Presentation

MYCIN Expert System

Presented by

Junaid Khan
Department of Computer Science
University of Peshawar Pakistan
Junaid_upesh@yahoo.com
 AI Continued…….
☻Also, the problem-solving methods themselves are
usually qualitative reasoning techniques that relate
items through judgmental rules, or heuristics, as well
as through theoretical laws and definitions.
☻An algorithm is a procedure that is guaranteed either
to find the correct solution to a problem in a finite
time or to tell you there is no solution.
☻ For example, an algorithm for opening a safe.
☻Few problems in medicine have algorithmic
solutions that are both practical and valid.
☻Physicians are forced to reason about an illness
using judgmental rules and empirical associations
along with definitive truths of physiology.
 What is MYCIN?
☻MYCIN is an expert system (BruceBuchanan and
Shortliffe, 1983).
☻By that we mean that it is an AI program designed
(a) to provide expert-level solutions to complex
problems,
(b) to be understandable, and
(c) to be flexible enough to accommodate
new knowledge easily.
 Because they have designed
☻MYCIN to provide advice through a consultative
dialogue, we sometimes refer to it as a consultation
system.
History of MYCIN
☻ MYCIN was an outgrowth of DENDRAL .

☻ The MYCIN team members

 Bruce Buchanan( Professor of Computer Science, Philosophy,

and Medicine, with the Department of Computer Science at the
University of Pittsburgh ).

 Stanley Cohen, then Chief of Clinical Pharmacology at the

Stanford University Medical School.

 Edward Shortliffe(a physician and computer scientist at

Stanford Medical School).

 George Forsythe, then Chairman of the Computer Science

Department.
Continued…..
☻After six months of collaborative effort on
MEDIPHOR, their discussions began to
focus on a computer program that would
monitor physician’s prescriptions for
antibiotics .
REQUIREMENTS
☻access data bases on three Stanford
computers:
 clinical laboratory
 bacteriology systems and
 the pharmacy
☻considerable knowledge about the general
and specific conditions that make one
antibiotic, or combination of antibiotics.
 Requirements continued….
☻Before a system could monitor for inappropriate
therapeutic decisions, it would need to be an
"expert" in the field of antimicrobial selection.

☻ Thus, a monitoring system could be modified to

provide consultations to physicians.

☻Another appeal of focusing on an interactive system

was that it provided us with a short-term means to
avoid the difficulty of linking three computers
together to provide data to a monitoring system.

☻ Thus their concept of a computer-based consultant

was born.
Whose EXPERTIES to be used?
☻ Cohen interested Thomas Merigan, Chief of the
Infectious Disease Division at Stanford, in
lending both his expertise
And that of
☻ Stanton Axline, a physician in his division.
☻ Shortliffe synthesized medical knowledge from
Cohen and Axline
And
☻ AI ideas from Buchanan and Cordell Green.
☻ Axline subsequently enlisted help from
infectious disease fellows to complement.
☻ The expertise of" Cohen’s clinical
pharmacology fellows”.
 MYCIN’s Task Domain Antimicrobial
Selection
☻ The sulfonamides in the 1930s and penicillin in 1943.

☻ In the 1950s it became clear that antibiotics were

being misused.

☻ At the time they began work on MYCIN, antibiotic

misuse was receiving wide attention .

☻ The studies showed that very few physicians go

through the methodical decision process.

☻ In the outpatient environment antibiotics are often

prescribed without the physician’s having identified or
even cultured the offending organism.
Structure of MYCIN
Structure of MYCIN Continued….
 Some main points about MYCIN
☻ It is a rule based system.

☻ MYCIN primarily uses backward chaining, or a goal-

directed control strategy.

☻ It is built using LISP language.

☻ CF’s are associated with all implications.

☻ Meta-level reasoning is allowed in order to increase

efficiency (e.g., using meta-rules or looking for a unity
path).

☻ MYCIN’s logic is incomplete in the sense that we know

there are propositions that can be expressed in the
language but are not provable as theorems.
 A typical consultation
1) Patient’s name: (first-last)
** FRED BRAUN (Questions 1-3 request background patient
data.)
2) Sex:
**M
3) Age:
** 55
4) Are there any cultures for Fred Braun which may be related to
the present illness, and from which organisms have been
grown successfully in the microbiology laboratory?
**y
........ CULTURE-1 ........
5) From what site was the specimen for CULTURE-1 taken?
** BLOD
= BLOOD (System corrects typing error.)
 Consultation continued…..
6) Please give the date and time when CULTURE-1 was
obtained. (mo/da/yr time)
** JUN 20, 1977

 The first organism isolated from the blood

culture of 20-JUN-77 (CULTURE-I) will be
referred to as:
........ ORGANISM-1 ........

7) Enter the laboratory-reported identity of ORGANISM-l:

**UNKNOWN (Typically identity is not yet
known.)
•
 Consultation continued……

8) The stain (gram or ZiehI-Neelsen acid-fast) of

ORGANISM-l:
** NEG (Preliminary lab results give some
clues.)

9) Is ORGANISM-1 rod or coccus (etc.):

** ROD

10) What is the form of the individual organisms (e.g.

lancet shaped for cocci, fusiform for rods, etc.)?
** FUSlFORM
 Consultation continued…..
(... more questions follow in order to gather sufficient
information to infer the identity and significance of the
infecting organisms... )

23) What is the suspected portal of entry of the gramneg

rod from the blood culture of 20-JUN-77 (ORGANISM-I)
into this sterile site?
** GI (6) (User indicates the uncertainty of his
response.)

24) Is Fred Braun a burn patient?

**N
 Consultation continued……
25) Is Fred Braun a compromised host(e.g.alcoholic,
sickle-cell disease,immunosuppressed)?
**N

26) Is Fred Braun febrile?

**y

(... after about 50-60 questions, MYClN prints the

diagnostic hypotheses on which therapy will be
based... )
 How it works?
Patient1

Positive culture

Culture1 Drug 6
Known organism Suspected infection Previous treatment Current treatment

organism1

Stain
Growth

Gram positive
A context tree for a patient
 Start
The structure of
MYCIN ‘s Examine the first
MONITOR part in the condition
part of the rule

No
Can the test
be evaluated? Examine the next test

Yes
Yes
Yes
Get the information
Is the test true? More tests?
needed with
FINDOUT
No
No
Add the conclusion of
Reject the rule the rule to the patient
context tree

Return Return
 The structure of MYCIN’s
FINDOUT
Start

No
Find the rules Is the item Yes
for chaining LABDATA?
with UPDATED BY
Ask the user
to carry out
the necessary
Apply MONITOR test and enter
to the rules in turn the result

Return Return
Test which was
conducted to evaluate
the performance of
MYCIN
Rule of checking….
Each judge was asked to score
each
recommendation as
a) equivalent to their own best
judgment,
b) not equivalent but acceptable,
or
c) unacceptable.
Result of the test
 Result Continued……
☻ What is the x axis of the graph?
☻ It is unlabeled because the factors that
determine performance have not been explicitly
identified.
☻ What could these factors be?
☻ Mycin certainly does mental arithmetic more
accurately and more quickly than Stanford
faculty; perhaps this is why it performed so well.
☻ Mycin remembers everything it is told; perhaps
this explains its performance.
☻ Mycin reasons correctly with conditional
probabilities, and many doctors do not ;
perhaps this is why it did so well.
Any
Question…?