Carotid Artery Segmentation on Ultrasound Image using Deep Learning based on Non-Local Means-based Speckle Filtering

摘要

Cardiovascular disease (CVD) causes significant deaths worldwide, of which 17.3 million deaths per year are due to CVD. The use of Ultrasound is necessary to see the abnormalities. The study will segment Carotid Artery segmentation on the Ultrasound image by using the U-Net-based architecture of non-local means-based speckle filtering (NLMBSF). The images will use NLMBSF to reduce speckles, and the data set will be divided into two parts, namely the dataset, which using NLMBSF and not NLMBSF. After that, doing training to create a U-net model, the training data model results will be searched with the best Accuracy. The obtained result of the study is an accuracy value of 97.74%, dice value is 87.22%, and a loss of 0.0107 on data that does not use NLMBSF. Still, it got different data results using NLMBSF, namely 97.6% accuracy, dice value is 84.06% and 0.0138 value loss.