A problem decomposition method for efficient diagnosis and interpretation of multiple disorders

Diagnosis of multiple disorders can be made more efficient by reasoning explicitly about problem decompositions. A diagnostic problem can be decomposed by hypothesizing about common and disjoint cause relationships among the given symptoms. The resulting structure exploits computational principles of causal intersection, subproblem independence, and minimal factorability to increase efficiency. By assigning structure to a problem, the symptom decomposition approach offers a new type of decision-support task called symptom interpretation. Experimental results indicate that symptom decomposition yields substantial increases in performance compared to existing methods for multidisorder diagnosis.