MEDICAL IMAGE CATEGORIZATIONUSINGA TEXTURE BASEDSYMBOLIC DESCRIPTION

Inthefield ofmedical imageindexation, automatic categorization provides themeansforextracting, otherwise unavailable, information fromimages. Ourworkisfocused oncontent-based automatic medical imagecategorization methods, intheon-line context oftheCISMeFhealth-catalogue. Inthis study wepropose andevaluate areduced symbolic image representation. Thecategorization ofmedical images according totheir modality, anatomic region andviewangle isbased ontexture andstatistical features. We useamedical image dataset of10322images, representing 33classes, manually annotated byanexperienced radiologist. A topclassification accuracy of92.43%isobtained using k-Nearest Neighbors classifier ona64-label symbolic representation. Thisshows that thecompact symbolic imagerepresentation wepropose conveys enough oftheinitial texture information toobtain highrecognition rates, despite thecomplex context ofmultimodalmedical imagecategorization. IndexTerms-Content-based ImageRetrieval (CBIR), Internet, Medical ImageClassification, Feature Extraction, Symbolic Features