Information processing in upper digestive endoscopy

Provided with evolved functionalities, a digestive endoscopy atlas can be used as a training tool and even a diagnostic aid for "junior" endoscopists. Compared to a classic illustration of diagnoses by means of noteworthy iconography, such an atlas must in addition integrate similar-case retrieval and diagnostic hypothesis evaluation. That is why a case-based reasoning (CBR) tool, organized around two bases, one of endoscopic knowledge and one of case iconography, is well adapted to the objectives of conceiving a system that is able to improve. This paper focuses on the creation of two bases, in other words on the knowledge representation and on the iconography indexing. The concept of "scenes with objects" is useful for describing the endoscopic diagnoses as well as the images or sequences of the various cases. The endoscopic diagnosis description obviously falls within the province of an expert. He generally prefers to express uncertainty and vagueness in a qualitative form rather than in a quantitative one. Moreover, thanks to a common description mode, the knowledge base takes the place of a thesaurus, which is valuable for the indexing of cases. Classification algorithms are tested on simulated endoscopic descriptions and the results are very promising.