In recent years, a growing number of methods and systems for automated diagnosis by means of statistical principles assisted by computer have been reported.
The application of stochastic decision theory to the medical diagnostics might have been foreseen much earlier since such possibility could be easily read in Bayesian conditional probability concept developed in as early as 1763.
The basic logic generally adopted to describing the diagnostics is of a deductive type, that is, the title of disease is given first and then the description on symptoms and signs generated by the disease follows.
In medical practice, however, the logic is reversed, that is, given first are the symptoms or signs this time and search for the title of the disease come next.
The latter logic, usually called as an inductive type, is the very concept which allows theapplication of mathematical disciplines such as probability theory and symbolic logic which are,, well accepted as the reasoning foundations of decision . making process in medical diagnosis.
From this point of view, we made an experiment on differential diagnosis´ in which the so~ called "Likelihood Method" is employed for discriminating the patients.
The diseases selected for this study are; (1) Peptic Ulcer Perforation (2) Appendix Perforation (3) Acute Pancreatitis and (4) Gall stone.
The symptom varables used for the differential diagnosis are; (1) Age (2) Sex (3) Gastricdistress (4) Nausea and vomiting (5) Location of sudden pain (6) Body temperature (7) Pulse: rate (8) Location of maximum tenderness (9) Muscle guardness (10) Bowel sound (11) W.B.C. (12) Hematocrit (13) Urine specific gravity (14) Urine sugar (15) Urine protein (16) Urine R. B. C (17) Urine W.B.C. (18) Free air in abdominal X-ray.
In order to apply the likelihood method to discriminate the patients, a "Discriminant ScoreChart" was constructed on the basis of distribution of symptom variables obtained from the-´ medical records of a total of 385 patients (peptic ulcer perforation: 99 cases, appendix perforation: 94 cases, acute pancreatitis: 93 cases, gallstone: 99 cases). The chart is presented inTable 19 in the main text. The scores recorded in the chart are the transformed values according to the following formula.
score= (logP2 j+ 1) x 10
where P2j: the proportion of patients with jth symptom for the ith disease.
When a series of clinical symptoms or signs are given the total score for each disease group is calculated by adding up the scores for 18 symptom variables and the disease group which totaled maximum score was selected as a diagnosed disease.
|