Tissue Image Ranking Model for Low-Burden Morphology Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The burden of annotation work for creating training data for machine learning models to evaluate morphological abnormalities in tissue images is high due to the qualitative and quantitative demands, especially when classifying and identifying various types of abnormalities with varying determination criteria and large numbers of images.
Innovation Solution
An image processing apparatus and method that uses a machine learning model trained with tissue image sets labeled with relative ranks based on morphological characteristics, allowing for the estimation of evaluation values that conform to these ranks, reducing the burden of annotation work.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is created by classifying and identifying abnormality levels with a unified scale, then the machine learning model can estimate evaluation values accurately, but the qualitative burden of annotation work becomes extremely high
Solution Approach 1:
The patent segments the annotation task into two parts: (1) annotating only the presence/absence of morphological abnormalities using predetermined criteria, and (2) allowing the machine learning model to automatically learn the evaluation value mapping from these simplified annotations. This segmentation reduces the qualitative burden while maintaining estimation accuracy through the model's learning capability.
Solution Approach 2:
The machine learning model performs self-service by automatically learning the relationship between morphological characteristics and evaluation values from the annotated training data. Instead of requiring experts to manually define unified classification criteria for all abnormality levels, the model autonomously discovers the mapping patterns during training, reducing dependency on expert annotation effort.
2Quantity of substance
If a large number of tissue images are annotated with detailed morphological abnormality classifications, then the training data quantity is sufficient, but the quantitative burden of annotation work becomes enormous
Solution Approach 1:
The patent applies partial action by annotating only the essential feature (presence/absence of morphological abnormalities) rather than providing complete detailed classifications for all images. The machine learning model compensates for this partial annotation by learning from the patterns in the annotated data and generalizing to produce accurate evaluation values for the entire training set, thus reducing the quantitative annotation burden.
3Measurement precision
If various types of morphological abnormalities are classified with unified determination criteria, then the annotation accuracy and uniformity improve, but the difficulty of classification and identification increases due to diverse abnormality types and varying criteria
Solution Approach 1:
The patent establishes predetermined criteria for determining the presence/absence of morphological abnormalities before the annotation process. These predetermined criteria serve as a preliminary framework that guides annotators, ensuring consistency and accuracy without requiring them to develop complex classification systems during annotation. The machine learning model then learns from these consistently annotated data.
Data Source
AI summary
An image processing apparatus includes a processor, in which an evaluation value in accordance with morphological characteristics of a tissue image is estimated using a machine learning model, the machine learning model is trained using a plurality of tissue image sets each including a plurality of training tissue images and provided with a relative rank based on the morphological characteristics of the training tissue image, the training is training in which the machine learning model is caused to estimate the evaluation value in accordance with the morphological characteristics of the training tissue image and the estimated evaluation value is made to conform to the relative rank, and the processor estimates the evaluation value of an acquired evaluation target using a trained machine learning model that has undergone the training.