Pattern matching models of veterinary diagnosis

Abstract

In a survey of veterinarians and veterinary students pattern matching, pathophysiological reasoning and probabilities were recognised by both groups as pattern recognition strategies used in diagnosis. Veterinary students stated that they used pathophysiological reasoning most often and the veterinarians replied that they used pattern matching most frequently. Logical exclusion was used provided the data was reliable. The veterinarians indicated that they used the signs observed to be present and the signs observed to be absent during pattern recognition.

Pattern recognition analysis using case reports identified that pattern recognition was a function of a pattern matching model and not a function of a Bayes' theorem probability model with cr without prevalence data. The pattern matching model most closely resembled the results of each veterinarian regardless of their experience level.

A pattern matching system for the identification of Bovine Spongiform Encephalopathy (B.S.E) was devised. This system contained four pattern matching models.The system used prototype descriptions of the differential diagnoses based upon the point prevalence frequencies of the signs within diseases. The most accurate model for the recognition of the prototype disease descriptions used the signs observed to be present and absent with logical exclusion.

The sensitivities of the B.S.E. pattern matching system and 25 final year veterinary students were tested with 50 confirmed B.S.E case reports. The model with the highest sensitivity used the signs observed to be present and logical exclusion. Three cf the models were significantly better than the veterinary students at diagnosing B.S.E in patients with the disease. The model which allowed for the greatest amount of uncertainty regarding the input data had the lowest sensitivity.

A hypothetico-deductive pattern matching model was devised using sign point prevalence frequencies. This hypothetico-deductive pattern matching model of diagnosis was compared to 5 veterinarians. The performance of the model was equivalent to cr better than the veterinarians.