Comparative Analysis of Expert and Machine-Learning Methods for Classification of Body Cavity Effusions in Companion Animals

Abstract

A rule-based expert system using CLIPS programming language was created to classify body cavity effusions as transudates, modified transudates, exudates, chylous, and hemorrhagic effusions. The diagnostic accuracy of the rule-based system was compared with that produced by 2 machine-learning methods: Rosetta, a rough sets algorithm and RIPPER, a rule-induction method. Results of 508 body cavity fluid analyses (canine, feline, equine) obtained from the University of California–Davis Veterinary Medical Teaching Hospital computerized patient database were used to test CLIPS and to test and train RIPPER and Rosetta. The CLIPS system, using 17 rules, achieved an accuracy of 93.5% compared with pathologist consensus diagnoses. Rosetta accurately classified 91% of effusions by using 5,479 rules. RIPPER achieved the greatest accuracy (95.5%) using only 10 rules. When the original rules of the CLIPS application were replaced with those of RIPPER, the accuracy rates were identical. These results suggest that both rule-based expert systems and machine-learning methods hold promise for the preliminary classification of body fluids in the clinical laboratory.

Keywords

Computer decision support effusion expert system laboratory information machine-learning system

References

Baker

Lumsden

: 2000, Color atlas of cytology of the dog and cat, 1st ed., pp. 159–176. Mosby Inc., St. Louis, MO.

Cowell

Tyler

Meinkoth

: 1999, Abdominal and thoracic fluid. In: Diagnostic cytology of the dog and cat, ed. Cowell

Tyler

Meinkoth

, 2nd ed., pp. 142–158. Mosby Inc., St. Louis, MO.

Duncan

Prasse

Mahaffey

: 1994, Veterinary laboratory medicine-clinical pathology, 3rd ed. Iowa State University Press, Ames, IA.

Edwards

Compton

Malor

: 1993, PEIRS: a pathologist-maintained expert system for the interpretation of chemical pathology reports. Pathology 25:27–34.

Giarratano

Riley

: 1998, Expert systems-principles and programming, 3rd ed., pp. 477–508. PWS Publishing, Boston, MA.

Innis

: 1997, Clinical problem solving-the role of expert laboratory systems. Med Inf 22:251–261.

Ivandic

Hofmann

Guder

: 1996, Development and evaluation of a urine protein expert system. Clin Chem 42:1214–1222.

Kononenko

Bratko

Kukar

: 1997, Application of machine learning to medical diagnosis. In: Machine learning and data mining-methods and applications, ed. Michalski

Bratko

Kubat

, pp. 389–408. John Wiley and Sons Ltd, Chichester, West Sussex, England.

Ohrn

Komorowski

Skowron

Synak

: 1998, The design and implementation of a knowledge discovery toolkit based on rough set-the ROSETTA system. In: Rough sets in knowledge discovery 1-methodology and applications, ed. Polkowski

Skowron

, pp. 376–399. Physica-Verlag, Heidelberg, Germany.

10.

Olivier

Willard

: 1994, Fluid accumulation disorders. In: Small animal clinical diagnosis by laboratory methods, ed. Willard

Tvedten

Turnwald

, 2nd ed., pp. 219–231. WB Saunders, Philadelphia, PA.

11.

Parry

Brownlow

: 1992, Peritoneal fluid. In: Cytology and hematology of the horse, ed. Cowell

Tyler

, pp. 121–150. American Veterinary Publications, Goleta, CA.

12.

Spackman

Connelly

: 1987, Knowledge-based systems in laboratory medicine and pathology. Arch Pathol Lab Med 111:116–119.

13.

Winkel

: 1989, The application of expert systems in the clinical laboratory. Clin Chem 35:1595–1600.

14.

Witten

Frank

: 2000, Data mining-practical machine learning tools and techniques with Java implementations, 1st ed., pp. 119–156. Morgan Kaufmann Publishers, San Francisco, CA.