INTRODUCTION
This special issue on “Computer Methods and Programs in Biomedicine” is a collection of
related papers selected and extended from those presented at the triennial IFAC Biological and
Medical Systems (BMS 2012) Symposium in Budapest, Hungary during August 29-31, 2012.
IFAC was founded in September 1957 and is a worldwide multinational federation of National
Member Organisations (NMOs). Each NMO represents the engineering and scientific societies
concerned with dynamic systems and control within its own country. The aims of the Federation
are to promote the science and technology, both theory and application, of systems, signal and
control in the broadest sense in all systems, whether, for example, biological, medical,
engineering, physical, social or economic.
Technical excellence is a main concern of IFAC. In a world where rapid and widespread
dissemination of information has become possible for everybody, maintaining the highest level
of scientific excellence is the only way in which an organisation like IFAC can prosper. Thus,
IFAC prides itself on promoting consistent and rigorous peer review in all its publications.
The IFAC BMS 2012 Symposium is sponsored by the IFAC Technical Committee on
Biological and Medical Systems (TC 8.2). There were 109 articles from 21 countries of which
approximately 50% were part of invited sessions across a range of topics. Each paper had a
minimum of 2 reviewers and the top 14 articles by reviewer grade were selected for this special
edition, where the revised articles underwent a further rigorous independent peer-review. From
this group of 14 and the journal’s review process 10 were accepted for this special edition.
Taken together, we believe these selected articles constitute a valuable source of information on
the broad state of the art in Biomedical Systems Modelling, Signal Processing and Control.
The articles break down into the 3 main areas of modern control systems, with relevant
application to the TC 8.2 focus on Biological and Medical Systems. Specifically, papers are
grouped into: 1) Physiological model development, the major element at the commencement of
analysing and developing any control system; 2) System identification and identifiability, which
is at the heart of subject and patient specific modelling and control; and 3) Control design and
methods that present novel approaches to physiological control. Overall, these three areas cover
the major aspects in the applications of dynamic systems modelling and control to medicine.
Physiological Modelling:
The first article by Evans and colleagues describes the modelling of tumour growth in
xenografts animal models used in drug development. Xenograft animal models are a first line
test in oncology studies for developing new drugs. However, the simple measurement of tumour
growth over time in these simple models often masks several potentially important
pathophysiological and biological changes that would help investigate the mode of drug action.
This article presents a first model of tumour growth including histological biomarkers, and is
designed to be structurally identifiable given the data available in such studies. Clinical data
from a first use are then used to validate the model and approach. However, the overall paper
presents a unique approach to model development for a targeted, and critical, clinical research
outcome.
The next article by Hall and colleagues presents a mathematical model for the pharmacokinetics
of the commonly used anti-malarial drug, artesunate. The research develops a series of
constraints to enable the model to be identified, as it is not structurally identifiable with typical
data. These constraints thus alter the model and these impacts are analysed and shown to still
capture the observed dynamics. Hence, this article presents a new approach to adding insight via
modelling assumptions, a key element of any model representation, to help address the
significant and growing problem presented by malaria and our difficulty in treating it.
Keir and colleagues present a two compartment model to capture the dynamics of myoglobin in
patients undergoing dialysis. The model is developed from data for a specific group of patients
based on ensuring structural identifiability of the resulting model given the data that would be
available. The model is validated predictively by its ability to capture observed data, and
presents a further improvement in using structured model development methods to apply
engineering principles to improving care, using models, in core therapeutic arenas.
Finally, Pretty and colleagues use a similar approach to creating and identifying an insulin
kinetics model from newly emerging micro-dialysis data. This approach uses very rarely
obtained data to develop and validate kinetics parameters for a well-known kinetics model that
are not otherwise identifiable with typical data. It also assesses their validity by their ability to
capture observed clinical data dynamics, and in relation to a range of prior insulin kinetics
models that have all tried to address this problem.
From the area of model development, and particularly model development with a focus on
ensuring identifiability of parameters for typically available data, there is a natural lead-in
towards methods of system identification, identifiability analyses and the effect of sensor noise
or timing on identified parameters.
System Identification and Identifiability:
Yu and colleagues present a novel identification method for a biomechanics problem. In
particular, they parameterise a two muscle arm model incorporating a Hill muscle model so that
it is identifiable. The method uses a mix of directly measure anthropomorphic data and inferred
values to identify an equivalent mechanics model. The method is validated comparing measured
and identified model results, and an error sensitivity analysis is presented that shows reduced
error and sensitivity resulting from the assumptions that have to be made around body segment
geometry in these models.
Grandjean and colleagues then examine the application of four different methods to determine
the structural identifiability and indistinguishability of parameters for a series of candidate
nonlinear pharmacokinetic models used to capture the hepatic uptake of a therapeutic agent. The
methods are analysed and validated using in vitro data. Structural identifiability analysis shows
that, in this case, all candidate models were structurally distinguishable given the typical
observations. The overall results present a clearly template for analysing and comparing models
and the resulting identified parameters.
Next, Docherty and colleagues assess the impact on parameter identification of the quality of
the simulation method involved. Thus, given a model and data, errors after identification are
often ascribed to the model or quality of data. However, this research shows that the simulation
methods used in parameter identification and optimisation can also have a significant impact in
certain cases. In particular, they point out that models with discrete step changes, such as are
often found in lung recruitment models, can yield significant errors if the user is not very careful
using otherwise well-accepted approaches to system identification.
Finally, Pretty and colleagues assess the impact of sensor errors and sensor measurement timing
errors on identified parameters for insulin sensitivity in a well-known, structurally identifiable
model. This article uses a range of observed and recorded sensor error models and a timing error
model based on clinical observations of the pharmacodynamic glucose-insulin model in clinical
use for decision support. Insulin sensitivity is the driving parameter in determining insulin
dosing decisions. Overall it is found that the impact of reasonable sensor and timing errors can
be notable, but are not likely to outweigh far larger variations in patient behaviour for the
critically ill cohort examined. Thus, while these effects are overshadowed by physiological
effects in the critically ill, their size may be more significant in use with ambulatory populations
and in assessing hour-to-hour insulin sensitivity variability. The overall work thus links the
impact of methods and errors in identification to potential impacts on model-based decision
support.
Given models and the ability to make them physiologically patient specific with system
identification, as well as understanding the impact of different identification approaches and
methods, the next critical step is their application in control and decision support.
Physiological Control Systems:
Nogueira and colleagues present a nonlinear positive control law, for reference tracking in
multi-input, single output positive systems, which is applied to an existing model and problem
in depth of anaesthesia. Anaesthesia is a well-known area where model-based and automatic
control has made great headway over the recent decade. The methods presented are shown to
provide good potential control and provides a novel approach to visualising and designing
control for this clinical scenario.
Finally, Kovacs and colleagues examine the model-based control of molecular targeted
therapies in treating cancer. These approaches have limited side effects. However, titration and
control of dosing are not well known nor well understood. In particular, anti-angiogenic therapy
is quite promising but not well understood. This research presents a modified form of an
existing angoigenic tumour growth and control model, and applies h-infinity methods to better
understand the potential to control and modulate these therapies effectively, as well as providing
greater understanding of the underlying pathophysiological processes.
In summary, with relevance to this entire issue, health care costs and challenges are growing
faster than health care systems and governments can manage, and the last 20 years has seen
health grow at an annualised rate of 8-10% to consume 10-18% of GDP in western nations.
These changes are unsustainable if we wish to maintain equity and access to care for all. The
scientific and technological expertise of IFAC research into applying dynamic systems
modelling, system identification and control to a range of health care problems is developing
new means to reduce, manage, optimise, or even eliminate, some of these issues. Thus, these
papers all embody the main goal of IFAC TC 8.2 on Biological and Medical Systems to
integrate engineering, physiology, biology and clinical medicine to create new scientific and
technological breakthroughs to improve our world.
Guest editors:
Professor J. Geoffrey Chase
Professor Ewart Carson