超音波画像における深層学習を用いた肝硬変の自動診断

抄録

In recent years, the development of medical image diagnostic devices has made it possible to perform diagnosis using a large amount of medical data, but the increased burden on physicians has become a problem. Therefore, in this study, we applied deep learning to the ultrasound B-mode images of the liver to diagnose the presence or absence of diseases using a computer. The estimation of the liver parenchymal region in the ultrasonic image realized the Pixel accuracy of 0.88 and the conformance rate of 0.84. The classification of liver cirrhosis using ROI resulted in 0.88 classification results. According to the results of the study, liver cirrhosis is classified from a single image, but estimation from moving images is necessary for automation, and it is necessary to improve the selection method of ROI in order to improve accuracy.