Computer Vision assisted Collision Avoidance for Radiation Therapy

Project Overview

Collision avoidance is a critical safety concern in radiation therapy, where the linear accelerator gantry and other treatment components move around the patient. Our research focuses on developing an advanced computer vision system that creates patient-specific 3D models to prevent collisions during treatment.

The Challenge

Traditional collision avoidance systems rely on fixed safety margins and simplified patient models, which can be overly restrictive and may not account for patient-specific anatomy or positioning changes during treatment. This can lead to suboptimal treatment plans or unnecessary treatment interruptions.

Our Approach

We are developing a novel system that combines:

• Computer vision algorithms to capture the patient's external contour
• A deformable skin-mesh model that creates a patient-specific 'avatar'
• Real-time tracking of patient positioning during treatment
• Predictive collision detection that simulates gantry and couch movements before they occur

Expected Outcomes

Our collision avoidance system aims to:

• Enhance patient safety by preventing treatment machine collisions
• Enable more optimal treatment plans by reducing overly conservative safety margins
• Improve treatment workflow efficiency by reducing false collision warnings
• Provide real-time monitoring of patient position to detect potential safety issues

Current Status

This project has completed initial validation in a clinical environment and is now being implemented in selected treatment rooms. We are collecting data on system performance and refining the algorithms based on clinical feedback.