CE-Net: Context Encoder Network for 2D Medical Image Segmentation.

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computer Vision

Background:

• Medical image segmentation is crucial for medical image analysis.
• Deep learning, particularly convolutional neural networks (CNNs), is increasingly applied to medical image segmentation tasks.
• Existing U-Net based methods can suffer from spatial information loss due to pooling and strided convolutions.

Purpose of the Study:

• To propose a novel Context Encoder Network (CE-Net) for 2D medical image segmentation.
• To capture high-level contextual information while preserving spatial details.
• To improve upon existing segmentation methods like U-Net.

Main Methods:

• Developed CE-Net with three main components: feature encoder, context extractor, and feature decoder.
• Utilized a pretrained ResNet block as a fixed feature extractor.
• The context extractor incorporates a dense atrous convolution block and a residual multi-kernel pooling block.

Main Results:

• CE-Net was applied to diverse 2D medical image segmentation tasks, including optic disc segmentation, vessel detection, lung segmentation, cell contour segmentation, and retinal optical coherence tomography layer segmentation.
• Demonstrated superior performance compared to the original U-Net and other state-of-the-art methods.
• Showcased effective capture of high-level information and preservation of spatial details.

Conclusions:

• The proposed CE-Net effectively addresses the spatial information loss issue in U-Net architectures.
• CE-Net offers improved accuracy and robustness for various 2D medical image segmentation applications.
• This deep learning model represents a significant advancement in medical image analysis.