Medical Image Region Extraction Using Self-Correcting Feedback Loops

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current medical image processing technologies face challenges in collecting high-quality correct answer data for machine learning, which is essential for accurate image processing, particularly in extracting and measuring specific regions within medical images.

Innovation Solution

An image processing apparatus and method that includes a first extractor for extracting measurement target regions using learned correct answer data, a measurement object determination unit for defining measurement objects, a measurement object correction unit for user-driven corrections, and a measurement target region correction unit for refining the extraction results, allowing for the use of corrected data as high-quality correct answer data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual collection of correct answer data is performed for machine learning, then high-quality training data can be obtained, but the workload and time required for selecting and storing correct answer images increases significantly

Engineering Contradiction:

Improvequality of correct answer dataVSAvoidtime for selecting and storing correct answer images

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system allows users to self-correct extraction results directly in the UI without requiring manual data collection workflows. Users can interactively adjust extracted regions and the system automatically stores these corrections as new training data, enabling self-service data generation that eliminates time-consuming manual selection and storage processes

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system implements a feedback loop where user corrections to extraction results are automatically captured and fed back into the machine learning training process. This creates a continuous improvement cycle where each user interaction generates high-quality training data that enhances future extraction accuracy, resolving the contradiction between data quality and collection time

Inventive Principle:

Principle #23Feedback

2Manufacturing precision

If interactive correction functionality is added to allow users to correct extraction results, then the quality of extracted measurement target regions improves, but the device complexity increases

Engineering Contradiction:

Improveaccuracy of region extractionVSAvoidcomplexity of correction system

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system introduces a measurement object determination unit as an intermediary that identifies key reference points (such as anatomical landmarks) to guide the correction process. This intermediary component simplifies user interaction by providing structured correction options based on medical knowledge, improving extraction accuracy without requiring complex free-form editing tools

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The correction functionality is implemented with local quality adjustments rather than global reprocessing. Users can correct specific portions of extracted regions while maintaining the accuracy of other areas, and the system applies corrections locally to minimize computational complexity while improving overall extraction precision

Inventive Principle:

Principle #3Local quality

3Measurement precision

If corrected extraction results are used as new correct answer data for retraining, then machine learning accuracy improves, but the processing time and computational resources increase

Engineering Contradiction:

Improveaccuracy of machine learning extractionVSAvoidtime for learning with corrected data

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary validation and preparation of corrected data before retraining, ensuring that only high-quality corrections are incorporated into the training set. This preliminary filtering action reduces the volume of data requiring retraining while maintaining accuracy improvements, thereby reducing the computational time and resources needed

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements continuous incremental retraining rather than complete retraining from scratch. Corrected data is continuously integrated into the training process in small batches, maintaining the model's useful actions while gradually improving accuracy. This continuous approach reduces computational overhead compared to periodic full retraining cycles

Inventive Principle:

Principle #20Continuity of useful action

Data Source

PatentUS11244458B2Image processing apparatus, image processing method, and program

Publication Date: 2022.02.08 FUJIFILM CORP

AI summary

Provided are an image processing apparatus, an image processing method, and a program that can collect high-quality correct answer data used for machine learning with a simple method. The image processing apparatus includes: a first extractor that extracts a measurement target region from a medical image, using a result of learning performed using correct answer data of the measurement target region; a measurement object determination unit that determines a measurement object used to measure the measurement target region; a measurement object correction unit that corrects the measurement object in response to a command from a user; and a measurement target region correction unit that corrects the measurement target region extracted by the first extractor, using a correction result of the measurement object. The first extractor performs learning using the measurement target region corrected by the measurement target region correction unit as correct answer data.