Biomedical subjects

B S Todd

Publications and source records attributed to B S Todd.

3 recordsLinked to original sources

PubMed1994

The relative accuracy of a variety of medical diagnostic programs.

Acute abdominal pain is one of the most widely studied applications of computer-aided diagnosis. The usual approach is to apply Bayes' theorem with the assumption of conditional independence ("independence Bayes"). We compared various approaches to designing diagnostic programs for abdominal pain of suspected gynaecological origin. The methods range from statistical to knowledge-based. All programs were evaluated using a database of 1,270 cases collected retrospectively. Our results suggest that in this application no significant improvement in accuracy can be made by taking interactions into account, either by statistical or by knowledge-based means; independence Bayes is near-optimal. As far as accuracy is concerned, there appears to be little point in pursuing knowledge-based approaches. However, the "nearest neighbours" method using a new metric appears to be at least as accurate as independence Bayes. We argue that the nearest neighbours method is more suitable than independence Bayes for clinical use because of greater accountability.

Abdominal Pain↗

PubMed1990

An algorithm for approximating conditional probabilities.

When diagnostic programs are constructed within a probabilistic framework, it is often the case that computation of joint probabilities of exhaustive combinations of events is easy, but computation of the kind of conditional probabilities the user wishes to know, is hard. This paper describes a simple algorithm for computing the required values, and then suggests several heuristic optimizations that may enable suitable approximations to be obtained in a feasible time when the task is otherwise intractable. An account is given of a specific application of the method in the construction of a medical diagnostic program, which is described in more detail elsewhere.

Algorithms↗

PubMedDate not supplied

A probabilistic simulation model to assist the localization of nerve lesions.

I present a formal, mathematical specification of a probabilistic expert system to assist the localization of nerve lesions. The program is based on an anatomical model of the peripheral nervous system of the human upper limb. The simulation model defines a joint probability distribution over the states of nerves and clinical manifestations. A simple, general-purpose heuristic algorithm is used to approximate conditional probabilities of interest. It is shown how an upper bound on the expected approximation error can be measured experimentally; this upper bound is 0.05 for the system described here, although the bound can be made arbitrarily small by expending more computational effort. The expert system is compared with the nearest-neighbour statistical classification rule on two databases of 26 and 25 cases respectively. The expert system makes fewer errors, although the observed difference does not reach statistical significance. Possible future refinements to the model are explored, and the advantages of specifying expert systems formally are discussed.

Algorithms↗