MDX and Related Medical Decision-Making Systems

MDX AND RELATED MEDICAL DECISIOtHW<IN:i SYS'J»15

B. Olandrasekaran, s. Hittal and J. w. Snith*

Department of Comp..lter & Information SCience The Ohio State University

Columbus, Cll 43210, u. s. A

!IDX is a medical diagnosis system, organized as a hierarchica collection of conceptual specialists, who cooperate to diagnose a case. 'nte current danain of ~D:lX is cholestasis, a syndrane involving the liver. MDX works in conjunction with a conceptually organized data base specialist (PATREC) and radiology consultant (RADEX) ·

PR(X;RJ\M DESCRIPI'ION

In the last IJCAI at TOkyo, we presented an approach to knowledge organization and problem solving for diagnostic tasks, and described a prototype system called ~IDX based on that approach [1]. Several aspects of system design and performance were discussed in (2], and the central theoretical ideas underlying the approach are presented, and further extensions elaborated, in [3].

There are two key aspects to the methodology : (1) 1he diagnostic knowledge is decomposed into a collection of hierarchically organized cooce~ua1 specialists; the scope of the specialists varies from more general to more particular as the hierarchy is traversed from top to bottom. (2) The problem solving proceeds in a top-down manner; the specialists in the hierarchy establish or reject their relevance to the case at hand, and those who have established their relevance pass the control to their successors, until all relevant specialists have added their explanation of abnormal data which the knowledge in their scope can explain.

The ~IDX system, which is wholly diagnostic in its knowledge, communicates with two auxiliary systems, PATREC and RADEX. PATREC is a data base

• *Acknowledgement During the performance of this research, J. w. Snith was supported by National Library of ~~icine Biomedical Computing Training Grant to the Ohio State University, 5-T01-LM07023-02. We are also pleased to acknowledge the use of the computer facility of the Rutgers University Laboratory for Computer SCience, made possible by the Biotechnology Resources Program of the NIH under Grant RR-00643.

assistant in the sense it aCX}uires the patient data, organizes them, and answers the queries of MDX concerning the patient data. In all these activities PATREC uses various types of inferential knowledge el1bedded in an underlying conceptual roodel of the danain of medical data. RADEX is a radiology consultant to MDX, and it suggests or confirms diagnostic possibilities by reasoning based on its knowledge of imagi11<J procedures and relevant anatomy. See [4] and [5] for further details about these subsystems.

The current danain of MDX is a syndrome called cholestasis, which covers a large number of specific liver diseases. The pilot system was restricted to extra-hepatic diseases only; however, in the last several months, a large number of intrahepatic diseases have been added. The system is continually tested on a number of cases both from journals and hospital records. MDX is written in LISP, and is quite large : it rw-IS in 120K words of DEC-20 memory, in addition to the memory requirements of the LISP system itself.

1. B. Olandrasekaran, F. Gomez, s. Mittal, and J. w. Snith,. "An approach to medical diagnosis based on conceptual structures,"Proc. IJCAI, 1979, pp. 134-142.
2. S. l1ittal, B. Chandrasekaran, and J. w. Snith, "OVerview of MOX--a system for medical diagnosis," Proc. III Ann. 5;wp. Coop.!ter Applications in Medical care, 1979.
3. (3] F. Gomez and B. Olandrasekaran, "Knowledge organization and distribution for medical diagnosis," IEEE Trans. 5M&C, Jan. 1981. pp. 34-42.
4. s. Mittal and B. Chandrasekaran, "Conceptual representation of patient data bases, · J. ~tedical Systems, vol. 4, no. 2., 1980.
5. B. Olandrasekaran, s. Mittal, and J. w. Snith, "RADEX--towards a computer-based radiology consultant," in Pattern Becognitign in Practice, Gelsema & Kanal, Ed., pp. 463-474, North-Holland, 1980.