A Bag of Features Approach for CEUS Liver Lesions Investigation

摘要: In this work a novel approach for CEUS based diagnosis is presented. We propose spatial/image-based method using parallel and hierarchical system architecture. As feature extraction stage, we the Bag of Features (BoF) algorithm which treats image features as bag visual words. It followed by multiclass SVM classifier trained separately each phase ultrasound investigation. A soft voting scheme has been proposed information fusion individual classifiers. The preliminary evaluation shows promising qualitative results our on samples newly introduced dataset. Using only 550 images, (5 liver lesions x 10 pictures/lesion 11 patients) an average accuracy 64% obtained leave-one patient-out procedure.