End-to-end 3D FCN prostate segmentation based on attention mechanism
题目：End-to-end3D FCN prostate segmentation based on attention mechanism
摘要：Thedetermination of prostate volume (PV) is helpful in assessing prostate disease.Especially, combined with other parameters, it can help to predict thepathological stage of prostate cancer, aiding diagnosis and treatment. We usedan end-to-end 3D fully convolutional network (FCN) based on the attentionmechanism to segment the prostate from MR Images. The model uses the attention mechanism tosuppress uncorrelated areas in the input image, meanwhile highlights thesalient features that are useful for a particular task. The advantage of themodel is that it reduces excessive and redundant use of computational resourcesand model parameters. At the same time, multi-scale image feature fusion isrealized by long skip connection. The model performed well in the validationset results, while the results of some test data are still not good. Probablereasons may be large differences between training samples,insufficient amount of training samples or poor image preprocessing etc.