An Improved Algorithm for Segmentation of Retinal Detachment Images
To achieve segmentation of lesions in ultrasound images and improve target segmentation accuracy, we propose a residual structured deep scale fusion attention network model, RDFA-Net. Using an improved residual structure based network model and deform-able convolution to increase the sampling range, using channel and spatial attention mechanisms to capture features, and integrating deep scale features for fusion to capture deep level image features, the model successfully segmented the ultrasound images of retinal detach-ment lesions and produced better segmentation results. The self-made segmented dataset provided by the cooperative hospital is used to evaluate RDFA- Net. Through the comparison of experimental results, we found that RDFA-Net is effective and achieves advanced per-formance.
万方数据
维普
