Endoscope Control Using Machine Learning for Fold Avoidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing endoscope control technologies struggle to properly operate the distal end of the insertion portion, particularly when encountering structures like folds and intestinal walls, leading to potential contact issues.
Innovation Solution
An endoscope control apparatus that includes an image acquisition unit, an operation detail determination unit, and a control unit. The operation detail determination unit uses machine learning-based operation selection models to determine appropriate operation details from endoscopic images, and the control unit controls the endoscope's movement based on these determined details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the distal end of the insertion portion is controlled to face the center of the lumen, then the lumen can be imaged, but the distal end cannot properly operate when folds or intestinal walls shield portions of the lumen
Solution Approach 1:
The bending angle of the bending portion is dynamically adjusted based on real-time endoscopic images. The control unit changes the bending angle in response to detected folds or intestinal walls, allowing the distal end to adapt its orientation and navigate around obstacles rather than maintaining a fixed posture aimed at the lumen center.
Solution Approach 2:
The system uses feedback from endoscopic images to control the bending portion. The control unit receives image data, detects obstacles such as folds or intestinal walls, and adjusts the bending angle accordingly. This closed-loop control enables the distal end to respond to environmental conditions and maintain proper operation capability.
2Measurement precision
If the bending angle is controlled to image the lumen center, then imaging is improved, but the distal end moves toward contacting folds without avoiding them
Solution Approach 1:
The control unit continuously monitors endoscopic images and detects folds or intestinal walls in real-time. When obstacles are detected, the system adjusts the bending angle to change the direction of the distal end, preventing contact with folds while maintaining the ability to image the lumen center when clear.
Solution Approach 2:
The system takes preliminary action by detecting folds or intestinal walls before the distal end can contact them. The control unit adjusts the bending angle in advance to redirect the distal end away from potential contact points, preventing harmful interactions before they occur.
3Device complexity
If the distal end moves directly toward the lumen center, then imaging is simplified, but the distal end cannot go around intestinal walls to access hidden lumens
Solution Approach 1:
The bending angle is dynamically adjusted based on real-time detection of intestinal walls and lumens. The control unit changes the bending angle in response to detected structures, allowing the distal end to navigate around intestinal walls and access hidden lumens rather than following a fixed simple path.
Solution Approach 2:
The system changes the bending angle parameter in response to detected anatomical structures. When intestinal walls or hidden lumens are detected, the control unit modifies the bending angle to enable the distal end to navigate around obstacles and access different lumens, providing navigation flexibility while maintaining relatively simple control.
Data Source
AI summary
An image acquisition unit acquires an endoscopic image imaged by an endoscope inserted inside a subject. An operation detail determination unit determines one or more operation details from among a predetermined plural number of operation details based on the endoscopic image acquired in the image acquisition unit. An operation control unit controls a movement of the endoscope based on the determined operation details. An operation detail determination unit determines one or more operation details by inputting input data acquired from the endoscopic image acquired in the image acquisition unit to one or more into operation selection models generated by machine learning using, as training data, an image for learning, which is an endoscopic image imaged in the past, and a label indicating an operation detail for an endoscope that has imaged the image for learning.