Method for detecting a binding of antibodies from a patient sample to double-stranded DNA using Crithidia luciliae cells and fluorescence microscopy

Patent metadata

Field	Value
Publication number	US-12372532-B2
Application number	US-202117478666-A
Country	US
Kind code	B2
Filing date	Sep 17, 2021
Priority date	Sep 17, 2020
Publication date	Jul 29, 2025
Grant date	Jul 29, 2025

Abstract

There is proposed a method for detecting a binding of autoantibodies from a patient sample to double-stranded deoxyribonucleic acid using Crithidia luciliae cells by means of fluorescence microscopy, including the steps of: provision of a substrate which has multiple Crithidia luciliae cells, incubation of the substrate with the patient sample which potentially has the autoantibodies, incubation of the substrate with secondary antibodies which have each been labelled with a preferably green fluorescent dye, acquisition of a fluorescence image of the substrate, identification by means of a first pretrained convolutional neural network of respective sub-images in the one fluorescence image that each represent a Crithidia luciliae cell, furthermore respective processing of at least one subset of the respective sub-images by means of a second pretrained convolutional neural network for determining respective binding measures which indicate a respective extent of a binding of autoantibodies in a respective kinetoplast region of a respective Crithidia luciliae cell of a respective sub-image, and determination of an overall binding measure with regard to a binding of autoantibodies from the patient sample to double-stranded deoxyribonucleic acid on the basis of the respective binding measures.

First claim

What is claimed is:

1. A method for detecting a binding of autoantibodies from a patient sample to double-stranded deoxyribonucleic acid using Crithidia luciliae cells by means of fluorescence microscopy and by means of digital image processing, the method comprising:
provision of a substrate which has multiple Crithidia luciliae cells;
incubation of the substrate with the patient sample which potentially has the autoantibodies;
incubation of the substrate with secondary antibodies which have each been labelled with a green fluorescent dye;
acquisition of a fluorescence image of the substrate in a green color channel which corresponds to the fluorescent dye;
identification by means of a first pretrained convolutional neural network of respective sub-images in the one fluorescence image that each represent a Crithidia luciliae cell;
respective processing of at least one subset of the respective sub-images by means of a second pretrained convolutional neural network for determining respective binding measures which indicate a respective extent of a binding of autoantibodies in a respective kinetoplast region of a respective Crithidia luciliae cell of a respective sub-image; and
determination of an overall binding measure with regard to the binding of autoantibodies from the patient sample to double-stranded deoxyribonucleic acid on the basis of the respective binding measures.

2. The method according to claim 1 , wherein:
the identification of the respective sub-images in the one fluorescence image is effected by assigning respective image segments of the fluorescence image to respective image segment classes from a group of image segment classes by means of the first pretrained convolutional neural network; and
the group of the image segment classes comprises at least the following image segment classes: cell and background.

3. The method according to claim 1 , wherein:
the identification of the respective sub-images in the one fluorescence image is effected by assigning respective image segments of the fluorescence image to respective image segment classes from a group of image segment classes by means of the first pretrained convolutional neural network; and
the group of the image segment classes comprises at least the following image segment classes: cell body, cell edge and background.

4. The method according to claim 1 further comprising:
for a respective sub-image:
selection of a respective subordinate image of the respective sub-image,
wherein the respective subordinate image represents a respective kinetoplast region of a respective Crithidia luciliae cell; and

determination of the respective binding measure on the basis of the respective subordinate image; and

determination of the overall binding measure on the basis of the respective binding measures.

5. The method according to claim 1 further comprising:
determination of a respective final feature map for a respective sub-image by means of the second convolutional neural network;
determination of a respective confidence measure with regard to a presence of a binding of autoantibodies in a respective kinetoplast region for the respective sub-image;
selection of a subset of the sub-images on the basis of the determined confidence measures;
respective processing of the respective feature maps of the respective selected sub-images for determining the respective binding measures; and
determination of the overall binding measure on the basis of the respective binding measures of the respective selected sub-images.

6. The method according to claim 5 further comprising:
for a respective sub-image from the selected subset:
selection of a respective subordinate image of the respective sub-image on the basis of a respective final feature map corresponding to the respective sub-image,
wherein the respective subordinate image represents a respective kinetoplast region of a respective Crithidia luciliae cell; and

determination of the respective binding measure on the basis of the respective subordinate image; and

determination of the overall binding measure on the basis of the respective binding measures.

7. The method according to claim 6 further comprising:
for a respective sub-image from the selected subset:
ascertainment of a respective masking operator on the basis of the respective final feature map;
selection of the respective subordinate image of the respective sub-image by means of application of the respective masking operator to the respective sub-image; and
determination of the respective binding measure on the basis of the respective subordinate image; and

determination of the overall binding measure on the basis of the respective binding measures.

8. The method according to claim 5 , wherein, in the course of a processing of a sub-image, the second convolutional neural network:
in a first processing level, generates a first set of resultant feature maps on the basis of the sub-image by means of at least one first convolutional layer and by means of application of multiple two-dimensional convolution kernels; and
in a second processing level:
generates a second set of resultant feature maps on the basis of the first set of two-dimensional feature maps by means of at least one second convolutional layer and by means of application of multiple three-dimensional convolution kernels; and

generates a third set of resultant feature maps on the basis of the second set of two-dimensional feature maps by means of at least one third convolutional layer and by means of application of multiple three-dimensional convolution kernels,
wherein the second set has a smaller number of resultant feature maps than the first set, and
wherein the third set has a larger number of resultant feature maps than the second set.

9. The method according to claim 8 , wherein:
in the second processing level:
the second convolutional layer and the third convolutional layer are in a sequence as sub-steps of a sequential processing path; and
there is in parallel to the sequential processing path a further processing path in which the second convolutional neural network generates a fourth set of resultant feature maps on the basis of the first set of two-dimensional feature maps by means of at least one fourth convolutional layer;

the second convolutional neural network generates, on the basis of the third and the fourth set of resultant feature maps, the final feature map corresponding to the sub-image; and
the number of successive convolution layers in the parallel processing path is smaller than the number of successive convolution layers in the sequential processing path.

10. The method according to claim 1 , further comprising:
acquisition of a first preliminary fluorescence image in the color channel using a predefined acquisition parameter;
establishment of whether a brightness of the first preliminary fluorescence image of the color channel exceeds a maximum brightness;
in the event of the first preliminary fluorescence image of the color channel not exceeding the maximum brightness, use of the first preliminary fluorescence image as the one fluorescence image; and
in the event of the first preliminary fluorescence image of the color channel exceeding the maximum brightness, acquisition of a second preliminary fluorescence image in the color channel and use of the second preliminary fluorescence image of the color channel as the one fluorescence image.

11. A device for detecting a binding of autoantibodies from a patient sample to double-stranded deoxyribonucleic acid using Crithidia luciliae cells by means of fluorescence microscopy and by means of digital image processing,

Assignees

• Euroimmun Medizinische Labordiagnostika Ag

Inventors

• Gerlach Stefan
• Krauth Jens
• Marzahl Christian
• Krause Christopher
• Danckwardt Maick
• Hahn Melanie
• Voigt Jörn

Classifications

• G06T2207/30024

  Cell structures in vitro; Tissue sections in vitro · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/20021

  Dividing image into blocks, subimages or windows · CPC title

• G06T2207/10064

  Fluorescence image · CPC title