Use this URL to cite or link to this record in EThOS:
Title: Time-domain system identification applied to non-invasive determination of cardiopulmonary quantities
Author: Bache, Richard Angus
ISNI:       0000 0001 3434 8353
Awarding Body: University of Glasgow
Current Institution: University of Glasgow
Date of Award: 1981
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis describes an application of the techniques of modelling and time-domain system identification to the processes of respiratory and inert gas transport in the human body. In particular, attention has been focussed on a new noninvasive method for measurement of the total blood flow through the lungs (the cardiac output in normal subjects). Determination of this quantity provides important clinical information on the state of the cardiovascular system. This work, being essentially multi-disciplinary, has involved close collaboration with medical personnel - in this case at the Centre for Respiratory Investigation, Glasgow Royal Infirmary. At this establishment much of the development of the homogeneous gas exchange model and practical experimentation have been carried out. The author has principally been concerned with the identification and accuracy aspects of the method. The starting point of the work involved the examination, in a general way, of data obtained prior to the author's full involvement in the project, which had produced results inferior in terms of reproducibility to that anticipated on the basis of use of a mathematical technique (see Chapter 3). At this stage, the author became convinced that the route to the solution of the troubles in the technique lay in viewing the problem as one in Statistical Time-Domain Identification. This represented a radical change in approach to the work since previously the model/data comparison and experimental design had been conceived in an 'ad hoc' manner and the identifi ability and accuracy implications poorly understood. Consequently, (in Chapter 4), the original data was viewed in this new light and certain deficiencies of the estimation method were made apparent by posing the problem in this probabilistic context. (ii) The analysis indicated that the cause of the disappointing results was the poor information content of the data rather than the nature of the model itself. This suggested that a better form of experiment be sought. This necessitated study of the area of optimal test signal design to maximise the amount of information encumbent in the resultant data. Utilising these concepts a new longer form of experiment, aimed at having better properties in respect of cardiac output estimates, was evolved. This work is reported in Chapter 7 where the results of a comprehensive set of reproducibility studies to test the new form of experiment are also presented. These results showed a marked improvement in the reproducibility of the technique, as was exemplified by the fact that the average reproducibility of the cardiac output estimates was 6.2% (4.6% if two rogue results are ignored), as opposed to 12.2% for the earlier studies. What is even more encouraging is that this figure even stands comparison with the average reproducibility of the earlier dye dilution estimates calculated at 6.8% and much of the published results for both invasive and non-invasive methods in the literature. In Chapter 8 the scope of the work is extended somewhat and here inhomogeneous gas transport models (applicable to diseased lungs) are considered. The concept of designing identification methods to optimally discriminate between these models and homogeneous models is tentatively introduced as a mechanism for quantitative diagnosis of lung dysfunction and some prefatory simulation work in this vein presented. A technique crucial to the mechanisation of the identification methods underlying much of the work detailed in this thesis is that of Function Minimisation. A large proportion of the time allocated for this Ph.D. project was spent on the investigation of recent, numerically stable computer algorithms for this purpose. (iii) Chapter 5 investigates the usefulness of generalised descent methods in the context of the particular application for this project, whilst Chapter 6 is devoted specifically to methods for sums of squares problems. A technique due to Gill and Murray (124) is shown to be superior to all others for estimating the parameters of the lung model and thus this algorithm is used to generate the results subsequently presented in the rest of the thesis. When creating the software for the Function Minimisation, great care was taken to configure it in a general manner. This philosophy thus led to the creation of a flexible Function Minimisation package as a useful by-product of this work at very little extra programming effort. This the author feels constitutes a piece of software which could be of general use in a wide number of different applications. Details of this package are outlined in Appendix B.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Bioengineering & biomedical engineering