SegUnXt+: A High-Performance Deep Learning Model for Thyroid Segmentation in Fully Automatic 3D USE Robotic...

Area of Science:

• Medical Imaging
• Oncology
• Robotics

Background:

• Thyroid cancer incidence is rising, particularly in women.
• Early detection is crucial to reduce overdiagnosis and enable less invasive treatments.
• Current ultrasound (US) imaging is limited by observer variability.

Purpose of the Study:

• To develop an advanced 3D imaging system for thyroid cancer diagnosis.
• To overcome the limitations of traditional ultrasound by integrating robotics and AI.
• To improve the accuracy and accessibility of thyroid cancer screening.

Main Methods:

• A novel 3D imaging system combining a US machine, robotic arm, depth camera, and custom software.
• Integration of automated robotic assistance for enhanced precision and repeatability.
• 3D reconstruction of ultrasound elastography (USE) and brightness mode (USB) with a deep learning model (SegUnXt+) for segmentation.

Main Results:

• The SegUnXt+ model achieved competitive thyroid gland segmentation (IoU 82.8%, DC 90.6%) and thyroid nodule segmentation (IoU 71.9%, DC 83.3%).
• The system enables flexible multi-view observation and quantitative stiffness analysis.
• Demonstrated potential to minimize observer dependency and enhance diagnostic accuracy.

Conclusions:

• The developed 3D imaging system offers precise 3D thyroid visualization.
• Automated, quantitative analysis of elastographic and morphologic data supports early thyroid cancer detection.
• The system has clinical relevance for mass screening and improving diagnostic accuracy and accessibility.