Deep learning volume quantifying methods and apparatus

Patent metadata

Field	Value
Publication number	US-11657593-B2
Application number	US-201816634541-A
Country	US
Kind code	B2
Filing date	Jul 25, 2018
Priority date	Jul 28, 2017
Publication date	May 23, 2023
Grant date	May 23, 2023

Abstract

A neural network-based method for quantifying a volume of a specimen. The method includes providing a specimen, capturing images of the specimen, and directly classifying to one of a plurality of volume classes or volumes using a trained neural network. Quality check modules and specimen testing apparatus adapted to carry out the volume quantification method are described, as are other aspects.

First claim

What is claimed is:

1. A method of quantifying a specimen, comprising:
providing a specimen container at an imaging location, the specimen container containing the specimen;
capturing multiple high dynamic range images of the specimen and storing the multiple high dynamic range images as image data;
processing the image data with a neural network, the neural network comprising a plurality of convolution and pooling layer pairs; and
classifying the specimen with the neural network into one of:
a plurality of volume classes, and
a volume of the specimen.

2. The method of claim 1 wherein the capturing multiple high dynamic range images comprises capturing a series of images of the specimen at multiple spectra having different nominal wavelengths.

3. The method of claim 2 wherein the series of images comprises multiple different exposures at each of the different nominal wavelengths.

4. The method of claim 3 comprising selecting optimally-exposed pixels from the multiple high dynamic range images at the multiple different exposures and at each of the multiple spectra to generate optimally-exposed image data for each of the multiple spectra.

5. The method of claim 1 , wherein the capturing multiple high dynamic range images is conducted from multiple different viewpoints, with an image capture device provided at each viewpoint.

6. The method of claim 5 , wherein the number of the plurality of volume classes comprises between 10 and 1,000.

7. The method of claim 1 , wherein the neural network comprises a convolutional neural network having a fully-connected layer following the plurality of convolution and pooling layer pairs.

8. The method of claim 1 , wherein the neural network is trained from multiple training sets, wherein each training set of the multiple training sets comprises a training image and a scalar volume annotation for each of the plurality of volume classes.

9. The method of claim 8 , wherein the scalar volume annotation is a volume or of one or more of a serum or plasma portion, a settled blood portion, and a gel separator.

10. The method of claim 1 , wherein the one of the plurality of volume classes is mapped to a volume by a look-up table.

11. The method of claim 10 , determining a ratio of a volume of serum or plasma portion to a volume of a settled blood portion from determined volumes.

12. The method of claim 1 , wherein the neural network is configured and trained to determine a volume class of a serum or plasma portion, a settled blood portion, or a gel separator.

13. The method of claim 1 , wherein the neural network is configured and trained to determine a volume class of a serum or plasma portion.

14. The method of claim 1 , wherein the neural network is configured and trained to determine a volume class of a settled blood portion.

15. The method of claim 1 , comprising:
identifying a volume of the serum or plasma portion that is contained in the specimen container.

16. The method of claim 1 , comprising:
identifying a volume of the settled blood portion that is contained in the specimen container.

17. The method of claim 1 , comprising:
identifying a volume of a gel separator that is contained in the specimen container.

18. The method of claim 1 , wherein the neural network comprises a pool layer, a convolution layer, and a soft max layer.

19. A quality check module, comprising:
one or more image capture devices configured to capture images of the specimen at multiple spectra having different nominal wavelengths, at multiple exposures, and from one or more viewpoints; and
a computer operatively coupled to the one or more image capture devices and operable to:
select optimally-exposed pixels from the images at the different exposures and at each of the multiple spectra and generate optimally-exposed image data for each of the multiple spectra,
process the image data with a neural network, the neural network comprising a plurality of convolution and pooling layer pairs; and
classify the specimen with the neural network into one of:
a plurality of volume classes, and
a volume of the specimen.

20. A specimen testing apparatus, comprising:
a track; and
a quality check module on the track, the quality check module including:
one or more image capture devices configured to capture images of the specimen at multiple spectra having different nominal wavelengths, at multiple different exposures, and from one or more viewpoints, and
a computer configured and operable to:
select optimally-exposed pixels from the images at the multiple different exposures at each of the multiple spectra to generate optimally-exposed image data for each of the multiple spectra and viewpoint,
process the image data with a neural network, the neural network comprising a plurality of convolution and pooling layer pairs; and
classify the specimen with the neural network into one of:
a plurality of volume classes, and
a volume of the specimen.

Assignees

• Siemens Healthcare Diagnostics Inc

Inventors

• Kluckner Stefan
• Chang Yao-Jen
• Ma Kai
• Singh Vivek
• Chen Terrence
• Pollack Benjamin S

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/454 Primary

  Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN] · CPC title

• G01N35/04

  Details of the conveyor system {(G01N35/021 - G01N35/028 take precedence)} · CPC title

• G06V10/82

  using neural networks · CPC title

• G01N35/00613

  Quality control · CPC title