An effective AI model for automatically detecting arteriovenous fistula stenosis

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Medical Acoustics

Background:

• Arteriovenous fistulas (AVFs) are crucial for hemodialysis access.
• Stenosis is a common complication that can lead to AVF failure.
• Current detection methods can be invasive or costly.

Purpose of the Study:

• To develop a novel, non-invasive AI model for detecting AVF stenosis.
• To utilize audio recordings as an inexpensive data source.
• To achieve high diagnostic performance for stenosis detection.

Main Methods:

• A hybrid AI model combining short-time Fourier transform (STFT) and sample entropy features.
• Integration with ResNet50 and Artificial Neural Network (ANN) classifiers.
• Hyper-parameter optimization using Design of Experiments (DOE).

Main Results:

• The AI model achieved high performance across all key metrics (sensitivity, specificity, accuracy, precision, F1-score > 0.90).
• Effective detection of stenosis greater than 50% was demonstrated.
• The model showed robust performance in identifying AVF stenosis.

Conclusions:

• The proposed AI model offers a promising, non-invasive, and cost-effective approach for AVF stenosis detection.
• The use of audio recordings and AI can significantly improve AVF monitoring.
• This technology has potential applications in home-care settings for early detection and management.