An Expert System on Detection, Evaluation and Treatment of Hypertension - Artificial Intelligence and Symbolic Computation - page 259

to the data (what is usually the case when dealing with medical information).
This is not the case here, as the input data have a True/False format and data
acquisition is precise (all input data can be clearly determined from the his-
tory, physical examination, laboratory tests, electrocardiograms...). Moreover,
proceeding with only part of the input data could be very dangerous (e.g. the
suggested medication could be incompatible).
As said in the introduction, this work is mainly based on the conclusions of
the committees of the ISH-WHO [7], JNC 6 [9] and JNC 7 [10, 11]. These commit-
tees release conclusions based on large-scale clinical trials and sound statistical
studies, and are considered as the main reference in the field. This widely ac-
cepted knowledge is what our program automatices. Our inference engine doesn't
add any new rule, it just incorporates the conclusions of these committees, so
we don't have to deal with confidence thresholds of new added knowledge.

6.1 Translation of the Corrected BP Classification Tables

to a Set of Production Rules

Now that the semialgebraic regions defined by Table 3 are disjoint, the informa-
tion contained in it 3 can be translated as the following set of rules:

IF SBP < 120 AND DBP < 80 THEN Normal
IF 120 SBP < 140 AND DBP < 90 THEN Prehypertension

IF SBP < 140 AND 80

≤

≤ DBP < 90 THEN Prehypertension

IF 140 SBP < 160 AND DBP < 100 THEN Stage 1 Hypertension

IF SBP < 160 AND 90

≤

≤ DBP < 100 THEN Stage 1 Hypertension

IF SBP ≥ 160 THEN Stage 2 Hypertension

IF DBP ≥ 100 THEN Stage 2 Hypertension

The information in Table 4 can be translated similarly.
In previous occasions we have implemented expert systems devoted to other
illnesses where the rules really had to be fired once and again and again, as
new derived facts were obtained each time. Then we used a Grobner bases-based
inference engine [5, 6, 12, 13] based on a application to RBES of these authors
[14], itself based on previous works for Boolean logic [15, 16] and modal multi-
valued logics [17]. Such an approach would be infraused here, as rules have to
be fired only one time. Therefore we have decided to implement simple proce-
dures in Maple using IF...THEN...ELSE... nested conditionals, like the following
procedure, that determines the particular BP case.

> class_BP:=proc()
> global BP;

> if SBP>=160 or DBP>=100

> then BP:=Hypertension2

> elif SBP>=140 or DBP>=90