Depth-Map Obstacle Avoidance for Surgical Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Surgical navigation systems face challenges in accurately tracking flexible anatomical structures and managing the complexity and cost associated with attaching trackers to multiple objects in the surgical workspace.
Innovation Solution
A navigation system that combines tracker-based localization with machine vision to generate depth maps, allowing for the identification of obstacles and controlling robotic manipulators to avoid them, using a localizer and vision device to create expected and actual depth maps for precise surgical navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If trackers are attached to each object adjacent the target volume, then tracking accuracy is improved, but workspace congestion increases and system complexity increases
Solution Approach 1:
The patent combines multiple tracking functions into a single camera-based vision system. Instead of using separate trackers on each object, the system uses monocular or stereo vision to detect and track multiple objects (surgical instruments, anatomical structures, implants) simultaneously through image processing and machine learning algorithms, thereby reducing system complexity while maintaining tracking accuracy.
Solution Approach 2:
The vision system serves multiple functions: it tracks surgical instruments, identifies anatomical landmarks, monitors implant positions, and detects obstacles all through a single camera system. This multi-functional approach eliminates the need for specialized trackers for each object type, reducing both workspace congestion and system complexity.
2Measurement precision
If trackers are attached to each object adjacent the target volume, then tracking coverage is improved, but cost increases
Solution Approach 1:
The system creates virtual models (digital twins) of physical objects through computer vision. Instead of requiring physical trackers on each object, the vision system captures images and generates corresponding virtual representations that can be tracked and manipulated in the surgical navigation software, providing the same tracking coverage at lower cost.
Solution Approach 2:
The patent replaces mechanical tracker systems with an optical vision-based system. By using camera imaging and computational algorithms instead of physical electromagnetic or optical trackers attached to each object, the system achieves comprehensive tracking coverage while significantly reducing hardware costs and system complexity.
3Measurement precision
If trackers are attached to flexible anatomical structures, then tracking accuracy is improved, but the flexible nature of structures makes tracking difficult
Solution Approach 1:
The vision system automatically adapts to the changing positions of flexible anatomical structures by continuously analyzing sequential images. The machine learning algorithms automatically identify and track anatomical landmarks on flexible structures without requiring manual tracker attachment, compensating for the structures' natural movement and flexibility.
Solution Approach 2:
The system dynamically adjusts to the flexible nature of anatomical structures by using continuous video imaging and real-time image processing. Instead of relying on static trackers that become misaligned with moving flexible structures, the vision system continuously updates the positions of anatomical landmarks based on each new frame, maintaining tracking accuracy despite structural flexibility.
Data Source
AI summary
Systems and methods are described herein wherein a localizer is configured to detect a position of a first object and a vision device is configured to generate a depth map of surfaces near the first object. A virtual model corresponding to the first object is accessed, and a positional relationship between the localizer and the vision device in a common coordinate system is identified. An expected depth map of the vision device is then generated based on the detected position of the first object, the virtual model, and the positional relationship. A portion of the actual depth map that fails to match the expected depth map is identified, and a second object is recognize based on the identified portion.