Medical images are an important tool for doctors to diagnose conditions and treat diseases, and performing accurate medical image segmentation is the basis for disease diagnosis and treatment planning. Currently convolutional neural networks have achieved significant results in the field of medical image segmentation. However, considering the actual usage scenario, the model needs to run on resource-constrained devices, so the model needs to be lightweighted. ESPNet is a lightweight segmentation model structure. The ESP module effectively reduces the number of model parameters and computation, but in this paper, we note that directly reducing the number of model parameters by point-wise convolution will lead to the loss of model feature map information, which in turn leads to the degradation of model performance. In this paper, in order to further reduce the number of model parameters based on the ESPNet model, the number of channels of all model feature maps of the ESPNet model is halved, and in order to mitigate the resulting degradation of model performance, the feature maps of the model are grouped in the channel dimension using the modified Shuffle-ESP module. In order to avoid the loss of information interaction between different grouped convolutional feature maps, the channel information is artificially interacted using a channel shuffle mechanism before entering the atrous convolution of different dilation rate. It is experimentally demonstrated that the model in this paper decreases 54.99% and 54.99% compared to the original model parameters on two tumor data, and that the model performance metrics decrease by 1.59% and 1.76% respectively. The superiority of the proposed model in this paper is demonstrated through experiments.
This paper proposes a new liver tumor segmentation method based on mixed domain attention mechanism. Firstly, by combining the well-known SENet with channel attention architecture, cross channels are interacted at their Excitations by using multiple one-dimensional convolution kernels. Then, multiple dilation convolution is used to increase the receptive field in the spatial attention module of BAM. After then, the channel and spatial attention feature maps are fused to recorrect the original feature map. Finally, a gating mechanism is introduced at the sampling skip connection on the decoder to filter important features. The experimental results show that in many evaluation indexes, the accuracy of this method is higher than the relevant segmentation methods, and thus the introduced segmentation method can provide some guidance in clinical diagnosis for liver tumor.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.