Earliest Medical Diagnosing AIs

DENDRAL system (1969), by Bruce Buchaman and Joshua Lederberg.

• Goal: Infer molecular structure from a mass spectrometer’s information
• First expert system
• Approach:
  • Based on a set of rules
  • AND-OR tree
  • AO* search algorithm

MYCIN expert system (1972), by Feigenbaum, Buchaman, and Edward Shortliffe

• Goal: Diagnose blood infections
• Novelty: Incorporate “Probabilistic Reasoning”, or “certainty factors”
• Approach:
  • Based on about 450 rules
  • Difference from DENDRAL: no theoretical model
  • Acquired knowledge from medical textbooks and expert’s interviews
  • Simple inference engine + knowledge base

Computer-aided Diagnosis (1972), by Horrocks et al.

• Goal: Diagnose acute abdominal illness

Comment: At this point, early Bayesian system suffered from a number of problems. Thus probabilistic methods for coping with uncertainty fell out of favor in AI from the 1970s to the mid-1980s

MUNIN (1989), by Andersen et al.

• Goal: Diagnose neuromuscular disorders

PATHFINDER (1991), by Heckerman

• Goal: Pathology

CPCS system (1994), by Pradham et al.

• Goal: internal medicine
• Approach:
  • 448 nodes, 906 links, and 8,254 conditional probability values

References:

• Stuart Russell, Peter Norvig 1995. Artificial Intelligence: A Modern Approach. Upper Saddle River, New Jersey.
• Horrocks et al. (1972). Computer-aided Diagnosis: Description of an Adaptable System, and Operational Experience with 2,034 Cases. British Medical Journal, 2, 5-9
• https://expertsystem101.weebly.com/dendral.html
• https://expertsystem101.weebly.com/mycin.html

Other resources:

• https://www.britannica.com/technology/artificial-intelligence/Expert-systems