深層学習を用いた肝血管腫の抽出手法

抄録

Hepatic hemangioma is lesion frequently found in ultrasonography. Although hepatic hemangioma is benign tumor, other tumors are subject to detailed examination by judging hepatic hemangioma. Therefore, in this study, we proposed a new discrimination method aimed at improving accuracy by using deep learning to classify hepatic hemangioma and blood vessels appearing on ultrasound images. This is prevents feature loss system using learning model that removes diaphragm and kidney, sorts out image features to increase the proportion of classification targets. In the experiment of this study, 132 ultrasound images were divided into 120 Training images and 12 Test images. As a result, it was possible to estimate with accuracy of 89.4% in classification of hepatic hemangioma and 79.7% in blood vessel classification by the proposed method.