Multiply sectioned Bayesian networks for neuromuscular diagnosis

Yang Xiang, B. Pant, A. Eisen, M.P. Beddoes, D. Poole
1993 Artificial Intelligence in Medicine  
A prototype neuromuscular diagnostic system (PAINULIM) that diagnoses painful or impaired upper limbs has been developed based on Bayesian networks. This paper presents nonmathematically the major knowledge representation issues that arose in the development of PAINULIM. Motivated by the computational overhead of large application domains, and the desire to provide a user with an interface that gives a focused display of a subdomain of current interest, we built PAINULIM using the idea of
more » ... ly sectioned Bayesian networks. A preliminary evaluation of PAINULIM with 76 patients has demonstrated good clinical performance. 1 This paper presents results of our research in developing the PAINULIM expert system for neuromuscular diagnosis involving a PAINful or impaired Upper LIMb. Our research involves the development of a general technique of multiply sectioned Bayesian networks and its application in the above mentioned medical domain. The emphasis of this paper is on the application. Readers are referred to 27] for mathematical details. Bayesian networks 18] combine probability theory with a graphical representation of domain models. Probability theory provides a language which embeds many intuitive inference patterns of reasoning under uncertainty and guarantees the consistency of inference made upon the representation. Graphical domain models convey directly to users the dependence and independence assumptions made in the domains, which facilitates knowledge acquisition and makes the representation more transparent. They also allow quick identi cation of dependence relations by tracing arcs in the networks and e cient computation in which di culty associated with general probabilistic reasoning 22] can be avoided when the networks are sparse. Some of the medical systems based on Bayesian networks include in internal medicine, MUNIN 1] in EMG, PATHFINDER 9] and INTELLIPATH 15] in pathology, and QUALICON 25] in nerve conduction studies. In the area of neuromuscular diagnosis, several (prototype) expert systems have appeared since the early 1980's: LOCALIZE 4] for localization of peripheral nerve lesions; MYOSYS 23] for diagnosing mono-and polyneuropathies; MYOLOG 6] for diagnosing plexus and root lesions; Blinowska and Verroust's system 3] for diagnosing carpal tunnel syndrome; ELECTRODIAGNOSTIC ASSISTANT 11] for diagnosing entrapment neuropathies, plexopathies, and radiculopathies; NEUROP for neuropathy diagnoses 19]; KANDID 5] and MUNIN 1] aiming at diagnosing the complete range of neuromuscular disorders. Most of the above systems in neuromuscular diagnosis are rule based. Satisfaction with system testing based on constructed cases has been reported, while ELECTRODIAGNOSTIC ASSISTANT reported clinical evaluation with a 78% agreement rate with electromyographers (EMGers), based on 15 cases. As medical diagnosis involves reasoning with uncertain knowledge, and limitations of rule-based systems for reasoning under uncertainty have been identi ed 7, 8, 18], we have chosen to build PAINULIM based on Bayesian belief networks. One exception in the above systems to rule-based structure is MUNIN which is based on Bayesian networks for its uncertain reasoning component. The MUNIN project started in the mid 80's in Denmark as part of the European ESPRIT program. MUNIN is planned to be a'full expert system' for neuromuscular diagnosis. Functionalities to be included are test-planning, test-guidance, test-set-up, signal processing of test results, diagnosis, and treatment recommendation. The intended users of MUNIN range from novice to experienced practitioners. The knowledge base ultimately will include full human neuroanatomy. MUNIN adopts Bayesian networks to represent probabilistic knowledge. Substantial contributions to Bayesian network techniques have been made (e.g., 14, 12]). The MUNIN system is to be developed in 3 stages. In the rst stage, a'nanohuman' model with 1 muscle and 3 possible diseases has been developed. In the second stage, a'microhuman' system with 6 muscles and corresponding nerves is to be developed. The last stage will correspond to a model of the full human neuroanatomy. PAINULIM Domain PAINULIM started in 1990 at the University of British Columbia with cooperation from the Neuromuscular Disease Unit (NDU) of the Vancouver General Hospital (VGH). Rather than attempting to cover the full range of diagnosis as does MUNIN, PAINULIM sets out to cover the more modest goal of performing diagnosis on patients su ering from a painful or impaired upper limb due to diseases of the spinal cord and/or the peripheral
doi:10.1016/0933-3657(93)90019-y pmid:8220685 fatcat:gshoatxernho3henmifu6qiwdq