Prof. J.P.G. Lisboa School of Computing and Mathematical SciencesLiverpool John Moores University Byrom Street Liverpool L3 3AF United Kingdom E mail: P.J.Lisboa@ljmu.ac.uk

madbrainedmudlickΤεχνίτη Νοημοσύνη και Ρομποτική

20 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

91 εμφανίσεις

Prof. J.P.G. Lisboa

School of Computing and Mathematical Sciences

Liverpool John Moores University

Byrom Street

Liverpool

L3 3AF

United Kingdom

E

mail
:
P.J.Lisboa@ljmu.ac.uk



Advances in systems biology, together with growing expectations about referring the
evidence base to the needs of the individual patient, have given rise to the so
-
called 4P
agenda of preventive, predictive, personalised and participatory healthcare. This

agenda gained momentum as a way to contain spiralling healthcare costs with more
timely and effective interventions, requiring better modelling of the phenomenology of
disease, in particular for cancer.


The following two talks are linked. They will revie
w the application of generic non
-
linear
models in two aspects of cancer care, namely:


* supervised modelling of failure time series, i.e. survival modelling of longitudinal
cohort data.


* robust methods for unsupervised clustering of disease sub
-
types an
d for cohort
-
based
visualisation of high
-
dimensional data.


Talk 1. Generic non
-
linear models of prognostic outcome


Longitudinal cohort studies provide evidence for patient stratification by outcome. This
has both interpretative value by characterizing o
utcome differentials and predictive
value by estimating expected survival. In this talk, linear statistical methods are
extended into a generic non
-
linear time dependent framework for outcome modelling. A
prototype clinical interface is shown and developme
nts of the model to integrate
proteomic data are discussed.


A detailed case study is presented of survival of patients with early breast cancer. This
shows the predictive power of new data (ER status and histological grade) and also
provides smooth estim
ates over time for the hazard ratios and covariate effects.
Stratification methods are shown to generalize well from the calibration data to data
from another clinical centre, on which the AdjuvantOnline model was validated.


Talk 2. Robust methodologies f
or partition clustering


Clustering issues are fundamental to exploratory data analysis. This process may
follow algorithms that give unique answers by making assumptions about the ability to
find global optima. Most methods, such as k
-
means, are sen
sitive to the initial
conditions. This talk presents a methodology to make the variability in cluster solutions
dependent on the initial conditions, to find a hierarchy of useful cluster partitions which
should be profiled for practical interpretation b
y domain experts.


The methodology is generic and it is presented in the talk by reference to a
bioinformatics, specifically the sub
-
typing of cancer using a protein expression data set
derived from resected biopsies (n=1076).


References.


Lisboa
. P.J.G., Vellido, A. and Wong, H. 'Bias Reduction in Skewed Binary Classification
with Bayesian Neural Networks' Neural Networks, 13, 407
-
410, 2000.


Etchells, T.A. and Lisboa, P.J.G. 'Orthogonal search
-
based rule extraction (OSRE) from
trained neural net
works: a practical and efficient approach' IEEE Transactions on Neural
Networks, 17 (2):374
-
384, 2006.


Lisboa, P.J.G. and Taktak, A.F.G. 'The use of artificial neural networks in decision
support in cancer: a systematic review', Neural Networks, 19: 408
-
4
15, 2006.


Aung, M.S.H, Lisboa, P.J.G., Etchells, T.A., Testa, A.C., Van Calster, B., Van Huffel, S.,
Valentin, L. and Timmerman, D. 'Comparing Analytical Decision Support Models
Through Boolean Rule Extraction: A Case Study of Ovarian Tumour Malignancy'
Lecture
Notes in Computer Science 4492:1177
-
1186 , 2007.


Taktak, A., Antolini, L., Aung, M.H., Boracchi, P., Campbell, I., Damato, B., Ifeachor,
E.C, Lama, N., Lisboa, P.J.G, Setzkorn, C., Stalbovskaya, V. and Biganzoli, E.M.
'Double
-
blind evaluation and
benchmarking of survival models in a multi
-
centre study'
Computers in Biology and Medicine, 37 (8): 1108
-
1120, 2007.


Lisboa, P.J.G., Etchells, T.A., Jarman, I.H., Aung, M.S.H., Chabaoud, S., Bachelot, T.,
Perol, D., Gargi, T., Bourdès, V, Bonnevay, S and
Négrier, S. 'Time
-
to
-
event analysis
with artificial neural networks: an integrated analytical and rule
-
based study for breast
cancer' Neural Networks, 21(2
-
3): 414
-
426, 2008.


Jarman, I.H., Etchells, T.A., Martín, J.D. and Lisboa, P.J.G. 'An integrated fra
mework for
risk profiling of breast cancer patients following surgery' Artificial Intelligence in
Medicine, 42: 165
-
188, 2008.


Lisboa, P.J.G., Ellis, I.O., Green, A.R., Ambrogi, F. and M.B. Dias 'Cluster
-
based
visualisation with scatter matrices', Pattern

Recognition Letters, 29(13): 1814
-
1823,
2008.


Lisboa, P.J.G., Etchells, T.A., Jarman, I.H., Arsene, C.T.C., Aung, M.S.H., Eleuteri, A.,
Taktak, A. F. G., Ambrogi, F., Boracchi, P. and Biganzoli, E.M. 'Partial Logistic Artificial
Neural Network for Compet
ing Risks Regularised with Automatic Relevance
Determination ' I.E.E.E. Transactions on Neural Networks, 20(9):1403
-
1416, 2009.


Fernandes, A.S., Fonseca, J.M., Jarman, I.H., Etchells, T.A., Lisboa, P.J.G., Biganzoli,
E.M. and Bajdik, C. 'Evaluation of mis
sing data imputation in longitudinal cohort studies
in breast cancer survival' International Journal of Knowledge Engineering and Soft Data
Paradigms, 1(3):257
-
275, 2009.


Lisboa, P.J.G., Vellido, A., Tagliaferri, R., Napolitano, F., Ceccarelli, M., Martín
-
Guerrero,
J.D. and Biganzoli, E. 'Data Mining in Cancer Research', Invited Paper, IEEE
Computational Intelligence Magazine, February 2010:14
-
18.


Soria, D., Garibaldi, J.M., Ambrogi, F., Green, A.R., Powe, D., Rakha, E., Douglas
Macmillan, R., Blamey, R.W
., Ball, G., Lisboa, P.J.G., Etchells, T.A., Boracchi, P.,
Biganzoli, E. And Ellis, I.O. 'A methodology to identify consensus classes from clustering
algorithms applied to immunohistochemical data from breast cancer patients' accepted
for Computers in Biol
ogy and Medicine, 40:318
-
330, 2010.