Expert System
Interpretation of
Tc-99m Sestamibi
Myocardial Perfusion
Tomograms:
Enhancements and
Validation
Ernest V. Garcia
Rakesh Mullick
|
|
|
| |
Abstract
To overcome the needed expertise and variability
in identifying the presence (P) and location (L) of
CAD from rest/stress myocarclial perfusion
distributions we have developed and enhanced over the
last 10 years an expert system (ES) based on
artificial intelligence tools to provide consistently
accurate interpretations. Using data from 461 patient
studies we implemented and refined 253 heuristic rules
into an ES shell which best correlated the P and L of
perfusion defects (PDs) on SPECT studies with CAD
demonstrated angiographically and with expert readings.
The PDs were identified from polar CEqual maps as
pixels below gender-matched normal limits. This ES
acts as an inference engine in which the certainty of
the L, size, shape and reversibility of the PDs, were
used to "fire" the rules to infer the certainty of
the P and L of CAD in a prospective group of 40 CAD
patients and 20 normals comprised of 30 males and 30
females.
Results:
| |
CAD |
LAD |
LCX |
RCA |
|
Sens |
95(38/40) |
2(23/25) |
100(11/11) |
96(22/23) |
|
Spec |
50(10120) |
46(16/35) |
71(35/49) |
76(28/37) |
|
when compared to the interpretation by human experts.
This expert system which justifies its results in
English and provides consistently accurate
interpretations should aid physicians in gaining
expertise in interpreting myocardial perfusion
studies.
The American Heart Association, Inc.
|
|
|
|