Multi-Receptive-Field CNN for Semantic Segmentation of Medical Images

Area of Science:

• Medical Image Analysis
• Computer Vision
• Deep Learning

Background:

• Context-based Convolutional Neural Networks (CNNs) are vital for semantic segmentation in medical imaging.
• Extracting comprehensive contextual information from complex medical images remains a significant challenge.
• Existing methods struggle with the variability and intricacy of medical image data.

Purpose of the Study:

• To introduce a novel Multi-Receptive-Field CNN (MRFNet) for improved medical image segmentation.
• To address the limitations in extracting rich contextual information from medical images.
• To enhance the performance and robustness of semantic segmentation in the medical domain.

Main Methods:

• Proposed a novel Multi-Receptive-Field CNN (MRFNet) architecture.
• Implemented an encoder-decoder module (EDM) with optimal receptive fields for subnets.
• Generated and fused multi-receptive-field context information at the feature map level using concatenation.

Main Results:

• MRFNet demonstrated outstanding performance on three public medical image datasets: SISS, 3DIRCADb, and SPES.
• The proposed method achieved superior results compared to other segmentation techniques.
• MRFNet outperformed existing methods on the 3DIRCADb dataset without requiring model pre-training.

Conclusions:

• MRFNet effectively extracts multi-receptive-field context information, significantly improving medical image segmentation.
• The proposed architecture offers a robust solution for complex and variable medical imaging data.
• MRFNet represents a significant advancement in deep learning for medical image analysis.