Fully-channel regional attention network for disease-location recognition with tongue images

作者：

Highlights：

• A new research task of disease-location recognition based on tongue image is researched, the first large-scale clinical tongue image dataset with the diagnostic label is established.

• A novel inner-imaging channel-wise attention mechanism is proposed, it reduces the redundancy of the CNNs model and improves its modeling efficiency.

• A dynamic regional pooling mechanism is proposed. Several specific areas are intercepted from one feature map to form multiple concentrated signals and remove edge noise.

摘要

•A new research task of disease-location recognition based on tongue image is researched, the first large-scale clinical tongue image dataset with the diagnostic label is established.•A novel inner-imaging channel-wise attention mechanism is proposed, it reduces the redundancy of the CNNs model and improves its modeling efficiency.•A dynamic regional pooling mechanism is proposed. Several specific areas are intercepted from one feature map to form multiple concentrated signals and remove edge noise.

论文关键词：Tongue image modeling,Disease-position recognition,Convolutional networks,Regional detailed features,Attention mechanism

论文评审过程：Received 6 May 2020, Revised 6 April 2021, Accepted 11 May 2021, Available online 26 May 2021, Version of Record 1 June 2021.

论文官网地址：https://doi.org/10.1016/j.artmed.2021.102110