Machine learning methods are innovative tools used to find patterns in medical data. Laboratory data is suited to computerized Interpretation because of its objective, quantitative nature. Body fluid analysis is a good model for evaluating machine learning in the laboratory. Pathologists spend a substantial amount of time analyzing and classifying body fluids, or effusions, which are abnormal accumulations of fluid within body cavities of human beings and animals, caused by diseases such as congestive heart failure. Fluid classification provides clinicians with important diagnostic information about the underlying disease process. Automation of body fluid analysis by a machine learning system would substantially increase the efficiency and profitability of a medical laboratory. In a pilot study, RIPPER (Repeated Incremental Pruning to Produce Error Reduction), a rule discovery tool, accurately classified effusions from animals into five standard categories, based on the physical, chemical, and cellular characteristics of the fluid. The purposes of this study are: 1) to determine the accuracy of RIPPER on a larger data set, to expand and strengthen the results of the pilot; 2) to test the accuracy of RIPPER's fluid classifications prospectively in a large veterinary teaching hospital laboratory, 3) to determine the acceptance rate or reason for rejection of RIPPER's classification by clinical pathologists; and (4) to use RIPPER to discover novel rules for classifying effusions by underlying disease process. The results of this study will validate and test the acceptance of a machine learning system applicable to fluid analysis in both human and veterinary clinical laboratories. By discovering new patterns in quantitative data that identify the specific underlying disease, RIPPER can greatly enhance the diagnostic value of laboratory analysis.
| Hotz, Christine S; Templeton, Steven J; Christopher, Mary M (2005) Comparative analysis of expert and machine-learning methods for classification of body cavity effusions in companion animals. J Vet Diagn Invest 17:158-64 |
