Abductive localization of brain damage: incorporating spatial adjacency relations

Stanley Tuhrim, Deborah R. Horowitz, James A. Reggia, Sharon M. Goodall

Research output: Contribution to journalArticlepeer-review

2 Scopus citations

Abstract

An important problem in neurology is to localize the site of damage to the nervous system given a patient's examination findings. While the reasoning processes involved in this neurologic localization task are a type of diagnostic reasoning, they are distinguished by their heavy use of anatomical (spatial) relationships. Previous attempts to automate neurological localization have met with limited success. This paper describes an abductive problem-solving method for neurological localization based on parsimonious covering theory (PCT). Basic PCT is augmented by adding spatial relationships between elementary anatomic units. Our model's localization for 100 stroke patients was compared to that of a neurologist specializing in stroke who was not involved with the model's development. In 99 cases, the problem-solving system based on the augmented PCT algorithm identified the location of nervous system damage (brainstem or either hemisphere) found by the stroke expert. In the one case of complete disagreement, the problem-solving system was proven correct. Examination of the detailed localizations in terms of the elementary anatomical units involved indicated a number of interesting differences between human and automated inference processes. These results demonstrate that an augmented PCT approach has substantial promise for neurological localization.

Original languageEnglish
Pages (from-to)75-85
Number of pages11
JournalArtificial Intelligence in Medicine
Volume3
Issue number2
DOIs
StatePublished - Apr 1991

Keywords

  • Neurological localization
  • abductive problem solving
  • diagnostic reasoning
  • parsimonious covering

Fingerprint

Dive into the research topics of 'Abductive localization of brain damage: incorporating spatial adjacency relations'. Together they form a unique fingerprint.

Cite this