ABEL

ABEL Acid-Base ELectolyte program

ABEL (1979) was developed at MIT to be the diagnostic component of an overall medical consulation system.

Claim: "A diagnostic or therapeutic program must consider a case at various levels of detail in order to integrate its overall understanding with its detailed knowledge."

Knowledge Representation

ABEL uses a patient-specific model (PSM) to capture both the data about the patient and the program's interpretation of the data on a mulitlevel causal basis.

The PSM is generated by instantiating portions of ABEL's general medical knowledge and filling in details from the specific case. It's instantiation is strongly influenced by the initial data (preliminary diagnostic hypothesis). This is to mirror the physican's general tactic of using a tentative diagnosis to drive his testing.

The general drive of the instantiation is to gradually shift the description of the patient case from physiological to syndromic knowledge. The syndromic knowledge can then be used to determine a diagnosis.

There are 5 operators for this instantiation:

1. Aggregation
2. Elaboration
3. Component Decomposition
4. Component Summation
5. Projection

Each node is a normal or abnormal physiological state with an associated list of attributes such as temporal characteristics and severity.

Each link represents some relation (causal, associational, etc) between different states. These links are more than just "pointers", they are a sort of multivariate mapping function which depends on the severity and duration of the cause, the patient's age, sex, weight, etc. and the current set of hypotheses. (annotated links)

Figure 14.1 (or 2.6): consistituent-of

Movement between the level's is accomplished by the application of the Aggregation and Elaboration operators.

Aggregation is a way of abstracting a set of states at one level into one state at a higher level. Elaboration is the opposite.

Figure 14-2:6 (or 2.7-12)

Particularly 14-6

Component Decomposition and Summation allow for additional information to be gained at the same level. A primary node (one that cannot be elaborated) may be broken down into components in order to better account for some data (decomposition). This allows for cumulative effects, i.e., 2 different things causing the same symptom and each contributing a certain portion to it.

Figure 14-7,8 (or 2.16, 17)

Projection provides for causal links to be built in the same level by hypothesizing links which are then checked for consistency with levels above and below. This operator is particularly useful during diagnosis to explore the implications of a given hypothesis. (guestion: is this how we reason - or is this a check? ... sorta like rough design? assume, then get focus...?)

Building PSM's and Diagnosing

1. Collect initial data, generally a blood test and patient history
2. Generate all possible acis-base disturbances that can account for the laboratory data
3. Based on the complexity and likelihood of each component, prune the list and rank order it.
4. Create a PSM for each acid-base disturbance on the pruned list by using the operators outlined above
5. Continue diagnosis by choosing tests based on the status of each PSM and adding the new information to the PSM's

Diagnosis of multiple diseases with overlapping or canceling effects is possible because of component decomposition and summation.

An explanation of the state of the patient can be obtained at any of the levels. These explanations represent varying levels of "understanding"

Figure 14-9.

Previous work (from Patil paper)

Deep vs Compiled reasoners

Model-based Reasoning