Granted Patent B2

US 12,243,644 · App. 17/868,605 · Granted Mar 4, 2025

Inventors: Thomas Engel (Aalen, DE); Gaby Marquardt (Hausen, DE); Jens-Peter Brock (Langenzenn, DE)

Assignee: Siemens Healthcare Diagnostics Inc.

G16H50/20G06T7/0012G16H30/40G06T2207/20081G06T2207/30024

Abstract

Claims (47)

1. A medical diagnosis assistance system, comprising:

an input interface configured to receive medical image data of a patient;

a computing device configured to implement:

a classification module configured to classify parts of interest, POI, comprising objects of interest, OOI, or regions of interest, ROI, within the received medical image data, and to assign a corresponding reliability metric to each of the classified POI; wherein:

the classification module is configured to implement a classifying artificial intelligence entity, CAIE, which is trained and configured to receive at least a portion of the received medical image data and to generate, based thereon, a CAIE output classifying the POI or the corresponding reliability metric,

the CAIE comprises a first-level classifying artificial intelligence sub-entity, FLCAISE, and a second-level artificial intelligence sub-entity, SLCAISE,

the FLCAISE is trained and configured to receive, as its input, at least a portion of the received medical image data and to generate, based thereon, a FLCAISE output classifying at least a part of the POI according to a broad classification scheme; and

the SLCAISE is trained and configured to receive, as its input, at least a portion of the received medical image data and the FLCAISE output, and to further classify at least the part of the POI according to a refined classification scheme, and

an analysis module configured to determine, based on the POI and the assigned reliability metric, an analysis of the medical image data; and

an output interface configured to output an output signal indicating the analysis.

2. The medical diagnosis assistance system of claim 1 , wherein the classification module comprises a cropping sub-module, CSM, configured to receive the FLCAISE output and to generate excerpts from the received medical image data, each excerpt associated with a single POI and comprising at least the corresponding associated POI, and to provide the generated excerpts to the SLCAISE as the input of the SLCAISE.

3. The medical diagnosis assistance system according to claim 1 , wherein the analysis module is configured to implement an analyzing artificial intelligence entity, AAIE, which is trained and configured to receive, as its input, the classified POI and the assigned reliability metric and to generate, based thereon, the analysis.

4. The medical diagnosis assistance system according to claim 1 ,

wherein the classification module and the analysis module are configured to perform at least one refinement loop comprising:

generating, by the analysis module, a preliminary analysis;

receiving, by the classification module, the preliminary analysis by the analysis module and re-classifying the POI or re-assigning the corresponding reliability metric based on the preliminary analysis, and

determining, based on the re-classified POI or the re-assigned corresponding reliability metric, another preliminary analysis; and

wherein after the last refinement loop the preliminary analysis is used as the analysis of the analyzing module.

5. The medical diagnosis assistance system according to claim 1 , wherein the input interface is configured to receive medical image data comprising at least 20 parts of interest, POI, to be classified, and wherein the classification module is configured to classify at least 20 of the at least 20 parts of interest, POI, comprised in the medical image data.

6. The medical diagnosis assistance system according to claim 1 , wherein the medical image data are magnified visual images of blood samples, and wherein the parts of interest, POI, comprise blood components within the blood samples as objects of interest, OOI.

7. A computer-implemented medical diagnosis assistance method, comprising:

receiving medical image data of a patient;

classifying parts of interest, POI, comprising objects of interest, OOI, or regions of interest, ROI, within the received medical image data via a classifying artificial intelligence entity, CAIE, which is trained and configured to receive at least a portion of the received medical image data and to generate, based thereon, a CAIE output classifying the POI or the corresponding reliability metric, wherein:

the CAIE comprises a first-level classifying artificial intelligence sub-entity, FLCAISE, and a second-level artificial intelligence sub-entity, SLCAISE,

the FLCAISE is trained and configured to receive, as its input, at least a portion of the received medical image data and to generate, based thereon, a FLCAISE output classifying at least a part of the POI according to a broad classification scheme; and

the SLCAISE is trained and configured to receive, as its input, at least a portion of the received medical image data and the FLCAISE output, and to further classify at least the part of the POI according to a refined classification scheme,

assigning a corresponding reliability metric to each of the classified POI; and

determining, based on the POI and the assigned reliability metric, an analysis of the medical image data.

8. The method of claim 7 , further comprising steps of:

generating, using the FLCAISE, a FLCAISE output classifying at least a part of the POI according to a broad classification scheme, based on at least a portion of the received medical image data; and

generating, using the SLCAISE, a SLCAISE output classifying the at least part of the POI according to a refined classification scheme, based on at least a portion of the received medical image data and on the FLCAISE output.

9. The method of claim 8 , further comprising steps of:

generating excerpts from the received medical image data, each excerpt associated with a single POI and comprising at least the corresponding associated POI, and

providing the generated excerpts to the SLCAISE as the input of the SLCAISE.

10. A computer-implemented training method for training a classifying artificial intelligence entity, CAIE, to receive at least a portion of received medical image data and to generate, based thereon, a CAIE output classifying parts of interest, POI, comprising objects of interest, OOI, or regions of interest, ROI, within the received medical image data, or a corresponding reliability metric, the method comprising steps of:

providing first labelled samples of medical image data;

providing a first-level classifying artificial intelligence sub-entity, FLCAISE, in a preliminary state;

performing supervised learning, using the first labelled samples, in order to train the first-level classifying artificial intelligence sub-entity, FLCAISE, to generate a FLCAISE output classifying parts of interest, POI, within an input of the first-level classifying artificial intelligence sub-entity, FLCAISE, according to a broad classification scheme;

providing second labelled samples of medical image data;

providing a second-level artificial intelligence sub-entity, SLCAISE, in a preliminary state;

performing supervised learning, using the second labelled samples and the FLCAISE output, in order to train the second-level artificial intelligence sub-entity, SLCAISE, to further classify parts of interest, POI, within an input of the second-level classifying artificial intelligence sub-entity, SLCAISE, according to a refined classification scheme; and

providing a trained classifying artificial intelligence entity, CAIE, comprising both the trained first-level classifying artificial intelligence sub-entity, FLCAISE, and the trained second-level artificial intelligence sub-entity, SLCAISE.

11. The method of claim 10 , wherein the second labelled samples comprise less parts of interest, POI, than the first labelled samples.

12. The method of claim 10 ,

wherein the first labelled samples comprise magnified visual images of blood samples wherein blood components, as objects of interest, OOI, are labelled according to blood component types, and

wherein the second labelled samples comprise magnified visual images of blood samples wherein blood components, as objects of interest, OOI, are labelled according to blood component sub-types.

13. The method of claim 10 , wherein the first labelled samples or the second labelled samples comprise magnified visual images of undyed or dyed blood samples.

Assignments (4)

Priority Claims (1)

Continuity (1)

References Cited (24)