Research Article
Object-Specific Four-Path Network for Stroke Risk Stratification of Carotid Arteries in Ultrasound Images
Figure 3
Architecture of the object-specific four-path network (OSFP-Net). (a) The OSFP-Net is comprised of four paths for inputs of the bilateral carotid plaque images in both the transverse and longitudinal sections. Each path contains a feature extraction subnetwork (FE) and a feature downsampling subnetwork (FD). (b) The FE employs the same 5 convolutional and pooling blocks as VGG16, which are mainly used for image feature extraction. (c) The FDLS applies a multilevel strip pooling (MSP) strategy for the carotid plaques in the longitudinal section. (d) The FDTS employs a spatial pyramid pooling (SPP) strategy for the carotid plaques in the transverse section.