Validation of AI-based result data

Patent metadata

Field	Value
Publication number	US-12412377-B2
Application number	US-202217888572-A
Country	US
Kind code	B2
Filing date	Aug 16, 2022
Priority date	Aug 20, 2021
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

In a method, comparison features are extracted from labeled reference image data. Features are also extracted from the image data. A statistical comparison of the comparison features with the features then takes place. On the basis of the statistical comparison and a quality criterion, the quality of the AI-based result data is determined. A method for correcting result data is additionally described. Furthermore, a method for AI-based acquisition of result data on the basis of measured examination data is described. Also described is a validation entity. An entity for correcting result data is additionally described. Furthermore, an entity for acquiring result data is described. Also described is a medical imaging entity.

First claim

What is claimed is:

1. A method for validating result data acquired via artificial intelligence (AI), the result data being based on image data, and the method comprising:
extracting comparison features from reference result data in a rule-based manner such that at least one of
the comparison features indicate at least one of a physical or anatomical characteristic of an examination object, or
the comparison features indicate a known characteristic present in an image of a given type, the result data being based on at least one image of the given type;

extracting features from the result data in the rule-based manner;
statistically comparing the comparison features with the features; and
determining a quality of the result data based on the statistical comparison and a quality threshold value.

2. The method as claimed in claim 1 , further comprising:
acquiring the result data.

3. The method as claimed in claim 1 , wherein the result data comprises:
image segmentation data;
detected objects;
classification data; or
regression data.

4. The method as claimed in claim 1 , wherein, in the event that the quality of the result data is below the quality threshold value, the method comprises:
outputting information indicating that a reliability of the result data is not sufficient, or
correcting the result data based on the comparison features and rules for the extracting of the comparison features.

5. The method as claimed in claim 1 , wherein
the comparison features include a reference image noise, and
the method further includes
comparing, as part of a procedure for reducing image noise of image data, the image noise of the image data with the reference image noise.

6. The method as claimed in claim 5 , wherein
the reference image noise is broken down into a number of frequency components, and
the statistical comparison of the comparison features with the features includes comparing frequency components of the reference image noise with frequency components of the image noise of the image data.

7. A method for correcting result data, the method comprising:
determining a validation result based on a method for validating result data as claimed in claim 1 ; and
generating corrected result data in the event that the validation result is negative.

8. The method for correcting result data as claimed in claim 7 , wherein the generating of the corrected result data comprises at least one of:
capturing substitute result data and replacing the result data with the substitute result data, or
correcting the result data directly based on the statistical comparison of the comparison features with the features, wherein the correcting includes
correcting an extracted feature such that the extracted feature becomes valid in a statistical comparison with the comparison features, and
generating corrected result data based on rules applied during the extracting of the comparison features.

9. A method for AI-based acquisition of result data based on measured examination data, the method comprising:
capturing the measured examination data from an examination object;
determining result data using AI based on the measured examination data;
determining a validation result based on a method as claimed in claim 1 ; and
determining the result data, which has been corrected based on the validation result.

10. A validation entity comprising:
at least one memory; and
processing circuitry configured to, with the at least one memory, cause the validation entity to
extract comparison features from labeled reference result data in a rule-based manner such that at least one of
the comparison features indicate at least one of a physical or anatomical characteristic of an examination object, or
the comparison features indicate a known characteristic present in an image of a given type, wherein result data is based on image data of at least one image of the given type,

extract features from the result data in the rule-based manner, the result data acquired via artificial intelligence (AI);
statistically compare the comparison features with the features; and
determine a quality of the result data based on the statistical comparison and a quality threshold value.

11. The validation entity as claimed in claim 10 , wherein the processing circuitry is further configured to cause the validation entity to:
determine a validation result based on the quality of the result data; and
generate corrected result data in the event that the validation result is negative.

12. The validation entity as claimed in claim 10 , wherein the processing circuitry is further configured to cause the validation entity to:
capture measured examination data from the examination object;
determine the result data based on the measured examination data;
determine a validation result based on the result data; and
generate corrected result data in the event that the validation result is negative.

13. The validation entity as claimed in claim 10 , wherein the processing circuitry is further configured to cause the validation entity to:
generate image data from an examination region;
determine the result data based on the image data; and
validate the result data.

14. A non-transitory computer program product including a computer program having program sections that, when executed by at least one processor at a medical imaging entity, cause the medical imaging entity to perform the method of claim 1 .

15. A non-transitory computer-readable medium storing program sections that, when executed by at least one processor of a computer unit, cause the computer unit to perform the method of claim 1 .

16. The method of claim 9 , wherein the measured examination data is medical image data.

17. A method for AI-based acquisition of result data based on measured examination data, the method comprising:
capturing the measured examination data from an examination object;
determining result data using AI based on the measured examination data;
determining a validation result for the result data; and
determining corrected result data based on the validation result, according to the method of claim 7 .

18. The validation entity as claimed in claim 11 , wherein the processing circuitry is further configured to cause the validation entity to:
generate image data from an examination region;
determine the result data based on the image data;
validate the result data; and
correct the result data.

19. The validation entity as claimed in claim 12 , wherein the processing circuitry is further configured to cause the validation entity to:
generate image data from an examination region;
determine the result data based on the image data; and
validate the result data.

20. The method as claimed in claim 4 , further comprising:
using, as part of a procedure for reducing image noise of image data, a reference image noise as a comparison feature; and
comparing the image noise of the image data with the reference image noise.

21. The method as claimed in claim 1 , wherein the comparison features indicate a distance between landmarks of the examination object.

Assignees

• Siemens Healthcare Gmbh
• Siemens Healthineers Ag

Inventors

• Fieselmann Andreas
• Huemmer Christian
• Biniazan Ramyar
• Ritschl Ludwig

Classifications

• G06T5/70

  Denoising; Smoothing · CPC title

• G06V10/751

  Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching · CPC title

• G06T5/50

  using two or more images, e.g. averaging or subtraction · CPC title

• G06V10/765

  using rules for classification or partitioning the feature space · CPC title

• G06V10/776 Primary

  Validation; Performance evaluation · CPC title