Medical Image Learning Data Creation via Reference Region Contact
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in efficiently creating learning data for machine learning models that recognize the size of target regions in medical images, particularly due to the inconvenience of using measure forceps and the time-consuming process of manually creating answer data.
Innovation Solution
A learning data creation apparatus that acquires medical images, detects target and reference regions, determines if they are in contact, measures the size of the target region based on the reference region, and stores the image and size data as learning data, thereby automating the creation of learning data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If measure forceps are inserted into the forceps port to measure the size of the target region, then the size measurement can be performed, but the operation becomes inconvenient and time-consuming
Solution Approach 1:
The patent replaces the mechanical measurement method (inserting measure forceps into the forceps port and manually reading gradations) with an automated image processing system. The learning model automatically detects the target region and reference region in the endoscopic image, calculates their contact relationship, and determines the target region size without any manual mechanical intervention.
Solution Approach 2:
The system enables self-service measurement by allowing the learning model to autonomously perform the entire measurement process. The model automatically identifies the target region, reference region, and their contact points, then calculates the size based on the known reference region dimensions, eliminating the need for operator intervention.
2Quantity of substance
If manual methods are used to create answer data for machine learning, then learning data can be created, but significant time and effort are required
Solution Approach 1:
The patent performs preliminary action by automatically creating learning data through the learning model during the normal diagnostic process. Instead of manually creating answer data after acquiring medical images, the system automatically generates labeled learning data (including target region size measurements) that can be stored and used for training the learning model further.
Solution Approach 2:
The patent replaces the manual labor-intensive process of creating learning data with an automated computational system. The learning model automatically processes medical images, detects regions, calculates measurements, and generates labeled learning data without human intervention, dramatically reducing the time and effort required for data creation.
3Ease of operation
If artificial objects with known sizes are used to generate gradations, then size measurement can be achieved without measure forceps, but the artificial object must be captured in the endoscopic image
Solution Approach 1:
The patent enhances universality by making the learning model adaptable to various measurement scenarios. The model can handle different types of reference regions (artificial objects with known sizes) and automatically detect their presence in the image. When a reference region is detected, the model calculates the target region size based on the contact relationship, making the system versatile for different measurement situations without requiring specific manual configuration.
Data Source
AI summary
A learning data creation apparatus, a method, a program, and a medical image recognition apparatus are provided. The learning data creation apparatus includes a first processor, and a memory that stores learning data for machine learning. The first processor acquires a first medical image from a modality, detects each of a target region and a reference region from the acquired first medical image, determines whether or not the detected target region and the reference region are in contact, measures a size of the target region based on a size of the reference region in a case where a contact is determined, and stores, in the memory, as the learning data, a pair of the first medical image including the target region of which the size is measured, and the measured size of the target region.