Coronary angiography image segmentation based on PSPNet.

Area of Science:

• Medical Imaging
• Computer Vision
• Deep Learning

Background:

• Coronary artery disease (CAD) presents a significant global health challenge due to high prevalence, disability, and mortality.
• Cardiovascular disease incidence and mortality are steadily increasing worldwide, necessitating advanced diagnostic tools.

Purpose of the Study:

• To develop an efficient image processing method for accurate vascular structure extraction from coronary angiography images.
• To combine computer vision and deep learning techniques for improved analysis of vascular images.

Main Methods:

• Coronary angiography image segmentation was performed using the PSPNet network, with comparisons to FCN.
• A parallel multi-scale convolutional neural network model based on PSPNet was implemented, incorporating small sample transfer learning.
• Performance was evaluated using precision, recall, and F1-score metrics.

Main Results:

• The proposed technique achieved an accuracy of 0.957.
• PSPNet demonstrated a 26.75% higher accuracy than traditional algorithms and 4.59% higher than U-Net.
• Transfer learning with PSPNet improved average segmentation accuracy to 0.936, sensitivity to 0.865, and specificity to 0.949, closely matching human expert segmentation.

Conclusions:

• The PSPNet network enhances diagnostic efficiency by reducing manual interaction and reliance on medical personnel.
• This approach provides valuable auxiliary strategies for subsequent medical diagnosis systems utilizing cardiac coronary angiography.