AIME'01 Workshop: Bayesian Models in Medicine

Scope

Bayesian models are used in medicine to assist in the diagnosis of disorders and to predict the natural course of disease or outcome after treatment (prognosis). They are also being used as part of models to determine the optimal treatments of a disorder in patients, or to predict outcome in groups of patients. The Bayesian approach has the advantage that evidence can be easily incorporated into statistical models. Another advantage is that evidence can be handled readily when using a Bayesian model to solve an actual medical problem. Furthermore, Bayesian models cannot only be developed by extracting probabilistic information from datasets; graphical models like Bayesian networks can also be constructed with the help of medical domain experts or by consulting relevant biomedical literature. Typically, Bayesian networks rely for their construction on causal, in particular (patho)physiological models of disease. Bayesian networks have also been used successfully for the construction of dynamic, temporal statistical models. The fact that Bayesian models allow for the easy incorporation of knowledge of background populations, explains that they are also increasingly used in research on risk models of disease, associating risk with spatial distribution of populations. When used for the prediction of life expectancy, they can be combined with traditional models of life expectancy. In the context of medical decision making, Bayesian models can be easily integrated with decision theory to yield models for the selection of optimal treatments, or to develop models for healthcare planning under uncertainty.

Topics

It is expected that papers will explicitly discuss one or more of the topics mentioned below in the context of medicine.

Modelling and Reasoning

• Construction of Bayesian network models in collaboration with medical domain experts, using qualitative knowledge or decision-theoretic principles
• Bayesian modelling of biological mechanisms
• Specification languages for Bayesian models, such as probabilistic logic and graph-like languages
• Problem solving in medicine with Bayesian networks
• Computational issues, such as space and time complexity of algorithms, in reasoning with realistic medical Bayesian models
• Representation and manipulation of time in Bayesian networks
• Representation of spatial knowledge in Bayesian models
• Markov processes (MPs, MDPs and POMDPs)
• The notion of causality in medicine, and its representation in Bayesian networks
• Sensitivity analysis of Bayesian models in medicine
• Modelling of medical domain characteristics by means of Bayesian networks

Learning

• Problems of using real-world biomedical datasets for learning Bayesian models from data
• Using background knowledge, e.g. causal knowledge, to enhance learning Bayesian models from data
• Learning Bayesian network topologies from data
• Probability assessment in the context of the construction of Bayesian models from data
• Refinement of Bayesian models using data
• Handling missing values in data

Evaluation

• Methodologies for the evaluation of Bayesian models
• Evaluation of Bayesian networks for concrete medical domains

Submissions have been refereed by at least two and in most cases three members of the programme committee. Accepted papers will appear in the working notes of the workshop "Bayesian Models in Medicine". The authors of the best papers of the workshop will be invited to submit an extended version of their paper to a special issue of the journal Artificial Intelligence in Medicine on Bayesian Models in Medicine.

Final Programme

---

08:45 - 09:00 Coffee

---

Session I: 09:00 - 10:50 (Chair: Peter Lucas)

• 09:00 - 09:15 Welcome
• 09:15 - 10:00 Invited Talk - Marco Ramoni and Paola Sebastiani: Bayesian bioinformatics
• 10:00 - 10:30 Rosa Blanco, Pedro Larranaga, Inaki Inza and Basilio Sierra: Selection of highly accurate genes for cancer classification by estimation of distribution algorithms
• 10:30 - 10:50 Vincent Labatut and Josette Pastor: Bayesian modeling of cerebral information processing

---

10:50 - 11:15 Coffee Break

---

Session II: 11:15 - 12:35 (Chair: Linda van der Gaag)

• 11:15 - 11:45 Wim Wiegerinck and Tom Heskes: Probability assessment with maximum entropy in Bayesian networks
• 11:45 - 12:15 Silvia Acid, Luis de Campos, Susana Rodriguez, Jose Maria Rodriguez and Jose Luis Salcedo: Represening health-care systems using Bayesian networks. An experimental comparison of learning algorithms
• 12:15 - 12:35 Carmen Lacave, Agnieszka Onisko and Francisco Javier Diez: Debugging medical Bayesian networks with Elvira's explanation facility

---

12:35 - 14:15 Lunch

---

Session III: 14:15 - 15:50 (Chair: Ameen Abu-Hanna)

• 14:15 - 15:00 Invited Talk - Andrew Lawson: Bayesian hierarchical modelling for spatial disease surveillance
• 15:00 - 15:30 Peter Antal, Geert Fannes, Frank De Smet, Joos Vandewalle and Bart De Moor: Ovarian cancer classification with rejection by Bayesian belief networks
• 15:30 - 15:50 Joachim Horn, Thomas Birkhoelzer, Oliver Hogl, Marco Pellegrino, Ruxandra Lupas Scheiterer, Kai-Uwe Schmidt, Volker Tresp: Medical knowledge acquisition and automated generation of Bayesian networks

---

15:50 - 16:15 Refreshments Break

---

Session IV: 16:15 - 17:45 (Chair: Peter Lucas)

• 16:15 - 16:45 Niels Peek: A notion of diagnosis in decision-theoretic planning
• 16:45 - 17:15 Agnieszka Onisko: Evaluation of the HEPAR II system for diagnosis of liver disorders
• 17:15 - 17:35 Ken McNaught, Sarah Clifford, Marilyn Vaughn, Anthony Fogg and Mike Foy: A Bayesian belief network for lower back pain diagnosis
• 17:35 - 17:45 Closing

---

Instructions to authors

The papers (up to 5 pages) are to be sent as a compressed Postscript file by e-mail before 10 June 2001 to all three co-chairs and should be written in English with a brief abstract. Formatting instructions are as follows: the abstract should be formatted in two-column format, with Times Roman type face, pointsize 10, with title and names of the authors in bold font. Left and right margins should be 2 cm, text height 23 cm, and text width 16.9 cm; the two columns should be separated by 0.5 cm white space. A sample paper is available.

Preferably the papers are produced using either

• LaTeX (preferred): Use
  \documenclass[fleqn,twocolumns]{article}
  \usepackage{aime01}
  (A sample LaTeX file is available.) or
• MS-Word

The style file for each of these two text processing systems is obtained by clicking on the hyperlinks.

Registration

Workshop price is only 50 Euros; workshop participants must also register for the main AIME'01 conference.

Workshop organization

Co-Chairs

Peter Lucas, University of Aberdeen, Aberdeen, UK
Linda van der Gaag, Utrecht University, The Netherlands
Ameen Abu-Hanna, University of Amsterdam, The Netherlands

Programme committee

A. Abu-Hanna, The Netherlands
K.-P. Adlassnig, Austria
R. Bellazzi, Italy
C. Berzuini, Italy
G.F. Cooper, USA
R.G. Cowell, UK
F.J. Diez, Spain
M.J. Druzdzel, USA
L.C. van der Gaag, The Netherlands
P. Haddawy, USA
D. Hand, UK
I.S. Kohane, USA
P. Larranaga, Spain
A. Lawson, UK
L. Leibovici, Israel
T.Y. Leong, Singapore
P.J.F. Lucas, UK
S. Monti, USA
L. Ohno-Machado, USA
K.G. Olesen, Denmark
M. Paul, Israel
M. Ramoni, USA
A. Riva, USA
P. Sebastiani, USA
G. Tusch, Germany
J. Wyatt, UK
B. Zupan, Slovenia

For more information about the workshop please contact one of the co-chairs.

---

Peter Lucas  Dept. of Comp. Science  Meston Building  University of Aberdeen  Aberdeen, AB24 3EU  Scotland, UK  Telephone: +44 1224 273829  Fax: +44 1224 273422/487048  plucas@csd.abdn.ac.uk

Linda van der Gaag  Dept. of Comp. Science  Utrecht University  Padualaan 14  3584 TB Utrecht  The Netherlands  Telephone: +31 30 2534113  Fax: +31 30 2513791  linda@cs.uu.nl

Ameen Abu-Hanna Dept. of Medical Inform., AMC  University of Amsterdam  Meibergdreef 15  1105 AZ Amsterdam  The Netherlands  Telephone: +31 20 5664511  Fax : +31 20 6912432  A.Abu-Hanna@amc.uva.nl

Peter Lucas | AIME'01 | Workshop Location
University of Aberdeen

Last updated: 28 May, 2001

peterl@cs.kun.nl