Model-based diagnosis of brain disorders: a prototype framework.

Abstract

This paper describes a prototype framework, named NEUROLAB, dedicated to research and diagnosis in the area of brain disorders. The diagnostic task uses a blending of factual knowledge, formal knowledge, and experiential knowledge. The prototype's first target clinical application is partial seizures in epilepsy. Diagnosis is carried out using qualitative electroencephalographic descriptions, clinical attack pattern descriptions, and pre- and post-ictal observations. From this information, the system builds explanations in the form of candidate epileptogenic foci and trajectories of the seizure spread. Hypothesis-testing and discrimination is based on minimal set coverage, and consistency-checking is performed using the general background knowledge. Upon completion, NEUROLAB will provide specific physiological knowledge for solving the so-called inverse problems in electroencephalography (EEG) and magnetoencephalography (MEG).