Automatic calculation method for stenosis ratio based on dialysis access ultrasound image segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Vascular Surgery

Background:

• Dialysis Access (DA) stenosis affects hemodialysis efficiency and patient health, requiring early detection via ultrasound.
• Current ultrasound assessment relies heavily on operator expertise, straining resources in large facilities.
• Automated analysis of DA ultrasound images is crucial for accurate stenosis ratio calculation, improving diagnostics and treatment.

Purpose of the Study:

• To employ image segmentation networks for precise segmentation of DA lumens in ultrasound images.
• To automatically classify stenosis types and calculate stenosis ratios using morphological processing.
• To enhance physician diagnostic efficiency and provide quantitative data for clinical decision-making.

Main Methods:

• Developed Vessel Lumen Segmentation and Classification-Net (VLSC-Net), a deep neural network for DA lumen segmentation and classification.
• Compared VLSC-Net against U-Net, TransUNet, MultiResUnet, and ResUNet using mIoU, Dice, Accuracy, HD, and ASSD metrics.
• Utilized morphological processing and feature extraction for lumen edge delineation and automatic calculation of LDSR and SASR.

Main Results:

• VLSC-Net achieved superior performance with mIoU of 0.9563, Dice score of 0.9777, and Accuracy of 0.9976.
• Demonstrated significant improvement over U-Net (p < 0.0125) with an average processing time of 164 ms per image.
• Reported average errors of 1.4% for LDSR and 7.8% for SASR in stenosis ratio calculations.

Conclusions:

• The VLSC-Net approach significantly enhances diagnostic efficiency for medical personnel.
• Provides reliable, objective evidence for clinical assessment and decision-making in DA stenosis treatment.
• Aims to reduce the risk of complications associated with Dialysis Access stenosis.