Convolution Pyramid Network: A Classification Network on Coronary Artery Angiogram Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Convolutional Neural Networks (CNNs) have advanced natural image classification using single feature maps.
• Coronary artery angiograms present challenges for CNNs due to significant photographing noise and small inter-class variations.
• Existing methods struggle to achieve optimal classification performance on angiographic data.

Purpose of the Study:

• To develop a novel network architecture for improved feature representation in coronary artery angiogram classification.
• To enhance the richness and relevance of features during the training process for better diagnostic accuracy.
• To create a network with strong generalization capabilities for diverse angiogram classification tasks.

Main Methods:

• Proposed a new network utilizing multiple convolutions with varying kernel sizes.
• Focused on enhancing feature extraction to address noise and subtle class differences in angiograms.
• Evaluated the network's performance against state-of-the-art image classification models.

Main Results:

• The proposed network demonstrated significantly improved feature representation for angiogram classification.
• Achieved a 30.5% increase in classification recall and a 19.1% increase in precision in optimal scenarios.
• Showcased superior performance compared to existing state-of-the-art image classification networks on angiographic data.

Conclusions:

• The novel network architecture effectively enhances feature richness and relevance for coronary artery angiogram classification.
• The proposed method offers a robust solution for overcoming challenges posed by noise and small class gaps in angiograms.
• The network exhibits strong generalization, outperforming current leading models in recall and precision for angiogram analysis.