Synthetic Medical Image Artifact Detection With Confidence Maps

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Medical images generated by machine learning models may contain errors, making it difficult for physicians to distinguish between real features and artifacts, which can lead to incorrect diagnoses or therapies.

Innovation Solution

A method and system for generating synthetic medical images that includes generating multiple sub-images from an original image, determining color value dispersion, and calculating a confidence value based on this dispersion to assess the trustworthiness of the synthetic image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If machine learning models generate synthetic medical images to accelerate examinations and reduce contrast agent usage, then productivity and resource efficiency are improved, but the images may contain artifacts that reduce reliability

Engineering Contradiction:
Improveexamination speedVSAvoidimage accuracy
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The system generates multiple synthetic images from the same input data and uses them to create confidence maps that provide feedback on the reliability of each image region. This feedback mechanism allows physicians to identify and disregard artifact-containing regions while maintaining the productivity benefits of synthetic image generation.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

Instead of treating synthetic images as uniform entities, the system applies local quality assessment by generating confidence maps that vary across different regions of the image. Each region receives a reliability score based on consistency with other synthetic images, allowing localized trustworthiness evaluation.

Inventive Principle:
Principle #3Local quality

2Loss of time

If physicians use synthetic medical images for diagnosis, then examination time is reduced, but the presence of undetectable artifacts increases the risk of incorrect diagnosis

Engineering Contradiction:
Improvediagnosis timeVSAvoiddiagnosis error risk
Core Design Contradiction:
Loss of timeVSObject-affected harmful factors

Solution Approach 1:

The system performs preliminary quality assessment by generating confidence maps before the physician makes a diagnosis. This preliminary action identifies reliable and unreliable regions in advance, allowing physicians to focus their attention on high-confidence areas and avoid potential diagnostic errors from artifacts.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The confidence map serves as an intermediary between the synthetic image and the physician's diagnostic decision. It provides an additional layer of information that mediates the interpretation of synthetic image features, helping physicians distinguish between real anatomical structures and generation artifacts.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If multiple synthetic images are generated to assess reliability, then measurement precision is improved, but device complexity increases

Engineering Contradiction:
Improveconfidence assessment accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments the reliability assessment task by generating multiple synthetic images and comparing them region-by-region to create confidence maps. This segmentation approach breaks down the complex reliability assessment into manageable comparisons between individual synthetic images, improving measurement precision through multiple samples.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4471710B1Detection of artifacts in synthetic medical records
Publication Date: 2025.12.17 BAYER AG
  • EP4471710B1 patent drawingFigure 1~2
  • EP4471710B1 patent drawingFigure 3
  • EP4471710B1 patent drawingFigure 4

AI summary

The present disclosure relates to the technical field of generating synthetic medical images. The subject matter of this disclosure includes a method, a computer system, and a computer-readable storage medium comprising a computer program for detecting artifacts in synthetic medical images.