Puncture site decision method for venipuncture robot based on near-infrared vision and multiobjective optimization

Area of Science:

• Biomedical Engineering
• Robotics
• Artificial Intelligence

Background:

• Venipuncture robots offer enhanced precision but lack autonomous decision-making for puncture site selection.
• Current limitations hinder the widespread adoption of robotic venipuncture systems.

Purpose of the Study:

• To develop a multi-information fusion method for determining optimal venipuncture sites, thereby increasing robot autonomy.
• To enable robots to make independent decisions regarding puncture site selection for venipuncture procedures.

Main Methods:

• A U-Net with a soft attention mechanism (SAU-Net) was trained on a forearm vein image dataset for accurate vein segmentation.
• Near-infrared vision was used to extract vein features, including depth, diameter, curvature, and length.
• A multiobjective optimization model was employed to determine the optimal puncture site based on extracted vein characteristics.

Main Results:

• The SAU-Net achieved a vein segmentation accuracy of 91.2% and a vein extraction rate of 86.7%.
• The multiobjective optimization model provided a Pareto solution set with an average decision time of 1.458 seconds.
• Real-time vein segmentation and puncture site determination were demonstrated on a venipuncture robot using a near-infrared camera.

Conclusions:

• The proposed method significantly enhances the autonomy of venipuncture robots by enabling intelligent puncture site selection.
• This approach holds the potential to dramatically improve the implementation and effectiveness of robotic venipuncture in clinical settings.