Diagnostic reasoning by classification in upper digestive tract endoscopy

The purpose of the authors´ project is to build an atlas of indexed endoscopical lesions that could be used in computer-assisted diagnostics as referring data. The analysis of both endoscopists´ experience and computerized record format developed by professional shows that the diagnostic reasoning in digestive endoscopy uses a scene-object approach. The objects correspond to the endoscopical findings and the medical context of examination, and the scene to the endoscopical diagnosis. Based on expert assessment, a computer-assisted diagnosis system is described using a classification reasoning approach. The classes of endoscopical findings and diagnoses are listed and the primitive characteristics of objects are extracted. Each class content is defined in an intensive way, and symbolic distances between descriptors and classes leads to an a priori description of classes. Symbolic distances are transformed by heuristic in numeric measures that allow estimation of the membership value of an observed case to a defined class. A simulation test with 50 randomized objects by class demonstrates a good classification of endoscopical findings. The correct class is the unique Response in 71% of the tested objects, the first of multiple Responses in 23% and one of the multiple responses in 6%. Four descriptors are of major importance in the classification algorithm: anatomic location, shape, color and relief. A computerized application is currently developed to archive endoscopical cases according to the classification method