CN-121999483-A - AI training and detection method for detecting nuclear erythrocyte sample

CN121999483ACN 121999483 ACN121999483 ACN 121999483ACN-121999483-A

Abstract

The AI training and detecting method for the nuclear erythrocyte-containing sample detection comprises the steps of preprocessing the nuclear erythrocyte-containing sample to obtain a detection sample, tiling the detection sample, suspending cells to sink, shooting the tiled detection sample to obtain a detection sample image, identifying and labeling the nuclear erythrocyte in the detection sample image to obtain a labeling picture, carrying out AI training by using the labeling picture to obtain an AI characteristic data set A, wherein the AI characteristic data set A comprises nuclear erythrocyte sample characteristics, and the AI characteristic data set A is combined with a corresponding AI algorithm and has the capability of identifying the nuclear erythrocyte. The method comprises the steps of identifying and detecting sample images by an AI identification algorithm, identifying nucleated red blood cells in samples in a selected area S1 of the images, and obtaining the total number NUM1 of the nucleated red blood cells in the samples in the selected images, wherein the AI identification algorithm identifies the nucleated red blood cells in the samples according to a nucleated red blood cell characteristic data set, and the nucleated red blood cell characteristic data set is obtained by training a labeled nucleated red blood cell picture.

Inventors

FANG XIANGFEI
LU JINGJIANG
WANG ZHIPING

Assignees

深圳安侣医学科技有限公司

Dates

Publication Date: 20260508
Application Date: 20241108

Claims (18)

1. An AI training method for detecting a sample containing nuclear red blood cells is characterized in that, Pretreating a sample containing nuclear red blood cells to obtain a detection sample; Spreading the detection sample, suspending the cells and sinking; shooting a tiled detection sample to obtain a detection sample image; Identifying and labeling nucleated red blood cells in the detection sample image to obtain a labeling picture, and performing AI training by using the labeling picture to obtain an AI characteristic data set A; the AI characteristic data set A comprises nucleated red blood cell sample characteristics; the AI characteristic data set A is combined with a corresponding AI algorithm and has the capability of identifying nucleated red blood cells.
2. The AI training method for detecting a sample containing nucleated red blood cells according to claim 1, wherein a mammalian non-nucleated red blood cell is added to the sample to obtain a detection sample; identifying and labeling the mammal non-nucleated red blood cells in the detection sample image to obtain a labeling picture, and performing AI training by using the labeling picture to obtain an AI characteristic data set A; The AI characteristic data set A comprises characteristics of mammal non-nucleated red blood cells, wherein the nucleated red blood cells are mammal nucleated red blood cells; The AI characteristic data set A is combined with a corresponding AI algorithm and has the capability of distinguishing the mammal non-nucleated red blood cells from the mammal nucleated red blood cells.
3. The AI training method for detecting a sample containing nucleated red blood cells according to claim 1, wherein white blood cells of a selected species are added to the sample to obtain a detected sample; identifying and labeling the white blood cells of the selected species in the detected sample image to obtain a labeling picture, and performing AI training by using the labeling picture to obtain an AI characteristic data set A; The AI feature dataset a includes leukocyte features of a selected species; the AI feature dataset a, in combination with the corresponding AI algorithm, has the ability to distinguish nucleated red blood cells from white blood cells of a selected species.
4. The AI training method for the detection of a sample containing nucleated red blood cells according to claim 1, wherein thrombocytes or thrombocyte clusters of a selected species are added to the sample to obtain a detection sample; Identifying and labeling thrombocytes or thrombocyte clusters of selected species in the detected sample image to obtain labeling pictures, and performing AI training by using the labeling pictures to obtain an AI characteristic data set A; AI feature dataset a includes thrombocyte or thrombocyte cluster features of a selected species; The AI feature dataset a, in combination with the corresponding AI algorithm, has the ability to distinguish nucleated red blood cells from thrombocytes or thrombocyte clusters of a selected species.
5. The AI training method for nuclear erythroid sample detection of claim 3 or 4, wherein the selected species comprises reptiles, fish, birds, amphibians, mammals.
6. The AI training method for nuclear erythrocyte-containing sample detection of claim 1, comprising any one of the following technical features: TA1, adding a coloring agent into a sample, and coloring the sample to obtain a detection sample; TA2, the nucleated red blood cell sample is from a reptile, and the label of the label picture is the reptile nucleated red blood cell; TA3, the nucleated red blood cell sample is from fish, and the labeling picture is labeled as the nucleated red blood cell of the fish; TA4, the nucleated red blood cell sample is from birds, and the labeling picture is labeled as the nucleated red blood cells of the birds; TA5, the nucleated red blood cell sample is from mammal, and the labeling picture is labeled as the nucleated red blood cell of mammal.
7. The AI training method for nuclear erythrocyte-containing sample detection of claim 2, wherein the pretreatment of the sample comprises any one of the following technical features: the sample of the non-nucleated red blood cells is from human beings, and the labeling picture is labeled as human non-nucleated red blood cells; the TB2 is derived from dogs, and the labeling picture is labeled as dogs' non-nucleated red blood cells; TB3, wherein the coreless red blood cell sample is from cats, and the labeling picture is labeled as cat coreless red blood cells; And TB4, wherein the non-nucleated red blood cell sample is from a mammal, and the labeling picture is labeled as the non-nucleated red blood cell of the mammal.
8. A method for detecting a sample containing nuclear red blood cells is characterized in that, Pretreating a sample to obtain a detection sample; Tiling a detection sample; shooting a tiled detection sample to obtain a detection sample image; Identifying the detected sample image by using an AI identification algorithm, and identifying nucleated red blood cells in the sample in the selected area S1 of the image to obtain the total number NUM1 of the nucleated red blood cells in the sample in the selected image; the nucleated red blood cell characteristic data set is obtained by training the marked nucleated red blood cell picture.
9. The method for detecting a sample of erythrocyte containing nuclei according to claim 8 wherein, The sample is derived from any one of blood of reptiles, fish, birds, amphibians, and mammals.
10. The method according to claim 8, wherein the selected area S1 corresponds to a volume of the sample to be tested of V1 and the sample to be tested has a nuclear erythrocyte unit volume content of NUM1/V1.
11. The method for detecting a sample of red blood cells containing nuclei according to claim 10, Identifying the detected sample image by using an AI identification algorithm, and identifying cells of a selected category in a sample in a selected area S1 of the image to obtain the total number NUM3 of the cells of the selected category in the sample of the selected image; the selected class of cell characteristic data set is obtained by training the labeled selected class of cells.
12. The method for detecting a sample of erythrocyte containing nuclei according to claim 11 wherein, The selected class of cells includes any one or more of thrombocytes, thrombocyte clusters, leukocytes, naive nucleated cells, reticulocytes, ghost red blood cells.
13. The method of claim 11, wherein the sample is NUM3/V1 in a selected cell type per volume.
14. The method for detecting a sample of erythrocyte containing nuclei according to claim 8 wherein, The sample is derived from mammal blood, and the nucleated red blood cell characteristic data set is combined with a corresponding AI algorithm, so that the capability of distinguishing the non-nucleated red blood cells of the mammal from the nucleated red blood cells of the mammal is provided.
15. The method for detecting a sample of erythrocyte containing nuclei according to claim 14, wherein the test sample is a mammalian sample; And identifying the normal red blood cells in the sample in the selected area S1 of the image by using an AI identification algorithm to obtain the total number NUM2 of the normal red blood cells in the sample in the selected image, wherein the AI identification algorithm identifies the normal red blood cells in the sample according to the characteristic data set of the normal red blood cells in the sample, and the ratio of the number of nucleated red blood cells to the number of the normal red blood cells=NUM1/NUM2.
16. A computing processing device is characterized in that, The method is characterized by comprising any one of the following technical characteristics: TD1 is used to run all or part of the method of any one of claims 1 to 15; TD2 the memory of the computing device comprises all or part of the data of the method according to any one of claims 1 to 15.
17. A data storage device, characterized in that, The method is characterized by comprising any one of the following technical characteristics: TE1 storing all or part of the program code for performing the method of any one of claims 1 to 15; TE2 storing all or part of the data of the method according to any one of claims 1 to 15.
18. A detection device is characterized in that, Part or all of the method for performing any one of claims 1 to 15; memory comprising all or part of the data of the method of any one of claims 1 to 15.

Description

AI training and detection method for detecting nuclear erythrocyte sample Technical Field The application belongs to the technical field of microscopic amplified image-based formed component analysis, and particularly relates to an AI (animal husbandry) detection training and sample detection method, a calculation processing device, a data storage device and a detection device for a nuclear erythrocyte sample. Background The variety of nucleated red blood cells is wide, and as the observation means are gradually upgraded, the variety of nucleated red blood cells that can be observed by humans is increasing. The red blood cells are the most abundant blood cells in blood, the red blood cells of mammals in most vertebrates are in the shape of a round cake with a concave center part, and the red blood cells of human beings are also in the shape of a round cake with a convex edge and a concave center. When an abnormality occurs in erythrocytes, it is often an indicator of some diseases. When nucleated red blood cells are present in the cells. Often a sign of disease. The characteristics of erythrocytes of different species vary, and the probability and extent of erythrocyte absence are greater for animals of higher stages of evolution. How to detect and identify various different types of nucleated red blood cells is a technical problem to be solved. However, in the blood cell analyzer based on the principles of the prior art, such as blood smear microscopy, fluorescence staining, flow analyzer and coulter electrical impedance, it is difficult to detect the nucleated red blood cells, because the white blood cells in the blood are nucleated cells, if the detection is based on the optical characteristics of the nuclei, it is impossible to distinguish whether the white blood cells are nucleated red blood cells or not, especially for some non-mammalian animals, the nucleated red blood cells are the main red blood cell morphology, and it is more difficult to classify, identify and count the nucleated red blood cells separately. Disclosure of Invention In the application, the inventor provides that AI training is carried out on a suspension picture containing a nucleated red blood cell sample, and the obtained AI characteristic data set A can effectively support AI detection of the nucleated red blood cells in the identification sample, and the identification of the nucleated red blood cells is finished by utilizing advanced AI calculation force, so that the analysis of the nucleated red blood cells of the blood balls is more efficient and more accurate. The technical scheme for solving the technical problems is that the AI training method for detecting the nuclear erythrocyte-containing sample comprises the steps of preprocessing the nuclear erythrocyte-containing sample to obtain a detection sample, tiling the detection sample, suspending cells to sink, shooting the tiled detection sample to obtain a detection sample image, identifying and labeling the nuclear erythrocyte in the detection sample image to obtain a labeling picture, carrying out AI training by using the labeling picture to obtain an AI characteristic data set A, wherein the AI characteristic data set A comprises nuclear erythrocyte sample characteristics, and combining the AI characteristic data set A with a corresponding AI algorithm to have the capability of identifying the nuclear erythrocyte. The method comprises the steps of adding the mammal non-nucleated red blood cells into a sample to obtain a detection sample, identifying and labeling the mammal non-nucleated red blood cells in an image of the detection sample to obtain a labeling picture, carrying out AI training by using the labeling picture to obtain an AI characteristic data set A, wherein the AI characteristic data set A comprises characteristics of the mammal non-nucleated red blood cells, the nucleated red blood cells are mammal nucleated red blood cells, and the AI characteristic data set A is combined with a corresponding AI algorithm and has the capability of distinguishing the mammal non-nucleated red blood cells from the mammal nucleated red blood cells. The method comprises the steps of adding white blood cells of a selected species into a sample to obtain a detection sample, identifying and labeling the white blood cells of the selected species in an image of the detection sample to obtain a labeling picture, and carrying out AI training by using the labeling picture to obtain an AI characteristic data set A, wherein the AI characteristic data set A comprises the characteristics of the white blood cells of the selected species, and the AI characteristic data set A is combined with a corresponding AI algorithm and has the capability of distinguishing nucleated red blood cells from white blood cells of the selected species. The method comprises the steps of adding selected species thrombocytes or thrombocytes clusters into a sample to obtain a detection sample, identifying and la