Introduction:
Imagine achieving very high accuracy (maybe sub-millimeter precision) in tracking bone movement, improving medical procedures like robotic knee replacements and wearable exoskeletons. Traditional methods like CT scans, skin markers, or implantable devices come with different challenges. Use A-mode ultrasound—a technology that, when combined with deep-learning, holds the potential to surpass the limitations of current methods. In this MS thesis project, you will work with an A-mode ultrasound dataset from a cadaver knee movement. Your goal is to develop a high-accuracy deep-learning solution to track bone movements in real-time.
Project Objectives:
- Study the entire process of bone movement tracking using A-mode ultrasound.
- Design a robust deep-learning framework that automatically detects bone positions with high accuracy.
- Achieve real-time bone registration and tracking (visualization), with a target of the sub-millimeter precision.
Research Questions:
- How can A-mode ultrasound be leveraged for accurate, real-time bone tracking (in cadaver and in-vivo)? What are the current limitations, and how can we overcome them?
- What deep-learning structures can be designed to improve detection accuracy for bone positions?
- How can post-processing and registration techniques enhance the system's overall precision?
- How will you evaluate the performance of both the deep-learning model and the final bone tracking? What kinematic results should be reported?
Expected Outcomes:
- A full understanding of the A-mode ultrasound bone tracking pipeline.
- A new deep-learning architecture designed specifically for high-accuracy bone position detection.
- Advanced bone registration processes with continuous accuracy improvements.
- Comprehensive performance metrics and visual results that showcase the superiority of your system.
Mr. Bangyu Lan will be the daily supervisor for this thesis work.