Real-time AI-guided ultrasound localization method for breast tumor rotational resection

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Surgical Technology

Background:

• Breast tumors pose a global health challenge, with vacuum-assisted biopsy systems (VABB) facing limitations in primary care due to surgeon dependency and positioning difficulties.
• Current AI solutions for breast tumor analysis often lack real-time capabilities required for surgical navigation.

Purpose of the Study:

• To develop and validate a real-time positioning system for enhanced breast tumor rotational resection using an optimized AI model.
• To improve surgical navigation accuracy and facilitate the adoption of minimally invasive techniques in primary healthcare.

Main Methods:

• An optimized YOLOv11n architecture was developed, integrating MobileNetV4 blocks to enhance efficiency and reduce computational load.
• The model was trained and validated on ultrasound video data from 167 patients, focusing on backbone, neck, and detection head optimization.
• Two clinical positioning algorithms were developed to support diverse surgical workflows within a real-time visualization system.

Main Results:

• The optimized YOLOv11n+ model demonstrated a 17.1% reduction in parameters and 27.0% decrease in FLOPS compared to the baseline.
• Mean Average Precision (mAP50) for cutter slot and tumor detection improved by 2.1%.
• The system provides millisecond-level tracking, precise annotation, and intelligent prompts for optimal resection timing.

Conclusions:

• The developed AI system offers technical support for minimally invasive breast tumor resection.
• Findings suggest the potential to reduce reliance on surgeon experience, promoting wider use in primary healthcare institutions.