FlowVM-Net: Enhanced Vessel Segmentation in X-Ray Coronary Angiography Using Temporal Information Fusion

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Accurate segmentation of coronary vessels in X-ray coronary angiography (XCA) is crucial for diagnosing and treating coronary artery disease.
• Motion artifacts and shadowing in XCA images present significant challenges to precise vessel segmentation.

Purpose of the Study:

• To develop an advanced deep learning model, FlowVM-Net, for enhanced segmentation of coronary vessels in XCA image sequences.
• To address limitations in current methods by incorporating dynamic temporal information and robust feature fusion.

Main Methods:

• Proposed FlowVM-Net, a novel encoder-decoder architecture integrating an optical flow module for temporal information.
• Introduced a wavelet dilated convolution visual state space model block (VMamba) and an attention-based optical flow feature fusion module.
• Employed a composite loss function, including boundary difference over union loss, to improve edge and thin vessel segmentation accuracy.

Main Results:

• FlowVM-Net achieved a Dice Similarity Coefficient (DSC) of 85.17%, sensitivity of 85.15%, and a quality score of 90.49% on a dataset of 542 samples.
• The model demonstrated effectiveness in preserving vessel continuity and accurately segmenting thin vessels.

Conclusions:

• FlowVM-Net successfully leverages dynamic context from XCA sequences for improved coronary artery segmentation.
• The proposed method offers a promising approach for enhancing the diagnosis and treatment of coronary artery disease through precise vessel segmentation.