Endoscope System with Trained Model for Landmark Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In surgical procedures using endoscopes, it is challenging for operators to accurately identify objects whose position and shape are not clearly displayed in images, particularly when landmarks like the common bile duct, cystic duct, Rouviere's sulcus, and base of the S4 are obscured or have ambiguous boundaries.
Innovation Solution
An information processing system that uses a trained model, based on machine learning with annotated training data, to detect and display the position and shape of objects in real-time endoscopic images, allowing for enhanced visualization of critical landmarks during surgeries like cholecystectomy through laparoscopy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning is used to detect objects in endoscopic images, then detection accuracy of unclear landmarks is improved, but the complexity of the system increases
Solution Approach 1:
The system performs preliminary training of the machine learning model using annotated training data before actual detection. The training phase prepares the model in advance to recognize unclear landmarks, separating the complex learning process from the real-time detection process, thereby improving detection accuracy without adding complexity during surgical operations.
Solution Approach 2:
The patent introduces an intermediate processing layer between the endoscopic image capture and the detection output. This intermediate layer includes feature extraction and annotation processing that bridges the raw image data and the final detection results, managing the complexity through structured intermediate steps while improving measurement precision.
2Measurement precision
If annotated training data is used to train the model, then detection precision of obscured landmarks is improved, but the time and resources required for training increase
Solution Approach 1:
The system performs annotation and model training in advance before actual surgical detection is needed. By preparing the trained model beforehand with pre-annotated training data, the system achieves high detection precision for obscured landmarks without consuming time during critical surgical procedures.
Solution Approach 2:
The patent uses annotated training data that creates simplified representations or copies of the complex anatomical structures. These annotated copies serve as training examples that capture essential features without requiring the full complexity of original medical images, reducing training time while maintaining detection precision.
3Productivity
If the trained model processes detection images in real-time, then surgical efficiency is improved, but computational load increases
Solution Approach 1:
The detection process is segmented into distinct stages: feature extraction, model inference, and result processing. By dividing the computational task into smaller segments, the system achieves real-time detection capability with manageable computational load at each stage, improving surgical efficiency without overwhelming computational resources.
Solution Approach 2:
The system extracts only the essential features from detection images that are relevant for landmark identification, rather than processing the entire image data. This feature extraction approach reduces the computational load significantly while maintaining real-time processing capability and surgical efficiency.
Data Source
AI summary
An information processing system includes a storage device that stores therein a trained model, and a processor. The trained model is trained to output a position and shape of an object in a training image based on training data. The training data is data in which the training image is provided with an annotation indicating the position and shape of the object. The training image is an image captured with an angle of view including the object whose position and shape are not clearly displayed in an image. The processor executes detection processing on a detection image to output detected information indicating the position and shape of the object. The processor then causes a display device to display the detected information superimposed on the detection image.