The next operator does the local processing on the PSM to customize it to the patient by adjusting the probabilities, asserting any nodes that have definite values, and checking for inconsistencies. The first step is to go through the links in the PSM and do all of the partial evaluation possible on the probabilities to take into account input information such as patient age and sex. Then the program can compute the probabilities along paths to prepare the PSM for the probabilistic reasoning of diagnosis.
The local determination of the state of nodes in the patient specific model can be done at several levels. The primary source of information for this determination is the value structures created by the evidence generator. If there is a value with only one cause (and it requires a cause), that cause node is asserted true. If there is a node that always produces a value and that value is asserted false or otherwise inconsistent with the cause, the node is false. Nodes asserted in this way are certain and the automatic state determination stops at this point. However, it is sometimes useful for the user to assume that nodes with high probability are also true. Because of the large number of loops in the causal model, it is not computationally feasible to compute the exact probability of the nodes. It is possible to compute a conservative estimate of the probability of a node by locally comparing the probability of the node being true from the immediate evidence, to the probability that the evidence is caused by other nodes. If the probabilities of the causes are conservatively estimated and the threshold for asserting the node is high, this method identifies nodes that have a high probability from which the user can select ones to be asserted.
These determinations are added to the PSM and constrain it. There is also a truth maintenance system (TMS)[18] that assures that the assertions added to the PSM are consistent and adds any further implications of the nodes. Technically a TMS is not necessary because the probabilities of such nodes would be 1 or 0. In fact, the TMS in effect precomputes when the probabilities would go to 1 or 0 and provides an efficient mechanism for asserting such facts. It also provides a mechanism to track the source of any inconsistencies in the input. As a result, all hypotheses considered later in diagnosis include the nodes known to be true and exclude those known to be false. If a considerable amount of information is known about the patient, this simplifies differential diagnosis. When this operator is finished, the PSM has all of the information from the input and all of its definite implications asserted.
This operator is also used after the differential diagnosis to assert the consensus nodes. That is, the hypotheses with any significant probability are intersected to determine what nodes are common to all of them. These are then added to the PSM with the appropriate justifications to simplify the task of information gathering for diagnosis refinement and therapy selection.