John R. Svirbely

RED: A red-cell antibody identification expert module

We describe a software module in an expert system RED, which interprets data related to red cell antibody identification. There are three portions to this module: (1) the problem-solving component, which incorporates the knowledge required for antibody identification as a hierarchy of programs. The programs in the hierarchy organize within themselves small pieces of knowledge represented in the form of production rules, which are capable of making judgments concerning a specific hypothesis; (2) an intelligent data base for storage of patient data, red cell attributes, and test results; (3) the “overview critic” portion, which combines the atomic hypotheses judged favorably by the antibody programs into a unified judgment concerning the case. Overview makes the decision to terminate processing with a conclusion about which antibodies are actually present and what specific further tests need to be performed to resolve any remaining ambiguities.

Coping with the complexities of multiple-solution problems: a case study

A model is proposed to account for the expertise of a skilled immunohematologist in solving multiple-solution problems. These problems, which he must deal with daily, are concerned with ensuring the safe transfusion of blood into patients. This model suggests that he copes with this difficult class of problems by: (1) Using patterns in the data to simplify the problem, hypothesizing the number of primitive solutions necessary to account for the test results and, when possibles, decomposing the problem into a set of less complex, single-solution problems. Such decompositions then enable more powerful reasoning processes; (2) making use of a mixture of data-driven and hypothesis-driven processes in order to reduce the chances that heuristic (and therefore error-prone) methods and cognitive biases will lead away from critical data; (3) Relying on a mixture of confirmatory and rule-out processes to provide converging evidence, thus reducing the changes of error; (4) uncovering his own errors through the use of “error models” that identify the conditions where one of his processes is likely to make an error (similar to the use of student models by expert tutors to diagnose mistakes made by students).

An empirical evaluation of the performance of antibody identification tasks.

Four empirical studies were conducted for better understanding of the nature of problem‐solving activities by medical technologists and medical technology students when performing antibody identification tasks. The results indicated the importance of strategies that ensure the collection of converging evidence, as these strategies protect against the fallibility of commonly used heuristics and against errors due to simple slips. The results also indicate that not only do students make significant numbers of errors, but so do practicing technologists. In one of the studies covering a 1‐year period, for instance, a group of 16 technologists made a total of 41 errors in 1057 cases. On the basis of these findings, several alternatives are proposed to reduce errors.

Qualitative representation of behavior in the medical domain

In medical knowledge-based systems, there is a need for models of the human body which can predict and explain behavior based on qualitative information about the structure and behavior of the body. We present a framework for representing the structure and behavior of physical systems and apply it to the human cardiovascular system. Our framework allows for hierarchical representation of physical systems and facilitates two kinds of reasoning processes: composition of behaviors and qualitative simulation.

The role of essential explanation in abduction

The abduction task is to infer the best explanation for a given set of data. One common subtask of abduction is to synthesize the best composite explanatory hypothesis from elementary hypotheses retrieved from memory. The synthesis of best composite explanations, however, is computationally costly. One general approach to controlling the computational cost of synthesizing explanations is to decompose the synthesis search space into smaller spaces that can be searched more efficiently and effectively. The essential hypotheses, that is, the hypotheses that are the only available explanations for specific subsets of the data set, provide one such decomposition. In this method, first the essential hypotheses are included in the composite explanation, and, then, non-essential hypotheses are included to account for the remaining unexplained data elements. In addition to providing a more efficient method for synthesizing composite explanations, this decomposition leads to the formation of more parsimonious explanations. In this paper, we report on a set of experiments in the domain of medical data interpretation that demonstrates that the essential/non-essential decomposition of the abduction search space results in more efficient synthesis of more parsimonious composite explanations.

Errors in abductive reasoning

Results of efforts to extract knowledge for an expert whose job is to detect the errors made by practicing technologies are examined. These errors are discussed in terms of possible cognitive biases. The example examined involves students, technologies, and experts performing antibody identification tasks in order to construct a critiquing and intelligent tutoring system.<<ETX>>

MDX-MYCIN: The MDX paradigm applied to the mycin domain

Comparison of different approaches to expert system design for a given task, such as diagnosis, is difficult since they are eoften embodied in systems for domains with very different characteristics. It is a priori difficult to decide if a given difference in the approaches is necessitated by the differences in the domain. For example, it might be suggested that MYCINs global and numeric uncertainty calculus is needed in domains such as MYCINs, apparently characterized by a great deal of uncertainty in knowledge and data, while the approach of MDX, another medical system, which uses local combinations of qualitative probabilities only, may be too weak in such domains. In order to study the relationship between the domain characteristics and problem-solving approaches of the two systems, we constructed an MDX-like system for a subdomain of MYCIN and conducted a number of experiments on the resulting system. The results demonstrate that the MDX paradigm is effective in this domain and, additionally, offers knowledge engineering advantages along the dimension of debugging ease and system extensibility.

The Transfusion Medicine Tutor: Methods and Results from the Development of an Interactive Learning Environment for Teaching Problem-Solving Skills

Investigations of students and practicing medical technologists indicate that both groups make significant numbers of errors on tasks such as antibody identification. One potential solution to help with this problem is to provide access to a computerized learning environment in which users can get exposure to a larger and much broader set of problems than would otherwise be possible. This paper describes such a learning environment, the Transfusion Medicine Tutor, and discusses the ways in which it supports guided discovery learning.

The transfusion medicine tutor: a case study in the design of an intelligent tutoring system

Empirical studies of medical technology students indicated that there is a considerable need for additional skill development in performing tasks such as antibody identification. The transfusion medicine tutor (TMT) explores the use of a model of expert problem-solving to direct the application of a variety of teaching strategies in such an environment. This tutoring system focuses on red cell antibody identification, which is a problem-solving task where multiple primitive solutions can be true simultaneously. The approach to tutoring is the design of a problem-solving environment in which the computer plays an active role in tutoring the student. TMT allows students to gain experience by exploring a wide range of patient cases. The design of TMT makes it easy for the computer to detect certain types of errors. TMT allows more advanced students to compare the conclusions they have drawn from a particular test result with the computers interpretation of the same data.<<ETX>>

Design concepts underlying the use of an expert system to teach diagnostic reasoning for antibody identification

Antibody identification is a laboratory task where medical technologists must select tests to run and interpret the results in order to determine the antibodies in a patients blood. It has the classical characteristics of an abduction task, including masking and problems with noisy data. Based on a cognitive analysis of both successful and error-producing performances, a tutoring system (the transfusion medicine tutor, TMT), was developed that uses expert systems technology to provide immediate, context-sensitive feedback as students solve actual patient cases. Development of this system required consideration of aspects of artificial intelligence, education, psychology, and human factors engineering, as well as the domain of study (i.e., allo-antibody identification). In a formal field evaluation, when used by an instructor as a tool to assist with tutoring in a class laboratory setting, use of TMT resulted in improvements in antibody identification performance of 87-93% (p<.001) as compared to a passive control version which improved performance by 20%.