The writer Dr. RK Umesh Singh, is a REGIONAL MANAGER for SIEMENS, MEDICAL INSTRUMENTS SYSTEM in JEDDAH (KSA), since last 6yrs. SIEMENS is a global Multinational (GERMANY) Co.

The development of computer programs that can perform diagnosis or make treatment decisions has been a key goal of medical informatics. A seminal paper published in science in 1959 first suggested that probability theory, propositional logic, and decision theory could serve as the computational basis for such programs.

The 1960's saw immediate application of this idea in the area of probabilistic diagnosis of congenital heart disease and of acute abdomen problems. Other programs were based on a variety of artificial intelligence(AI) techniques (e.g. MYCIN family of a well known expert system for treatment of bacterimia and meningitis developed at Stanford university, USA), and INTERNIST a diagnostic program in the domain of general internal medicine developed at the University of Pittsburgh(USA).

Several of these projects have spawned commercially available products. A directory of Artificial intelligence systems in routine clinical use is maintained at www-uk.hpl.hp.com/people/ewc/list.html.

Efforts to program computers to perform diagnosis have led to an understanding of how Clinicians (mis)interprete clinical data and more importantly, development of techniques for the correct interpretation of clinical data (e.g. the effect of sensitivity and specificity of a test on the posterior probability of a diagnosis, given the result of a test).This work has also yielded quantitative techniques foe measuring and incorporating patient preferences in high risk decisions. Many of these topics are now taught in medical school curricula and permeate the published clinical literature (in developed world).

Before adopting any of these programs in the state, the practicing physicians should be aware that each program is based on a particular model of diagnosis that embodies simplifying assumptions. For example, many programs make the assumptions that the patient has one of the diseases that the program knows about (e.g. the program cannot make a diagnosis of normal).Some programs assume that the patient has only one disease and most cannot attribute a finding (e.g., a rash) to a drug and therefore attempt to find a disease to explain the finding. The net effect of these modelling assumptions is that the programs can make diagnostic errors.

The safest policy is to understand the limitations of the program and, moreover, to use the suggested diagnosis solely as a memory jogger. An additional issue to consider is whether the program fits into he clinical routine. Typically, diagnostic program require the physician to type in symptoms, physical findings and laboratory test results. The time required to enter data often affects the usefulness of these systems.