Granted Patent B2

US 12,299,172 · App. 17/755,463 · Granted May 13, 2025

Inventors: Venkatesh NarasimhaMurthy (Hillsborough, NJ); Vivek Singh (Princeton, NJ); Yao-Jen Chang (Princeton, NJ); Benjamin S. Pollack (Jersey City, NJ); Ankur Kapoor (Plainsboro, NJ); Rayal Raj Prasad Nalam Venkat (Princeton, NJ)

Assignee: Siemens Healthcare Diagnostics Inc.

G06F21/6254G06T5/77G06T7/0004G06T11/60G06T2207/20081G06T2207/20084

Abstract

Claims (34)

1. A method of characterizing a specimen in an automated diagnostic analysis system, comprising:

capturing an image of a specimen container with an image capture device;

identifying a label affixed to the specimen container in the image with an anonymization network, wherein the anonymization network is a pre-trained network trained, with training images, to identify labels on specimen containers; and

editing the image with the anonymization network to mask the label such that information present in the label is removed from the image to produce an edited image, wherein information present in the label is removed by altering pixel data within the image.

2. The method of claim 1 , wherein the anonymization network is part of a quality check module.

3. The method of claim 1 , further comprising outputting the edited image for a segmentation or an HILN determination of the specimen in the specimen container with an HILN network of a quality check module.

4. The method of claim 1 , wherein the anonymization network comprises a general adversarial network.

5. The method of claim 1 , wherein the anonymization network comprises a variational auto encoder.

6. The method of claim 1 , wherein the anonymization network is configured to inpaint the label in the image of the specimen and the specimen container in order to erase all barcodes and printed text on the label and form an inpainted label in the edited image.

7. The method of claim 6 , wherein the inpainted label appears as a white label in the edited image.

8. The method of claim 1 , wherein a segmentation convolutional neural network (SCNN) receives as input one or more edited images from the anonymization network.

9. The method of claim 1 , further comprising training the anonymization network in a training phase prior to any characterization.

10. The method of claim 1 , wherein the editing of the image with the anonymization network preserves imaged fluid characteristics of the specimen in the edited image.

11. The method of claim 1 , wherein captured images containing labels with patient information are not permanently stored in order to protect patient information.

12. The method of claim 1 , wherein the information is patient information.

13. The method of claim 1 , wherein the editing of the image with the anonymization network removes all information present on the label.

14. The method of claim 1 , further comprising capturing images of the specimen container from other viewpoints.

15. The method of claim 1 , wherein the anonymization network is configured to inpaint the label in the image in order to mask all barcodes on the label.

16. The method of claim 1 , wherein the anonymization network is configured to inpaint the label in the image in order to mask all printed text on the label.

17. A method of characterizing a specimen container in an automated diagnostic analysis system, comprising:

capturing an image of the specimen container, the image including fluid characteristics of a specimen within the specimen container;

identifying in the image a label affixed to the specimen container;

editing the image with an anonymization network to mask the label such that some or all printed information present on the label is removed from the image to form an edited image, wherein information present in the label is removed by altering pixel data representing the label and wherein pixel data representing imaged fluid characteristics of the specimen is preserved;

storing the edited image; and

employing a characterization network to access the stored image and characterize at least one imaged fluid characteristic of the specimen, wherein the characterization network is trained to identify one or more fluid characteristics of the specimen within the image.

18. A quality check module of an automated diagnostic analysis system, comprising:

a plurality of image capture devices arranged around an imaging location configured to capture multiple images from multiple viewpoints of a specimen container; and

a computer coupled to the plurality of image capture devices, the computer configured and operative via programming instructions to:

input a captured image taken by one of the plurality of image capture devices to an anonymization network executing on the computer, the captured image depicting at least the specimen container and a label affixed to the specimen container,

identify in the captured image the label affixed to the specimen container via the anonymization network, wherein the anonymization network is a pre-trained network trained, with training images, to identify labels on specimen containers; and

edit the captured image to produce an edited image via the anonymization network to mask the identified label such that information present in the label is removed from the captured image, wherein information present in the label is removed by altering pixel data within the image.

19. The quality check module of claim 18 , wherein the computer is further configured and operative via the programming instructions to:

output the edited image for segmentation or an interferent determination of a specimen in the specimen container via an HILN network executing on the computer.

20. The quality check module of claim 19 , wherein the computer is further configured and operative via the programming instructions to display the edited image with the information on the label being removed from the captured image.

Assignments (3)

Continuity (2)

References Cited (59)