Unified Training Data Collection for Endoscopic Image Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Collecting vast amounts of training data for endoscopic image analysis is labor-intensive, requiring significant effort to acquire and associate endoscopic images with corresponding diagnosis information for AI model training.
Innovation Solution
A training data collection apparatus and method that displays endoscopic images and diagnosis information on a device, allowing users to select and associate images with diagnosis information using tabs, and record the data for model training, facilitating efficient data grouping and selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a vast amount of training data is collected manually from multiple systems, then the quality and quantity of training data improve, but the time and labor required for data collection increases significantly
Solution Approach 1:
The patent merges endoscopic images and diagnosis information that were stored in separate systems (endoscopy management system, endoscopy filing system) into a unified training data set. The data collection apparatus integrates multiple data sources and associates images with diagnosis information through a single interface, eliminating the need to manually collect from separate systems and significantly reducing data collection time while maintaining comprehensive training data quantity
Solution Approach 2:
The patent introduces a data collection apparatus as an intermediary between existing endoscopy systems and the AI training process. This apparatus includes a display control unit that presents endoscopic images and diagnosis information in a unified interface, allowing automatic association and recording of training data. The intermediary tool streamlines the data extraction process from multiple sources without requiring manual intervention for each data point
2Quantity of substance
If endoscopic images and diagnosis information are collected from multiple separate systems, then comprehensive training data is obtained, but the complexity of data acquisition and format conversion increases
Solution Approach 1:
The patent creates a universal data collection apparatus that can access and process endoscopic images and diagnosis information from multiple different systems (endoscopy management system, endoscopy filing system) through a single unified interface. The apparatus performs multiple functions including displaying images, presenting diagnosis information, associating data pairs, and recording training data, thereby reducing acquisition complexity while maintaining comprehensive training data coverage
3Measurement precision
If manual selection and association of endoscopic images with diagnosis information is performed, then data accuracy is maintained, but the workload and time required for data preparation increases
Solution Approach 1:
The patent implements a self-service data association mechanism where the display control unit presents endoscopic images and diagnosis information in a structured format, and the data collection apparatus automatically records associated pairs once the user selects them. This automated recording process eliminates manual data entry and format conversion tasks, maintaining data association accuracy while significantly improving data preparation efficiency by reducing repetitive manual work
Data Source
AI summary
Provided are a training data collection apparatus, a training data collection method, and a program that easily collect training data for use in training of a model that discriminates a lesion from an endoscopic image, a training system, a trained model, and an endoscopic image processing apparatus. Endoscopic images are associated with respective pieces of findings-diagnosis information. When the endoscopic images and the respective piece of findings-diagnosis information that are associated with each other are recorded in a recording apparatus as training data for use in training of a model that discriminates a lesion from an endoscopic image, the endoscopic images associated with the respective pieces of findings-diagnosis information are classified into corresponding groups of the respective pieces of findings-diagnosis information. The endoscopic images are displayed and switched on a group-by-group basis by using tabs.


