A method and system for extracting morphological characteristics of bone marrow cells

Abstract

The present invention discloses a method and system for extracting morphological features of bone marrow cells, including: 1) obtaining a bone marrow cell image obtained after staining with a chemical reagent, and performing image preprocessing; 2) using a UNet++ model to segment the bone marrow cell image to obtain a number of single bone marrow cell images; 3) improving the marker-based watershed algorithm in combination with medical diagnosis knowledge, performing cell nucleus segmentation on a single bone marrow cell image according to a nucleus numerical description, and obtaining an accurate cell nucleus area; 4) extracting morphological features of a single bone marrow cell image, and performing feature selection to screen out morphological features with higher importance. The present invention breaks through the problem that the traditional marker-based watershed algorithm is difficult to determine the cell nucleus area when performing nuclear cytoplasm segmentation on a single bone marrow cell image, and obtains the cell nucleus area ratio information of different types of bone marrow cells in advance according to the knowledge of bone marrow cell morphology inspection, thereby extracting more accurate morphological features.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image feature extraction, and in particular to a method and system for extracting bone marrow cell morphological features. Background Art

[0002] Bone marrow cell morphology examination is one of the simplest and most important means to diagnose blood system diseases, judge the efficacy of treatment and detect changes in the disease. In clinical testing, the stained bone marrow cells are observed under a microscope to directly understand the number, proportion, morphology and presence of abnormal cells of various blood cells in the bone marrow. To date, manual microscopy is still the main method for bone marrow morphology examination, but its inspection process is time-consuming, labor-intensive and inefficient. In recent years, artificial intelligence technology has developed rapidly in the field of medical imaging, providing many new ideas in bone marrow cell segmentation and recognition.

[0003] Bone marrow cell morphology extraction generally includes the following steps: 1) image preprocessing of bone marrow cell images; 2) cell segmentation of the preprocessed bone marrow cell images to obtain a single bone marrow cell image; 3) cell morphology feature extraction of a single bone marrow cell image. The cell region and the nucleus region of bone marrow cells are important bases for calculating the morphological features of bone marrow cells. How to obtain accurate cell regions and nucleus regions is crucial. Image segmentation algorithms such as the UNet model and its improved models that emerged under the background of artificial intelligence technology have been very successful in the field of medical images. Among them, the UNet++ model enables the depth of the UNet model to adapt to the scale of the data set, with obvious improvement effects, and can accurately obtain cell regions. Traditional image segmentation methods such as the region-based watershed algorithm are widely used and have achieved good results in various tasks. Its improved algorithms such as the marker-based watershed algorithm limit the number of regions allowed to exist by incorporating preprocessing steps, and optimize the problem that the original watershed algorithm is very easy to over-segment images with noise and irregular gradients. However, in the task of nuclear and cytoplasmic segmentation of bone marrow cells, there are still many segmentation regions and it is difficult to determine the nucleus region, making it difficult to obtain an accurate nucleus region. Summary of the invention

[0004] The first purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and provide a method for extracting morphological features of bone marrow cells, breaking through the problem of difficulty in determining the nucleus area when performing nuclear cytoplasm segmentation on a single bone marrow cell image using the traditional marker-based watershed algorithm. Based on the knowledge of bone marrow cell morphology examination, the nucleus volume ratio information of different types of bone marrow cells is obtained in advance, thereby improving the accuracy of nucleus area determination and extracting more accurate morphological features.

[0005] The second object of the present invention is to provide a system for extracting morphological characteristics of bone marrow cells.

[0006] The first object of the present invention is achieved by the following technical solution: A method for extracting morphological characteristics of bone marrow cells, comprising the following steps:

[0007] 1) Obtaining an image of bone marrow cells after staining with chemical reagents and performing image preprocessing;

[0008] 2) Use the UNet++ model to segment the bone marrow cell image and obtain several single bone marrow cell images;

[0009] 3) Improve the marker-based watershed algorithm based on medical diagnosis knowledge, perform nucleus segmentation on a single bone marrow cell image according to the nucleus numerical description, and obtain accurate nucleus regions;

[0010] 4) Extracting the morphological features of a single bone marrow cell image, and performing feature selection to screen out morphological features of high importance.

[0011] Further, in step 1), the chemical reagent is the Wright dye required in the Wright staining process of bone marrow cells; the bone marrow cell image is a microscopic image taken by a microscope camera after microscopic examination of a bone marrow smear; the image preprocessing includes: ① grayscale processing, converting a color image into a grayscale image; ② filtering processing, removing noise through a filtering function; ③ edge detection, obtaining edge information in the bone marrow cell image through an edge detection operator.

[0012] Further, in step 2), the UNet++ model is an image segmentation algorithm, which proposes a UNet architecture that is adaptive to data sets of different sizes based on the original UNet model.

[0013] Further, the step 3) comprises the following steps:

[0014] 3.1) Based on the knowledge of bone marrow cell morphology test, the volume proportion information of the nucleus of different types of cells in each lineage and growth stage of bone marrow cells is obtained and converted into a numerical description that can be quantified and calculated; the numerical description is a set of volume proportion parameters, including a one-to-one correspondence between bone marrow cell types and their volume proportion parameters, expressed as Areas = {type_i:areaInfo_i, i = 1, ..., n}, where Areas represents the numerical description, type_i represents a certain type of bone marrow cell, areaInfo_i represents the numerical description of the volume proportion parameter corresponding to type_i, and n represents the range threshold of the bone marrow cell type;

[0015] 3.2) According to the numerical description obtained in step 3.1), the watershed algorithm based on markers is improved to determine the nucleus region of a single bone marrow cell image; when the watershed algorithm based on markers performs nuclear cytoplasm segmentation on a single bone marrow cell image, there is a problem of too many segmented regions, and because the morphologies of different types of cells are very different, the volumes of the nuclear regions are not the same, and a single volume ratio parameter cannot be applied to all types of bone marrow cells, making the determination of the nuclear region more difficult; therefore, when performing nuclear cytoplasm segmentation on single bone marrow cell images of different types, the corresponding volume ratio parameter is used to determine the region where the nucleus is located, thereby improving the accuracy of the determination of the nuclear region;

[0016] 3.3) Output the total number of nucleus pixels and cell pixels of a single bone marrow cell image.

[0017] Further, in step 4), the morphological features include shape features, texture features and color features, and the feature selection is to perform a feature screening operation on the extracted morphological features.

[0018] The second object of the present invention is achieved by the following technical solution: A system for extracting morphological characteristics of bone marrow cells, comprising:

[0019] A bone marrow cell image acquisition and preprocessing module is used to acquire the bone marrow cell image obtained after being stained with chemical reagents and perform image preprocessing;

[0020] The bone marrow cell image segmentation module uses the UNet++ model to segment the bone marrow cell images and obtain several single bone marrow cell images;

[0021] The single bone marrow cell image nucleus segmentation module combines medical diagnosis knowledge to improve the marker-based watershed algorithm, and performs nucleus segmentation on a single bone marrow cell image according to the nucleus numerical description to obtain a more accurate nucleus area;

[0022] The morphological feature extraction and screening module extracts the morphological features of a single bone marrow cell image, performs feature selection, and screens out morphological features of high importance.

[0023] Furthermore, in the bone marrow cell image acquisition and preprocessing module, the chemical reagent is the Wright dye required in the bone marrow cell Wright staining process; the bone marrow cell image is a microscopic image taken by a microscope camera after microscopic examination of a bone marrow smear; the image preprocessing includes: ① grayscale processing, converting a color image into a grayscale image; ② filtering processing, removing noise through a filtering function; ③ edge detection, obtaining edge information in the bone marrow cell image through an edge detection operator.

[0024] Furthermore, in the bone marrow cell image segmentation module, the UNet++ model is an image segmentation algorithm, which proposes a UNet architecture that is adaptive to data sets of different sizes based on the original UNet model.

[0025] Further, the single bone marrow cell image cell nucleus segmentation module performs the following operations:

[0026] According to the knowledge of bone marrow cell morphology test, the volume proportion information of the nucleus of different types of cells in each lineage and growth stage of bone marrow cells is obtained and converted into a numerical description that can be quantified and calculated; the numerical description is a set of volume proportion parameters, including a one-to-one correspondence between the bone marrow cell type and its volume proportion parameter, expressed as Areas = {type_i:areaInfo_i, i = 1, ..., n}, where Areas represents the numerical description, type_i represents a certain type of bone marrow cell, areaInfo_i represents the numerical description of the volume proportion parameter corresponding to type_i, and n represents the range threshold of the bone marrow cell type;

[0027] According to the obtained numerical description, the watershed algorithm based on markers is improved to determine the nucleus region of a single bone marrow cell image. When the watershed algorithm based on markers performs nuclear cytoplasm segmentation on a single bone marrow cell image, there is a problem of too many segmented regions. In addition, due to the large differences in morphology of different types of cells, the volume of the nuclear region is not the same. A single volume ratio parameter cannot be applied to all types of bone marrow cells, making the determination of the nuclear region more difficult. Therefore, when performing nuclear cytoplasm segmentation on single bone marrow cell images of different types, the corresponding volume ratio parameter is used to determine the region where the nucleus is located, thereby improving the accuracy of nuclear region determination.

[0028] Output the total number of nucleus pixels and cell pixels of a single bone marrow cell image.

[0029] Furthermore, in the morphological feature extraction and screening module, the morphological features include shape features, texture features and color features, and the feature selection is to perform a feature screening operation on the extracted morphological features.

[0030] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0031] 1. The present invention combines the knowledge of bone marrow cell morphology examination, converts the volume ratio of the nucleus of different types of cells in each lineage and growth stage of bone marrow cells into a quantifiable numerical description as a priori knowledge for determining the nuclear area.

[0032] 2. The present invention improves the marker-based watershed algorithm, taking into account the differences in the nucleus regions of different types of bone marrow cells, thereby obtaining a more accurate nucleus region. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The figure is a logical flow chart of the method of the present invention.

[0034] Figure 2 Schematic diagram of the system of the present invention. DETAILED DESCRIPTION

[0035] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0036] Example 1

[0037] like Figure 1 As shown, this embodiment discloses a method for extracting morphological characteristics of bone marrow cells, which comprises the following steps:

[0038] 1) Obtain the bone marrow cell image obtained after dyeing with chemical reagents, and perform image preprocessing. Wherein, the chemical reagent is the Wright dye required in the process of Wright staining of bone marrow cells; the bone marrow cell image is a microscopic image taken by a microscope camera after microscopic examination of a bone marrow smear; the image preprocessing includes: ① grayscale processing, converting the color image into a grayscale image; ② filtering processing, removing noise through a filtering function; ③ edge detection, obtaining edge information in the bone marrow cell image through an edge detection operator. In this embodiment, a total of 1000 stained bone marrow cell images are obtained, all in RGB image format. First, 1000 images are grayscaled to obtain 1000 grayscale images; then, a Gaussian filter is used to process the 1000 grayscale images, and the Gaussian kernel size is set to 7*7. Finally, the Canny operator is used for edge detection, and the maximum and minimum thresholds of edge detection are 200 and 60, respectively.

[0039] 2) Use the UNet++ model to segment the bone marrow cell image to obtain several single bone marrow cell images. The UNet++ model is an image segmentation algorithm that proposes a UNet architecture that is adaptive to different scale data sets based on the original UNet model. Because of its simple structure and friendliness to small sample data sets, it is more suitable for image segmentation tasks in the field of medical images. In this embodiment, a UNet++ model based on the PyTorch framework is used to divide 1000 preprocessed images into training sets and test sets in a ratio of 8:2. By inputting the training set into the model, a trained UNet++ model is obtained, and then the trained model is used to predict the test set to obtain several single bone marrow cell images.

[0040] 3) Improve the marker-based watershed algorithm by combining medical diagnosis knowledge, and segment the nucleus of a single bone marrow cell image according to the numerical description of the nucleus to obtain an accurate nucleus region, including the following steps:

[0041] 3.1) According to the morphological examination knowledge of bone marrow cells, obtain the volume ratio information of the nuclei of different types of cells in each lineage and growth stage of bone marrow cells, and convert it into a numerically calculable numerical description. The numerical description is a set of volume ratio parameters, including the one-to-one correspondence between bone marrow cell types and their volume ratio parameters, which can be expressed as Areas = {type_i: areaInfo_i, i = 1, …, n}, where Areas represents the numerical description, type_i represents a certain type of bone marrow cell, areaInfo_i represents the numerical description of the volume ratio parameter corresponding to type_i, and n represents the range threshold of bone marrow cell types. The morphological examination knowledge of bone marrow cells in this embodiment is shown in Table 1:

[0042] Table 1: Morphological knowledge of bone marrow cells

[0043] Serial number Myeloid cell type names Department Description of nuclear size 1 Proerythroblast Red Series The nucleus is usually larger than 3 / 4 of the cell size. 2 Mesocytic erythrocyte Red Series The nucleus is usually larger than half the size of the cell 3 Metaerythroblast Red Series The nucleus is usually less than half the size of the cell 4 Myeloblasts Granular system Less cell mass 5 Neutrophil Granular system The nucleus is usually larger than half the size of the cell 6 Lymphocytes Lymphatic system The nucleus is usually larger than half the size of the cell 7 Monocytes Single core system The nucleus is usually larger than half the size of the cell 8 Plasma cells Pulp system The nucleus is approximately half the size of the cell

[0044] Convert Table 1 into a numerical description, that is, numerical description = {early erythroblast: 0.75 < x < 1; intermediate erythroblast: 0.5 < x < 1; late erythroblast: 0 < x < 0.5; promyelocyte: 0.8 < x < 1; neutrophilic myelocyte 0.5 < x < 1; lymphocyte: 0.5 < x < 1; monocyte: 0.5 < x < 1; plasma cell: 0.4 < x < 0.6}. Where x represents the volume ratio parameter.

[0045] 3.2) Improve the marker-based watershed algorithm according to the numerical description obtained in step 3.1), and determine the nucleus region of a single bone marrow cell image. When the marker-based watershed algorithm performs nuclear-cytoplasmic segmentation on a single bone marrow cell image, there is a problem of too many segmented regions. Moreover, due to the large morphological differences of different types of cells, the volumes of the nucleus regions are not the same, and a single volume ratio parameter cannot be applied to all types of bone marrow cells, making it difficult to determine the nucleus region. Therefore, when performing nuclear-cytoplasmic segmentation on different types of single bone marrow cell images, use the corresponding volume ratio parameter to determine the region where the nucleus is located, thereby improving the accuracy of nucleus region determination.

[0046] 3.3) Output the total number of nucleus pixels and the total number of cell pixels of a single bone marrow cell image.

[0047] 4) Extract the morphological features of a single bone marrow cell image, perform feature selection, and screen out morphological features of high importance. Wherein, the morphological features include shape features, texture features and color features. The feature selection is to perform feature screening operations on the extracted shape features. In this embodiment, a total of 143 morphological features are extracted, of which shape features include cell volume (total number of cell pixels), nuclear volume (total number of nuclear pixels), volume ratio of nucleus to cytoplasm, volume ratio of nucleus to cell, minimum circumscribed rectangular volume of nucleus, roundness of nucleus, rectangularity of nucleus, etc., a total of 14; there are 102 texture features and 27 color features. The extracted morphological features are selected using principal component analysis, and the principal components with cumulative contribution rates>0.99 are selected, a total of 16, as the final features.

[0048] Example 2

[0049] This embodiment discloses a system for extracting morphological characteristics of bone marrow cells, such as Figure 2 As shown, it includes the following functional modules:

[0050] A bone marrow cell image acquisition and preprocessing module is used to acquire the bone marrow cell image obtained after being stained with chemical reagents and perform image preprocessing;

[0051] The bone marrow cell image segmentation module uses the UNet++ model to segment the bone marrow cell images and obtain several single bone marrow cell images;

[0052] The single bone marrow cell image nucleus segmentation module combines medical diagnosis knowledge to improve the marker-based watershed algorithm, and performs nucleus segmentation on a single bone marrow cell image according to the nucleus numerical description to obtain a more accurate nucleus area;

[0053] The morphological feature extraction and screening module extracts the morphological features of a single bone marrow cell image, performs feature selection, and screens out morphological features with higher importance.

[0054] Furthermore, in the bone marrow cell image acquisition and preprocessing module, the chemical reagent is the Wright dye required in the bone marrow cell Wright staining process; the bone marrow cell image is a microscopic image taken by a microscope camera after microscopic examination of a bone marrow smear; the image preprocessing includes: ① grayscale processing, converting a color image into a grayscale image; ② filtering processing, removing noise through a filtering function; ③ edge detection, obtaining edge information in the bone marrow cell image through an edge detection operator.

[0055] Furthermore, in the bone marrow cell image segmentation module, the UNet++ model is an image segmentation algorithm. Based on the original UNet model, it proposes a UNet architecture that is adaptive to data sets of different sizes. Because of its simple structure and friendliness to small sample data sets, it is more suitable for image segmentation tasks in the field of medical images.

[0056] Further, the single bone marrow cell image cell nucleus segmentation module performs the following operations:

[0057] According to the knowledge of bone marrow cell morphology test, the volume proportion information of the nucleus of different types of cells in each lineage and growth stage of bone marrow cells is obtained and converted into a numerical description that can be quantified and calculated; the numerical description is a set of volume proportion parameters, including a one-to-one correspondence between the bone marrow cell type and its volume proportion parameter, expressed as Areas = {type_i:areaInfo_i, i = 1, ..., n}, where Areas represents the numerical description, type_i represents a certain type of bone marrow cell, areaInfo_i represents the numerical description of the volume proportion parameter corresponding to type_i, and n represents the range threshold of the bone marrow cell type;

[0058] According to the obtained numerical description, the watershed algorithm based on markers is improved to determine the nucleus region of a single bone marrow cell image. When the watershed algorithm based on markers performs nuclear cytoplasm segmentation on a single bone marrow cell image, there is a problem of too many segmented regions. In addition, due to the large differences in morphology of different types of cells, the volume of the nuclear region is not the same. A single volume ratio parameter cannot be applied to all types of bone marrow cells, making the determination of the nuclear region more difficult. Therefore, when performing nuclear cytoplasm segmentation on single bone marrow cell images of different types, the corresponding volume ratio parameter is used to determine the region where the nucleus is located, thereby improving the accuracy of nuclear region determination.

[0059] Output the total number of nucleus pixels and cell pixels of a single bone marrow cell image.

[0060] Furthermore, in the morphological feature extraction and screening module, the morphological features include shape features, texture features and color features, and the feature selection is to perform a feature screening operation on the extracted morphological features.

[0061] In summary, after adopting the above scheme, the present invention provides a new method and system for extracting the morphological features of bone marrow cells. Combined with the knowledge of bone marrow cell morphological examination, the marker-based watershed algorithm is improved to obtain more accurate bone marrow cell morphological features, effectively improve the efficiency of bone marrow cell morphological examination, have practical utilization value, and are worthy of promotion.

[0062] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for extracting morphological characteristics of bone marrow cells, characterized in that: The following steps are involved: 1) Obtaining bone marrow cell images after staining with chemical reagents and performing image preprocessing; 2) Use the UNet++ model to segment the bone marrow cell images and obtain several single bone marrow cell images; 3) Combine medical diagnosis knowledge to improve the marker-based watershed algorithm, perform nucleus segmentation on a single bone marrow cell image according to the nucleus numerical description, and obtain accurate nucleus regions, including the following steps: 3.1) Based on the knowledge of bone marrow cell morphology, the volume percentage information of the nuclei of different types of cells in each lineage and growth stage of bone marrow cells is obtained and converted into a numerical description that can be quantified and calculated; the numerical description is a set of volume percentage parameters, including a one-to-one correspondence between bone marrow cell types and their volume percentage parameters, expressed as Areas ={type_i : areaInfo_i, i = 1, …, n}, where Areas represents the numerical description, type_i represents a certain type of bone marrow cell, areaInfo_i represents the numerical description of the volume percentage parameter corresponding to type_i, and n represents the range threshold of the bone marrow cell type; 3.2) According to the numerical description obtained in step 3.1), the watershed algorithm based on the marker is improved to determine the nucleus area of ​​the single bone marrow cell image; when performing nuclear cytoplasm segmentation on different types of single bone marrow cell images, the corresponding volume ratio parameter is used to determine the area where the nucleus is located; 3.3) Output the total number of nucleus pixels and cell pixels of a single bone marrow cell image; 4) Extract the morphological features of a single bone marrow cell image and perform feature selection to screen out the morphological features with high importance.

2. The method for extracting morphological characteristics of bone marrow cells according to claim 1, characterized in that: In step 1), the chemical reagent is the Wright dye required in the process of Wright staining of bone marrow cells; The bone marrow cell image is a microscopic image taken by a microscope camera after microscopic examination of a bone marrow smear; the image preprocessing includes: ① grayscale processing, converting a color image into a grayscale image; ② filtering processing, removing noise through a filtering function; ③ edge detection, obtaining edge information in the bone marrow cell image through an edge detection operator.

3. The method for extracting morphological characteristics of bone marrow cells according to claim 1, characterized in that: In step 2), the UNet++ model is an image segmentation algorithm, which proposes a UNet architecture that is adaptive to data sets of different sizes based on the original UNet model.

4. The method for extracting morphological characteristics of bone marrow cells according to claim 1, characterized in that: In step 4), the morphological features include shape features, texture features and color features, and the feature selection is to perform a feature screening operation on the extracted morphological features.

5. A system for extracting morphological characteristics of bone marrow cells, characterized in that: include: A bone marrow cell image acquisition and preprocessing module is used to acquire the bone marrow cell image obtained after being stained with chemical reagents and perform image preprocessing; The bone marrow cell image segmentation module uses the UNet++ model to segment the bone marrow cell images and obtain several single bone marrow cell images; The single bone marrow cell image nucleus segmentation module combines medical diagnosis knowledge to improve the marker-based watershed algorithm, and performs nucleus segmentation on a single bone marrow cell image according to the nucleus numerical description to obtain a more accurate nucleus area; The morphological feature extraction and screening module extracts the morphological features of a single bone marrow cell image and performs feature selection to screen out the morphological features with high importance. The single bone marrow cell image cell nucleus segmentation module performs the following operations: According to the knowledge of bone marrow cell morphology test, the volume proportion information of the nucleus of different types of cells in each lineage and growth stage of bone marrow cells is obtained and converted into a numerical description that can be quantified and calculated; the numerical description is a set of volume proportion parameters, including a one-to-one correspondence between bone marrow cell types and their volume proportion parameters, expressed as Areas ={type_i : areaInfo_i, i = 1, …, n}, where Areas represents the numerical description, type_i represents a certain type of bone marrow cell, areaInfo_i represents the numerical description of the volume proportion parameter corresponding to type_i, and n represents the range threshold of the bone marrow cell type; According to the obtained numerical description, the watershed algorithm based on markers is improved to determine the nucleus area of ​​a single bone marrow cell image; when performing nuclear-cytoplasmic segmentation on different types of single bone marrow cell images, the corresponding volume ratio parameters are used to determine the area where the nucleus is located; Output the total number of nucleus pixels and cell pixels of a single bone marrow cell image.

6. A system for extracting morphological characteristics of bone marrow cells according to claim 5, characterized in that: In the bone marrow cell image acquisition and preprocessing module, the chemical reagent is the Wright dye required in the bone marrow cell Wright staining process; The bone marrow cell image is a microscopic image taken by a microscope camera after microscopic examination of a bone marrow smear; the image preprocessing includes: ① grayscale processing, converting a color image into a grayscale image; ② filtering processing, removing noise through a filtering function; ③ edge detection, obtaining edge information in the bone marrow cell image through an edge detection operator.

7. A system for extracting morphological characteristics of bone marrow cells according to claim 5, characterized in that: In the bone marrow cell image segmentation module, the UNet++ model is an image segmentation algorithm, which proposes a UNet architecture that is adaptive to data sets of different sizes based on the original UNet model.

8. The system for extracting morphological characteristics of bone marrow cells according to claim 5, characterized in that: In the morphological feature extraction and screening module, the morphological features include shape features, texture features and color features, and the feature selection is to perform feature screening operations on the extracted morphological features.

Images