LEXMED

Computer Aided Diagnosis of Appendicitis
The aim of this project is to build a learning medical expert system for computer
aided diagnosis of acute abdominal pain, especially in the case of a suspected
appendicitis (LEXMED). Since June 1999 LEXMED is being tested at the '14
Nothelfer Hospital' in Weingarten / Baden-Wrttemberg / Germany, with very good
results. Additionally LEXMED is available (for physicians) on the internet (see
'registration' to obtain an account). It is easy to use and to access (free of charge).
Given the symptoms (maximal 14) of a patient with acute abdominal pain, LEXMED
first gives the probability of four different diagnoses [inflammed appendicitis,
perforated appendicitis, non specific abdominal pain (nsap), something else (other
illness) ]. The second step delivers a proposal for a treatment (1 of 5 possibilities:
operation, emergency operation, observe patient at home (out-patient), observe
patient in hospital (in-patient), perform further examinations (it may be something
else).
Method:
Its knowledge is based on a set of probability rules (500 at the moment) 400 of which
stem from a database of 15.000 patient records (operations carried out on suspected
appendicitis in 1995 in Baden-Wrttemberg / Germany) and 100 from our medical
experts. The methods used to extract (uncertain) rules from data and to compile these
rules together with those of the medical experts (to generate a complete probability
model) are based on probability theory and its extension using the method of
maximum entropy (for more details see the literature on our website). Due to its
construction, new rules (from data or medical experts) could be compiled and used by
LEXMED within minutes. From time to time we include new data (collected by our
partners) and rules, which is the reason for calling it a learning expert system. The
method of its construction is not just applicable for the diagnosis of abdominal pain
or even restricted to the medical area. It could be applied for many different tasks of
diagnosis in uncertain domains.
Future
We plan to extend LEXMED to diagnose other causes of acute abdominal pain. To
achieve this (but also for the current discussion) we are looking for interested
physicians
and
other
interested
persons.
Funding:
This project was funded by the "Ministerium fr Wissenschaft, Forschung und
Kunst" in Baden-Wrttemberg / Germany . The project started in November 1997
and was finished in December 1999. Ongoing research activities are being funded by
the University of Applied Sciences Ravensburg-Weingarten.
Making a Diagnosis
First case: You do not have a personal access authorisation. LEXMED can be used
without user identification. Use the Button 'LEXMED-Diagnosis' without making
entries in the fields 'username' and 'password' . This anonymous access is restricted. A
personal account can be applied for through the registration program (see
'Registration') Your access authorisation will be sent to you within one week by
email. Please note: There are two forms of accounts available:
-Account without a patient information system. You receive a question sheet, which
contains the symptoms which are presently used for diagnosis in LEXMED.
-Account including a 'Patient information system': You have the additional possibility
to store the symptoms of the (anonymous) patient (which is of advantage when
creating variations of symptoms). The accordingly used storage area is only
accessible to you (and the system operator). In addition you find on this page also
symptoms that are presently not been used but we forsee being applied to a question
sheet on a future version of the system. This type of account is especially intended for
doctors, who want to make their cases available to the project (for a future version,
after agreement of a definite procedure to draw up the data)
-Second case: You do have a personal access authorisation. Please enter 'username'
and 'password' (and be aware of of small and capital letters)
-Press
the
button
'Lexmed-Diagnosis'
For Problems (e.g. Button not visible, not useable, huge fontsize) check settings in
your browser (Netscape recommended). You need
-JavaScript: enabled
-StyleSheets: enabled
-Enter results of investigation in the question sheet. (Accounts including the 'patient
information system' have the possibility to save the patient findings)
-Note 1: The system operates under the assumption that the patient suffers from acute
abdominal pain (for less than 5 days and more than 2 hours).
-Note 2: The system always makes a diagnosis. This also takes place when only a few
(extreme case: no) investigation results are entered. It is only natural that the
proposed diagnosis is more meaningful when a larger number of investigation results
are recorded. The degree of uncertainty of the given diagnosis could be taken from
the size of average error, expressed by the cost matrix . (If high (e.g. > 2000), then
uncertainty also high)).
Press Button 'Diagnosis'

Wait for result page (Page with probabilities for different diagnostic
findings)
.
(assuming normal usage of the server only a few seconds).
Staring from this page, it is possible to request a proposal for a treatment. (Use
the link 'Proposal for Decision'). This proposal is based on the calculated
probabilities and a cost-matrix, which contain the (general) costs of wrong
classifications. (e.g: It is very wrong, to send a patient home, if he has a
perforated appendicitis; in this case there are high 'error costs'). Further
information abount the cost matrix is also available from this page.
finish or continue as required.
System Resources
Database (1995) with 15000 patients (Landesrztekammer Baden
Wrttemberg)
graphical model of the dependencies of the symptoms on these data
Frequency statements of (cooperating) surgeons
Costmatrix of wrong classifications (Matrix for penalty points for wrong
decisions) Source: Assessment of consequential damages)
System Versions:
System Version 1.02
Rulebase:
v2.0; approximately 400 probability statements (Jun 1999)
v2.1; approximately 100 probability statements from surgeons (Dr.
Rampf, Dr. Hontschik) (Jun 1999)
Costmatrix : v1.02 (Jun 1999)
Patient Information System: v3.00 (Jan 2000)
Accountmanagement: v2.00 (Jan 2000)
PIT-Scripts v4.0 (Nov 99) , imap v2.0 (Jun 99)

LEXMED

Uploaded by

Copyright:

Available Formats

LEXMED

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

LEXMED

Uploaded by

Copyright:

Available Formats

Computer Aided Diagnosis of Appendicitis

Press Button 'Diagnosis'

You might also like