Expert systems in medicine

Early systems

In my view the earliest expert system in medicine was developed in the mid eighteenth century. (This view is contentious!) This early system was developed by a doctor in the royal navy as a method for diagnosis of the illnesses likely to be encountered in a tropical voyage. The problem faced by the navy was that many of its ships were too small to carry a doctor. There was also a shortage of doctors willing to serve onboard a ship.

The system developed comprised of a number of sticks which each had a symptom, like "high temperature" written on them. Further down the stick was a series of painted groves. The ships captain would select the sticks for the symptoms that were present line up the sticks and note the number of the groves where one set of groves lined up. The number gave the disease or condition. A book was used to describe treatment needed.

MYCIN

Most computer people would say that the first expert system was MYCIN. This was developed in the mid 1970s at Stamford university. It was a research tool and was never used in active medicine. MYCIN took 50 man years to encode and carried the knowledge of ten well respected experts in the field of diagnosis of meningitis, and related bacterial infections of the blood and membranes. Meningitis is often not diagnosed until it is too late, when this happens the patient often dies. Correct early treatment is therefore vital. Usually a sample of the blood will be incubated to grow the bacteria then identify them. This process takes upto a week by which time the disease may have become dominant.

MYCIN ran on a DEC 20 mainframe with a dumb terminal in the surgery. The program controlled the interview asking a series of 46 detailed questions. Some of the questions required an answer that could only be given by a doctor with a detailed knowledge of medical conditions. At then end of the questions MYCIN used a rule base to make a prescription. It explains its reasoning and recommends further tests.

MYCIN was extensively tested against human specialists. It correctly diagnosed 65% of the time. This compared with highly qualified human experts ranging from 42.5% to 62.5%. the success of MYCIN was based on for factors

1. Its knowledge base was detailed and based on the knowledge of the most highly placed experts in the field.
2. MYCIN never overlooks any of the details it is completely methodical.
3. MYCIN never jumps to "obvious" conclusions.
4. MYCIN is completely upto date it is maintained in a teaching university by doctors who constantly keep abreast of the latest methods, most doctors do not have time to keep upto date.

In the early eighties MYCIN spawned EMYCIN (empty MYCIN) a knowledge engineering tool designed to allow the development of unrelated expert systems. Subsequently MYCIN spawned GUIDON a rearranged MYCIN used to train doctors in diagnosis of these diseases.

One of the main criticisms of MYCIN is that is lacked the credibility that a consultant. People didn't trust a computer. The computer offers no empathy which is an important part of treatment.

NHS Direct

NHS direct offers a website (linked in the image above right) and has a phone-line diagnosis system that runs on an expert system for general diagnosis. It attempts to overcome the problem of trust and empathy by placing a nurse on the other end of the phone line to the patient. The nurse works the computer.

©LEV