A Deep Neural Network Method for Arterial Blood Flow Profile Reconstruction

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Fluid Dynamics

Background:

• Arterial stenosis reduces blood flow, leading to cardiovascular diseases.
• Current diagnostic methods are costly and complex, hindering early prediction.
• Electromagnetic effects of arterial blood flow offer a potential diagnostic avenue.

Purpose of the Study:

• To develop a deep neural network-based method for accurate arterial blood flow profile reconstruction.
• To enable early prediction and monitoring of arterial stenosis and related cardiovascular diseases.
• To overcome limitations of existing diagnostic techniques in terms of cost and complexity.

Main Methods:

• Utilized a deep neural network, incorporating a convolutional auto-encoder (CAE) and a convolutional neural network (CNN).
• Input data included potential difference and weight matrix derived from arterial blood flow models.
• Simulations performed using COMSOL on carotid artery models with varying stenosis rates in a uniform magnetic field.

Main Results:

• Achieved an average root mean square error of 0.0333 and an average correlation coefficient of 0.9721.
• Demonstrated superior performance compared to Tikhonov, back propagation (BP), and standard CNN methods.
• High accuracy in reconstructing blood flow velocity distribution was confirmed through simulation.

Conclusions:

• The proposed deep neural network method accurately reconstructs arterial blood flow profiles.
• This technique holds significant potential for the early diagnosis of arterial stenosis.
• Offers a promising, high-accuracy, and potentially cost-effective approach for cardiovascular disease monitoring.