Neural Network for Distinguishing Anatomical Structures in Surgery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-assisted systems struggle to accurately detect and distinguish critical anatomical structures in dynamic and unpredictable surgical environments, such as laparoscopic cholecystectomy, due to challenges in processing extended fields of view and handling occluded parts.
Innovation Solution
A computer-implemented method using a neural network with two configurations to detect and identify anatomical structures in a video of a laparoscopic surgical procedure. The first configuration detects a plurality of structures, while the second configuration, using temporal models and weak labels, identifies specific types of anatomical structures, such as the cystic artery and cystic duct, and generates an augmented video with annotations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If computer-assisted systems process extended field of view in surgical procedures, then the information available to surgeons is improved, but the ability to accurately detect and distinguish critical structures deteriorates due to environmental variability and occlusions
Solution Approach 1:
The system segments the complex surgical environment into multiple processing streams: one for detecting all structures in the extended field of view, and another for specifically identifying critical anatomical structures. This segmentation allows the system to handle the large volume of visual information while maintaining precision in identifying critical elements like vessels and ducts, even when occluded or in variable lighting conditions.
Solution Approach 2:
The system transitions from 2D video feed analysis to 3D spatial understanding by implementing depth estimation and spatial relationship analysis. This dimensional enhancement allows the system to distinguish critical structures from background elements more effectively, resolving the contradiction between processing extended field of view and maintaining detection accuracy in variable surgical environments.
2Device complexity
If the neural network uses a single configuration to detect all structures, then the system complexity is reduced, but the ability to distinguish specific types of anatomical structures deteriorates
Solution Approach 1:
The system implements dynamic configuration switching of the neural network based on the surgical phase and detected structures. The network adapts its parameters and focus during the procedure, transitioning from a general detection mode to a specialized identification mode for critical structures. This dynamic adjustment allows the system to maintain high identification accuracy without requiring permanently complex multi-configuration systems.
Solution Approach 2:
The system performs preliminary detection of all structures in the surgical field before focusing on specific anatomical identification. This two-stage approach allows the simpler initial detection phase to guide the subsequent more complex identification phase, reducing overall system complexity while maintaining high accuracy in distinguishing specific anatomical structures like cystic arteries and common bile ducts.
3Loss of information
If the system annotates all detected structures, then the completeness of information is improved, but the visual clutter and difficulty in identifying critical structures deteriorates
Solution Approach 1:
The system applies different annotation qualities and levels of detail to different types of structures based on their clinical significance. Critical anatomical structures receive prominent, distinct annotations with higher visual weight, while less critical structures receive simpler or no annotations. This local differentiation maintains information completeness while ensuring that critical structures stand out clearly to the surgeon, improving visual interpretation ease.
Solution Approach 2:
The system extracts and highlights only the most critical anatomical structures for prominent display, separating them from the complete set of detected structures. This extraction approach maintains full information availability in the system's internal representation while presenting a simplified, clutter-free view to the surgeon that focuses attention on critical elements needing identification or avoidance.
Data Source
AI summary
Technical solutions are provided to facilitate computer assistance during a surgery to prevent complications by detecting, identifying, and highlighting specific anatomical structures in a video of the surgery using machine learning. According to some aspects, a computer vision system is trained to detect several structures in the video of the surgery, and further to distinguish between the structures despite their similar appearance.