ||Battaglini, Piero Paolo|
|Co-Advisor: ||Lucangelo, Umberto|
|Issue Date: ||17-Apr-2012|
|Publisher: ||Università degli studi di Trieste|
|Abstract: ||In the present Ph.D. thesis main attention is focused in searching effective methods to improve the coupling of mechanical ventilators to critical care patients breath requirements, exploiting in a statistical framework the neural respiratory drive information.
The first Chapter is devoted to offer an outlook, within the neuroscience perspective, on some relevant aspects of mechanical ventilation. Chapter starts recalling the neuroanatomy of human respiration, both in normal lung function and in respiratory disease condition. Consequently, the conventional mechanical ventilation methodology is briefly recalled, presenting also the risks associated to it, posing a particular accent on dyssynchronies which affect the correct interaction between the patient and the ventilator. The current technology of neural control of mechanical ventilation (NAVA) is therefore outlined, together with a review of the technical steps which have characterized its realization. Chapter ends stating the main aim of our Ph.D. research project about the possibility to exploit the random effects modelling in detecting ventilatory dyssynchronies.
The second Chapter provides a complete description of the mixed-effects model capabilities in analysing neuroscience experiments, both in neurobiology and in cognitive/psychological sectors. The repeated measures and the longitudinal design experimental schemes are considered; current approaches in literature are discussed as well. The theory of the linear mixed model is illustrated by means of datasets of increasing complexity; the analysis is performed by means of the open source statistical package R. Datasets are mainly drawn from some of our co-authored papers.
The third Chapter deals with the frailty models, inherent to random effects within time-to-event experimentations: after a brief recall on survival analysis, both theory and a worked example of frailty model are presented.
In the fourth Chapter the limitations in applying mixed model techniques to the digital signal analysis are discussed, focusing also on some limitations still present in the available softwares. The result of our research, i.e. the Analyzer library written by means of Rcode, is presented in details and it is outlined how to import a NAVA Servo Tracker dataset into R, how to plot and how to summarize dataset information. Our Analyzer library represents the core of a machine learning software acting in the state-of-art Neuroscience-informed learning research field. A mixed model technique in analysing the NAVA signals is discussed and compared with an unpublished algorithm able to detect a widespread dyssynchrony known as 'ineffective expiratory effort' with an optimal reliability in terms of sensitivity and specificity. The algorithm is supported also by a mathematical proof, completely discussed in an Appendix of the thesis. The volume ends drawing some conclusions and prompting the path of further researches.|
|PhD cycle: ||XXIII Ciclo|
|PhD programme: ||SCUOLA DI DOTTORATO DI RICERCA IN NEUROSCIENZE E SCIENZE COGNITIVE|
neuroscience informed algorithm
|Main language of document: ||en|
|Type: ||Tesi di dottorato|
|Scientific-educational field: ||MED/01 STATISTICA MEDICA|
|Appears in Collections:||Scienze mediche|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.