Improving SLAM-Based Navigation in Flexible Ureteroscopy by Kidney Stone and Surgical Tool Segmentation

Area of Science:

• Medical Engineering
• Computer Vision
• Surgical Navigation

Background:

• Flexible ureteroscopy is key for kidney stone treatment.
• Accurate ureteroscope tracking is vital for complete stone extraction.
• Existing visual SLAM systems struggle with dynamic surgical environments.

Purpose of the Study:

• To improve the robustness and accuracy of visual SLAM for flexible ureteroscopy.
• To adapt SLAM algorithms for dynamic intraoperative conditions, including moving stones and instruments.
• To integrate real-time object segmentation into surgical navigation systems.

Main Methods:

• Utilized YOLOv7-E6E with Segment Anything (SAM) and YOLO11m-seg for real-time segmentation of kidney stones and surgical tools.
• Modified the SLAM framework to treat surgical instruments as dynamic elements to be discarded.
• Classified kidney stone pixels as potentially dynamic map points, allowing for their disappearance within the SLAM framework.
• Evaluated the approach using recorded datasets for segmentation and ureteroscope pose estimation.

Main Results:

• Achieved an average improvement of 35.4% in ureteroscope pose estimation using YOLOv7-E6E with SAM.
• Demonstrated a significant average improvement of 52.49% in ureteroscope pose estimation using YOLO11m-seg.
• The refined SLAM approach showed enhanced accuracy and robustness in dynamic surgical scenarios.

Conclusions:

• Real-time segmentation of dynamic elements significantly improves visual SLAM performance in flexible ureteroscopy.
• The proposed method enhances ureteroscope navigation accuracy, crucial for complete kidney stone removal.
• This advancement offers a more reliable navigation tool for surgeons performing ureteroscopic procedures.