Oral Endoscopic Image Labeling via Multi-Expert Consensus
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high rate of misdiagnosis and delayed detection of oral cancer due to insufficient specialists skilled in reading endoscopic images necessitates a more precise method for analyzing oral endoscopic images, which requires reliable training data for machine learning algorithms.
Innovation Solution
A method and apparatus for generating training data by acquiring oral endoscopic images and clinical information, providing them to multiple medical staffs for reading, selecting target images based on consensus classification results, and determining labels for these images, ensuring that only images with consistent classification results from multiple specialists are used for training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are designed to improve oral cancer diagnosis accuracy, then diagnostic precision can be improved, but reliable training data becomes insufficient
Solution Approach 1:
The patent merges the diagnostic judgments of multiple medical staff members by collecting reading results from several specialists and using consensus-based labeling. This combination approach creates more reliable training labels by aggregating expert opinions, thereby resolving the contradiction between improving diagnostic precision through machine learning and ensuring the reliability of training data.
Solution Approach 2:
The patent implements a feedback mechanism where reading results from multiple medical staff are collected, analyzed for consistency, and used to determine final labels. The system provides feedback loops including consistency checking, dispute resolution processes, and iterative label refinement, which enhances training data reliability while supporting improved diagnostic precision through machine learning.
2Reliability
If multiple medical staff are involved in reading results to improve label accuracy, then training data quality can be improved, but processing time increases
Solution Approach 1:
The patent applies preliminary action by having multiple medical staff provide reading results in advance before the final label determination process. This allows the system to collect all necessary expert opinions beforehand, enabling efficient batch processing and consensus calculation without extending the overall processing time excessively, thus balancing label accuracy with processing efficiency.
Solution Approach 2:
The patent uses partial action by selectively involving multiple medical staff only when needed for consensus building, rather than requiring all possible experts for every case. The system determines the appropriate number of readers based on case complexity and uses excessive action only when initial readings show high disagreement, thereby improving label accuracy while controlling processing time.
Data Source
AI summary
Disclosed are an apparatus and method for generating training data. An apparatus for generating training data according to an embodiment includes a reading result acquirer that acquires one or more oral endoscopic images and clinical information related to each of the one or more oral endoscopic images, provides the one or more oral endoscopic images and the clinical information to a user terminal of each of a plurality of preset medical staffs, and receives a reading result for each of the one or more oral endoscopic images from the user terminal of each of the plurality of medical staffs, and a labeling unit that selects one or more labeling target images from among the one or more oral endoscopic images based on the reading result received from the user terminal of each of the plurality of medical staffs and determines one or more labels for each of the one or more labeling target images based on the reading result.