EAR-UNet: A deep learning-based approach for segmentation of tympanic membranes from otoscopic images
作者：
Highlights：
• This study presents a deep learning based-approach for fully automatic image segmentation of tympanic membranes.
• The proposed EAR-Unet composes of three main paradigms: EfficientNet for encoder; Attention gate for skip connection; ResNet for decoder.
• A new loss function relying on shape distance between prediction and reference is proposed.
• The stochastic weight averaging is exploited to avoid being trapped in local minima for network training.
• The proposed approach achieves better performances than state-of-the-arts.
摘要
•This study presents a deep learning based-approach for fully automatic image segmentation of tympanic membranes.•The proposed EAR-Unet composes of three main paradigms: EfficientNet for encoder; Attention gate for skip connection; ResNet for decoder.•A new loss function relying on shape distance between prediction and reference is proposed.•The stochastic weight averaging is exploited to avoid being trapped in local minima for network training.•The proposed approach achieves better performances than state-of-the-arts.
论文关键词：Tympanic membrane segmentation,Unet,Attention gate,Efficientnet,Resnet
论文评审过程：Received 15 August 2020, Revised 2 April 2021, Accepted 5 April 2021, Available online 8 April 2021, Version of Record 18 April 2021.
论文官网地址：https://doi.org/10.1016/j.artmed.2021.102065
