Artificial neural networks in pathology and medical laboratories

Abstract

Artificial neural networks in pathology and medical laboratories Clinical pathology laboratories are busier than ever.'Small pathology laboratories are merging with larger ones, which concentrates more work of the same repetitive nature in one place.As in the car industry, laboratory managers have turned to machines for these repetitive tasks because it is easier to recover the initially high capital cost of automation with high turnover.Machines now handle many types of clinical sample and are almost invariably linked to inexpensive high-performance microcomputers; these in turn have so enhanced the housekeeping aspects of a modern clinical pathology laboratory that "laboratory life before the computer" is almost impossible to comprehend.This integration of the computer within pathology laboratories, as record keeper, process controller, data tracker, and number cruncher, owes its success to the fact that computers excel in doing the same job very quickly without error, often unsupervised, 24 hours a day.It is easy to see how this leads to decreased sample turnaround time, a commodity increasingly demanded by managers and valued by clinicians.These roles for computers are perfectly tailored to the repetitive actions of the computer microprocessor chip; but they are really doing nothing more than acting as very fast electric filing cabinets.This article describes how artificial neural networks can extend the capabilities of computers within clinical pathology.Classifying data: when is more than speed needed?Programming a computer for tasks such as database management is straightforward, and this essentially "dumb" role for computers is now widely accepted by pathologists and laboratory managers.Other tasks relevant to clinical pathology laboratories are not so easy to specify as a computer algorithm.One of these is decision making.To involve computers in these tasks would be advantageous.This could not only increase sample throughput but also reduce the frequency of error.In pathology, decisions on samples are nothing more than the ability to classify data: "Is the histology benign or malignant?""Is the isolation of Staphylococcus aureus from this site relevant?""Is this lipid profile acceptable?".