Authors: Mina Amiri , Rupert Brooks , Hassan Rivaz
DOI: 10.1109/TUFFC.2020.3015081
Keywords:
Description: One way of resolving the problem scarce and expensive data in deep learning for medical applications is using transfer fine-tuning a network which has been trained on large set. The common practice to keep shallow layers unchanged modify deeper according new This approach may not work when U-Net moving from different domain ultrasound (US) images due their drastically appearance. In this study, we investigated effect sets pretrained US image segmentation. Two schemes were analyzed, based two definitions layers. We studied simulated images, as well human sets. also included chest X-ray results showed that choosing fine-tune critical task. particular, they demonstrated last network, classification networks, often worst strategy. It therefore be more appropriate rather than segmentation U-Net. Shallow learn lower level features are automatic images. Even set available, observed faster compared whole network.
.The resource attribute category is marked as computer automatic recognition, which may not be accurate. You can try clicking the link to view the resource details.