The  Weizmann  Institute  of  Science
                  Faculty of Mathematics and Computer Science


                            Computer Science Seminar

                                Michael M. Segal
                        Dept. of Surgery (Neurosurgery)
                             Harvard Medical School
                                      and
                            CTO, SimulConsult, Inc.


                                 will speak on


                    Harnessing probabilistic information for
                            medical decision support

Abstract:
A novel method of parsing verbal descriptions of probabilities will be shown.
The context is a project involving doctors on several continents who are
collecting probabilistic information using collaborative medical decision
support software.  Much of the knowledgebase for medical diagnosis consists of
descriptions of findings in diseases using verbal descriptions such as ``not
infrequently present".  We have developed and implemented a mathematical
analysis of such verbalisms.  The parsing turns out to have a remarkable
mathematical simplicity.  It allows automated parsing of thousands of different
verbal expressions, taking into account multiple negations and adjectives.  The
model reproduces knowledge that is so widespread that it is used correctly by
young children, yet is so unconscious that intelligent adults are unaware of
the rules.


This parsing of verbal descriptions of probability will be demonstrated in the
context of a tool that assists doctors in reaching medical diagnoses.  The
doctors use a Java applet to collect and share probabilistic information about
findings in diseases, including the temporal onset of the findings. The
information is used not only to suggest diagnoses but also suggest what other
information would be most useful and cost effective to obtain.  This
collaborative tool uses an ``open database" approach in which users can examine
all the probabilities in the database and submit changes or new information,
subject to peer review, creating a collaborative database of best available
information.




                      The lecture will take place in the 
                     Lecture Hall, Room 1, Ziskind Building
                           on Monday, August 20, 2001
                                    at 14:30