A learnable Gabor Convolution kernel for vessel segmentation.

Area of Science:

• Medical image analysis
• Computer vision
• Biomedical engineering

Background:

• Vessel segmentation is crucial for diagnosing vascular diseases.
• Current methods using convolutional neural networks (CNNs) can be parameter-heavy due to limitations in predicting learning direction.
• Existing CNNs often require large channels or depth for adequate feature extraction, leading to redundancy.

Purpose of the Study:

• To develop an optimized Gabor convolution kernel for enhanced vessel segmentation.
• To integrate Gabor convolution kernels into a novel CNN architecture, Gabor ConvNet.
• To evaluate the performance of Gabor ConvNet against state-of-the-art methods on multiple vessel datasets.

Main Methods:

• Designed and optimized a Gabor convolution kernel with parameters updated via backpropagation.
• Integrated the Gabor convolution kernels into a CNN architecture, creating Gabor ConvNet.
• Tested Gabor ConvNet on three distinct vessel segmentation datasets.

Main Results:

• Gabor ConvNet achieved top rankings on three vessel datasets with scores of 85.06%, 70.52%, and 67.11%.
• The proposed method demonstrated superior performance compared to advanced existing models.
• Ablation studies confirmed the enhanced vessel extraction capability of the Gabor kernel over regular convolution kernels.

Conclusions:

• The Gabor convolution kernel offers a more efficient and effective approach to vessel segmentation.
• Gabor ConvNet represents a significant advancement in the field of medical image analysis for vascular characterization.
• The optimized Gabor kernel can be seamlessly integrated into various CNN architectures for improved performance.