CN111476754A - A system and method for bone marrow cell imaging artificial intelligence-aided grading diagnosis - Google Patents

Abstract

The invention relates to the technical field of medical assistant diagnosis, and discloses a bone marrow cell image artificial intelligence assistant grading diagnosis system and method, comprising a data acquisition unit, an automatic identification and labeling unit, a manual labeling unit, a cell data statistics unit, a data analysis and grading unit, An image-assisted grading diagnosis unit, a display, a storage module and a processor, including a data acquisition unit, an automatic identification and labeling unit, a manual labeling unit, a cell data statistics unit, a data analysis and grading unit, an image-assisted grading diagnosis unit, a storage module and The displays are all connected in communication with the processor. The present invention can not only automatically identify, label, count analysis and grade diagnosis for the bone marrow cell image, but also manually mark the bone marrow cell image, and store the manually marked results in the storage module, and the system can perform the manual marking information. Memory, and gradually improve the system's ability to distinguish bone marrow cell images.

Description

A system and method for bone marrow cell imaging artificial intelligence-aided grading diagnosis

technical field

The invention relates to the technical field of medical auxiliary diagnosis, in particular to a bone marrow cell image artificial intelligence auxiliary classification diagnosis system and method.

Background technique

Research shows that the annual growth rate of medical imaging data in my country is about 30%, while the annual growth rate of the number of diagnostic imaging physicians is only about 4.1%, which means that diagnostic imaging physicians will be overworked in the future; It reduces the diagnostic efficiency of doctors and even reduces the accuracy of diagnosis; in addition, because medical imaging diagnosis has high requirements on the diagnostic experience of doctors, and in areas with low development level, the resources of experienced diagnostic doctors are relatively scarce.

With the development of modern medical science and technology, various new technologies have gradually penetrated into the medical field. In order to reduce the pressure of imaging diagnosis physicians to a certain extent, an artificial intelligence-assisted diagnosis system for imaging has been proposed. Artificial intelligence + medical imaging is the specific application of artificial intelligence technology to the diagnosis of medical imaging. Currently, it is mainly divided into two parts. One is image recognition, which is applied to the perception link. Its main purpose is to analyze non-institutionalized data such as images. , to get some meaningful information. The second is deep learning, which is applied to learning and analysis, and is the core link of artificial intelligence applications. Through a large amount of image data and diagnostic data, deep learning training is continuously carried out on neuron networks to enable them to master the ability of "diagnosis".

The patent with the application number CN 201810866088.3 discloses an auxiliary diagnosis system for tumor malignancy risk stratification of artificial intelligence medical images, including: a data acquisition module, a data preprocessing module, a model establishment module, a model verification and optimization module, and a hierarchical diagnosis module and database platform. The patented auxiliary diagnosis system for tumor malignancy risk stratification is based on artificial intelligence technology, which can realize the successive stratification of the malignant risk of tumors, simulate the clinical diagnosis idea, and use the artificial intelligence model to detect benign lesions and malignant tumors with high precision. It can automatically diagnose space-occupying lesions with clear imaging features, which can substantially assist the clinical management decision-making of space-occupying lesions, improve the existing workflow of clinical diagnosis, increase the physician’s confidence in diagnosis, reduce work pressure, and reduce the risk of low malignancy. The anxiety of patients with lesions greatly improves the diagnosis rate of benign lesions and malignant tumors.

Although the above-mentioned patent realizes the intelligent auxiliary diagnosis of images, it is only applicable to the intelligent auxiliary diagnosis of lung cancer, liver cancer and other cell images. For the images of bone marrow cells, it is necessary to identify the complexity of the size, granules and nuclei of each cell in the bone marrow cells. Since there are three major systems of granulocyte lineage, erythrocyte lineage and megakaryocyte lineage in bone marrow cells, some of the various cells have small differences in size and other data, so the intelligent auxiliary diagnosis of the above scheme The recognition accuracy of the system for bone marrow cell images is still low.

SUMMARY OF THE INVENTION

In view of this, the purpose of the present invention is to provide a bone marrow cell image artificial intelligence-assisted grading diagnosis system and method, which can not only automatically identify, label, count, analyze and grade the bone marrow cell image, but also can perform the bone marrow cell image. Manually mark and store the manually marked results in the storage module, the system memorizes the manually marked information, and gradually improves the system's ability to distinguish bone marrow cell images.

The present invention solves the above-mentioned technical problems through the following technical means:

A bone marrow cell imaging artificial intelligence-assisted grading diagnosis system, comprising a data acquisition unit, an automatic identification and labeling unit, a manual labeling unit, a cell data statistics unit, a data analysis and grading unit, an image-assisted grading diagnosis unit, a display, a storage module and a processor , the data acquisition unit, the automatic identification and labeling unit, the manual labeling unit, the cell data statistics unit, the data analysis and classification unit, the image-assisted classification and diagnosis unit, the storage module and the display are all connected with the processor in communication; the data collection unit It includes an image acquisition module for reading information on bone marrow cell images, a clinical information acquisition module for inputting patient clinical information, and a gene information acquisition module for inputting patient genetic information; the automatic identification and labeling unit includes a The cell morphology standard module for standard cell morphology, the bone marrow cell morphology extraction module for automatically extracting the morphology of bone marrow cells, the cell morphology comparison module for comparing the extracted information of bone marrow cell images with the cell morphology standard module, and the identified bone marrow cells. A cell morphology automatic labeling module for automatically labeling the morphology; the manual labeling unit includes a cell morphology manual labeling module for manual operation by a doctor to manually label bone marrow cell images, and a labeling storage module for manually labeling information by the doctor. The data analysis unit includes a granulocyte ratio analysis module for calculating the ratio of nucleated cells to mature red blood cells, and a cell ratio analysis module for calculating the proportion of the number of each cell in the total number of bone marrow cells.

Further, the image-assisted grading diagnosis unit includes first-level, second-level, third-level, and fourth-level, the first-level includes normal bone marrow images and abnormal bone marrow images, and the second-level is a large category of diseases, including anemia bone marrow images. , Proliferative anemia bone marrow pattern, the third level is a certain disease, including IDA bone marrow pattern, CL bone marrow pattern, AL bone marrow pattern, the fourth level is a certain gene-containing disease, including APL bone marrow pattern, CML bone marrow pattern .

Further, it also includes a data preprocessing unit, the data preprocessing unit includes a data noise reduction module that performs noise reduction and data normalization processing on the data collected by the image acquisition module, searches for lesions corresponding to the pathological results, and performs lesions on the images. Annotated lesion annotation module and lesion segmentation module.

Further, the cell data statistics unit includes a cell classification module and a cell counting module.

Further, the cell counting module divides the bone marrow cells into: erythroid cells, metamyelocytes, other erythroid cells, blast cells, mature lymphocytes, other lymphoid cells, monocyte lineage cells, promyelocytic cells, and juvenile cells. Granulocytes, metamyelocytes, rod-shaped nuclear cells, segmented nuclear cells and other myeloid cells.

Further, the cell data analysis unit further includes an error calibration module for eliminating unnecessary cell interference and classifying the bone marrow cells accurately and efficiently.

A method for a bone marrow cell imaging artificial intelligence-assisted grading diagnosis system, comprising the following steps:

A1. The image acquisition module reads information from bone marrow cell images, the clinical information acquisition module inputs the patient's clinical information, and the gene information acquisition module inputs the patient's genetic information;

A2. The data noise reduction module performs noise reduction and data normalization processing on the data collected by the image acquisition module, finds the lesions corresponding to the pathological results, and marks the lesions on the images;

A3. The cell morphology extraction module automatically extracts the morphology of bone marrow cells, the cell morphology comparison module compares the extracted information of the bone marrow cell image with the cell morphology standard module, and the cell morphology automatic labeling module automatically labels the identified bone marrow cell morphology;

A4. Physicians perform manual operations, use the manual labeling module to manually label the bone marrow cell images, and the labeling storage module stores the manually labeling information of the physicians;

A5. The error calibration module eliminates useless cell interference and accurately and efficiently classifies bone marrow cells;

A6. The cytometry module divides the bone marrow cells into: erythroid cells, metamyelocytes, other erythroid cells, blast cells, mature lymphocytes, other lymphoid cells, monocyte lineage cells, promyelocytes, and granulocytes , metamyelocytes, rod-shaped nuclear cells, lobulated nuclear cells and other myeloid cells;

A7. The granulocyte ratio analysis module calculates the ratio of nucleated cells to mature red blood cells, and the cell ratio analysis module calculates the proportion of each type of cell in the total number of bone marrow cells;

A8. The image-assisted grading diagnosis unit divides the bone marrow images into: first, second, third and fourth grades. The first grade includes normal bone marrow images and abnormal bone marrow images, and the second grade includes anemia bone marrow images and proliferative anemia bone marrow images. The third level includes IDA bone marrow, CL bone marrow, and AL bone marrow, and the fourth level includes APL bone marrow and CML bone marrow.

Further, the error calibration method of the error calibration module includes the following steps:

B1. Obtain the stained bone marrow cell image, map the bone marrow cell image to the HSV space, and isolate the S channel;

B2. Draw the histogram of the S channel image, binarize the S channel image according to the threshold range, obtain the binary image of the bone marrow cells, and perform morphological processing on the binary image;

B3. Extract the edge pixels of the morphologically processed bone marrow cell image by the method of connected domain, find the edge pixels of the upper, lower, left, and right edge of the bone marrow cells, and then segment the bone marrow cells;

B4. Select images of bone marrow cells after segmentation, and input cell images with obvious features in each category as training cells into the deep residual network to train the network;

B5. Use the remaining cells after selection as the test cells, and use the softmax classifier to score the test cells; if the maximum score is greater than or equal to the set threshold, it is classified into a certain category; if the maximum score is less than the set threshold, it is classified as a sub-class middle;

B6, take any pixel point of the training cell and the cell edge in the sub-classification in step S5 as a pole, establish a polar coordinate system, and map all the pixel points to a rectangular coordinate system one by one with polar coordinate transformation;

B7. For the training cells, traverse the edge pixels of the cells, each pixel generates a transformed image, each image has n pixels, that is, transform n times; for the cells in the sub-category, each image only transforms once;

B8. Use the transformed image of the training cell as the input of the sub-classification network, retrain the deep residual network, and save the network parameters; take the transformed image of the cell in the sub-classification as the test data, and use the softmax classifier to score again: if the cell If the maximum score of the image is greater than or equal to the set threshold, the cell image is classified as a subclass; if the maximum score is less than the set threshold, the cell image is classified as unclassified cells.

Further, if the proportion of unclassified cells is greater than the set unclassified threshold, the unclassified cells are then segmented and graded, including the following steps, C1, for the classified cells to obtain the edge pixels of the morphologically processed bone marrow cell image , find the upper, lower, left, and right edge pixels of the bone marrow cells, and segment the bone marrow cells;

C2. Input the segmented bone marrow cell images as training cells into the deep residual network to train the network;

C3. Then use the softmax classifier to score the test cells. If the maximum score is greater than or equal to the set threshold, it will be classified into a certain category; if the maximum score is less than the set threshold, it will be classified into the final unclassified cell;

C4. If the final proportion of unsorted cells is less than the set unsorted threshold, stop; if the final unsorted cell proportion is greater than the set unsorted threshold, continue step C1 until the final unsorted cell proportion is less than the set unsorted threshold The unclassified threshold of .

Beneficial effects of the present invention:

(1) The present invention utilizes the data acquisition unit to read the information of the bone marrow cell image, input the clinical information of the patient, and input the genetic information of the patient; and then use the data noise reduction module to perform noise reduction and The data is normalized to find the lesions corresponding to the pathological results and mark the lesions on the images; then the morphology of the bone marrow cells is automatically extracted, and the cell morphology comparison module compares the extracted information of the bone marrow cell images with the cell morphology standard module. The morphological automatic labeling module automatically labels the morphology of the identified bone marrow cells; and the cell counting module divides the bone marrow cells into: middle erythrocytes, metamyelocytes, other erythroid cells, blast cells, mature lymphocytes, and other lymphoid cells , monocytes, promyelocytes, myelocytes, metamyelocytes, rod-shaped nucleated cells, lobulated nucleated cells and other myeloid cells; calculate the ratio of nucleated cells to mature red blood cells, and the proportion of cells The analysis module calculates the proportion of the number of each type of cells in the total number of bone marrow cells; finally, the image-aided grading diagnosis unit divides the bone marrow image into: first, second, third, and fourth. Intelligent grading diagnosis reduces the workload of radiologists.

(2) In the present invention, through the setting of the manual labeling unit and the data storage unit, when in use, the accuracy rate of the cell identification by the auxiliary classification diagnosis system of the present invention can be identified one by one, and the correct identification can be confirmed and stored, and the wrong identification can be confirmed and stored. Identify and correct errors; in this way, in the process of using the present invention, the identification ability of the bone marrow cell image can also be gradually improved, so as to achieve the purpose of truly complete artificial intelligence image identification.

Description of drawings

1 is a schematic diagram of a bone marrow cell imaging artificial intelligence-assisted classification diagnosis system of the present invention;

FIG. 2 is a flowchart of the error calibration method of the error calibration module of the present invention.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings:

As shown in Figure 1-2:

A bone marrow cell imaging artificial intelligence-assisted classification diagnosis system, as shown in Figure 1, includes a data acquisition unit, an automatic identification and labeling unit, a manual labeling unit, a cell data statistics unit, a data analysis and classification unit, an image-assisted classification and diagnosis unit, and a display. , a storage module and a processor, including a data acquisition unit, an automatic identification and labeling unit, a manual labeling unit, a cell data statistics unit, a data analysis and grading unit, an image-assisted grading diagnosis unit, a storage module and a display that are all connected in communication with the processor; The acquisition unit includes an image acquisition module for reading bone marrow cell images, a clinical information acquisition module for inputting patient clinical information, and a gene information acquisition module for inputting patient genetic information; the automatic identification and labeling unit includes a The cell morphology standard module for standard cell morphology, the bone marrow cell morphology extraction module for automatically extracting the morphology of bone marrow cells, the cell morphology comparison module for comparing the extracted information of bone marrow cell images with the cell morphology standard module, and the identified bone marrow cells. The cell morphology automatic labeling module for automatically labeling the morphology; the manual labeling unit includes a cell morphology manual labeling module for manual operation by physicians and manual labeling of bone marrow cell images, and a labeling storage module for manually labeling information for physicians. Storage module; data The analysis unit includes a granulocyte ratio analysis module that calculates the ratio of nucleated cells to mature red blood cells, and a cell ratio analysis module that calculates the proportion of each type of cell in the total number of bone marrow cells; the image-aided grading diagnosis unit includes first-level, second-level Grade 3, Grade 4 and Grade 4, Grade 1 includes normal bone marrow and abnormal bone marrow, Grade 2 includes bone marrow of anemia, and bone marrow of proliferative anemia, Grade 3 includes bone marrow of IDA, CL, and AL, and Grade 4 Including APL bone marrow, CML bone marrow. It also includes a data preprocessing unit. The data preprocessing unit includes a data noise reduction module that performs noise reduction and data normalization processing on the data collected by the image acquisition module, and a lesion label that searches for lesions corresponding to the pathological results and labels the images. module and lesion segmentation module. The cell data statistics unit includes a cell classification module and a cell counting module. The cytometry module classifies bone marrow cells into: erythroblasts, metamyelocytes, other erythroid cells, blasts, mature lymphocytes, other lymphoid cells, monocytes, promyelocytes, myelocytes, late Myeloid cells, rod-shaped nuclear cells, segmented nuclear cells and other myeloid cells. The cell data analysis unit also includes an error calibration module for removing unwanted cell interference and classifying bone marrow cells accurately and efficiently.

A method for auxiliary diagnosis of a bone marrow cell imaging artificial intelligence-aided grading diagnosis system of the present invention includes the following steps:

A1. The image acquisition module reads information from bone marrow cell images, the clinical information acquisition module inputs the patient's clinical information, and the gene information acquisition module inputs the patient's genetic information;

A2. The data noise reduction module performs noise reduction and data normalization processing on the data collected by the image acquisition module, finds the lesions corresponding to the pathological results, and marks the lesions on the images;

A3. The cell morphology extraction module automatically extracts the morphology of bone marrow cells, the cell morphology comparison module compares the extracted information of the bone marrow cell image with the cell morphology standard module, and the cell morphology automatic labeling module automatically labels the identified bone marrow cell morphology;

A4. Physicians perform manual operations, use the manual labeling module to manually label the bone marrow cell images, and the labeling storage module stores the manually labeling information of the physicians;

A5. The error calibration module eliminates useless cell interference and accurately and efficiently classifies bone marrow cells;

A6. The cytometry module divides the bone marrow cells into: erythroid cells, metamyelocytes, other erythroid cells, blast cells, mature lymphocytes, other lymphoid cells, monocyte lineage cells, promyelocytes, and granulocytes , metamyelocytes, rod-shaped nuclear cells, lobulated nuclear cells and other myeloid cells;

A7. The granulocyte ratio analysis module calculates the ratio of nucleated cells to mature red blood cells, and the cell ratio analysis module calculates the proportion of each type of cell in the total number of bone marrow cells;

A8. The image-assisted grading diagnosis unit divides the bone marrow images into: first, second, third and fourth grades. The first grade includes normal bone marrow images and abnormal bone marrow images, and the second grade includes anemia bone marrow images and proliferative anemia bone marrow images. The third level includes IDA bone marrow, CL bone marrow, and AL bone marrow, and the fourth level includes APL bone marrow and CML bone marrow.

Among them, the error calibration method of the error calibration module in this system, as shown in Figure 2, includes the following steps:

B1. Obtain the stained bone marrow cell image, map the bone marrow cell image to the HSV space, and isolate the S channel;

B2. Draw the histogram of the S channel image, binarize the S channel image according to the threshold range, obtain the binary image of the bone marrow cells, and perform morphological processing on the binary image;

B3. Extract the edge pixels of the morphologically processed bone marrow cell image by the method of connected domain, find the edge pixels of the upper, lower, left, and right edge of the bone marrow cells, and then segment the bone marrow cells;

B4. Select images of bone marrow cells after segmentation, and input cell images with obvious features in each category as training cells into the deep residual network to train the network;

B5. Use the remaining cells after selection as the test cells, and use the softmax classifier to score the test cells; if the maximum score is greater than or equal to the set threshold, it is classified into a certain category; if the maximum score is less than the set threshold, it is classified as a sub-class middle;

B6, take any pixel point of the training cell and the cell edge in the sub-classification in step S5 as a pole, establish a polar coordinate system, and map all the pixel points to a rectangular coordinate system one by one with polar coordinate transformation;

B7. For the training cells, traverse the edge pixels of the cells, each pixel generates a transformed image, each image has n pixels, that is, transform n times; for the cells in the sub-category, each image only transforms once;

B8. Use the transformed image of the training cell as the input of the sub-classification network, retrain the deep residual network, and save the network parameters; take the transformed image of the cell in the sub-classification as the test data, and use the softmax classifier to score again: if the cell If the maximum score of the image is greater than or equal to the set threshold, the cell image is classified as a subclass; if the maximum score is less than the set threshold, the cell image is classified as unclassified cells. If the proportion of unsorted cells is greater than the set unsorted threshold of 5%, the unsorted cells are then segmented and graded, including the following steps:

C1. Take the edge pixels of the morphologically processed bone marrow cell image for the classified cells, find the edge pixels of the upper, lower, left and right edge of the bone marrow cells, and segment the bone marrow cells;

C2. Input the segmented bone marrow cell images as training cells into the deep residual network to train the network;

C3. Then use the softmax classifier to score the test cells. If the maximum score is greater than or equal to the set threshold, it will be classified into a certain category; if the maximum score is less than the set threshold, it will be classified into the final unclassified cell;

C4. If the final proportion of unsorted cells is less than 5% of the set unsorted threshold, stop; if the final unsorted cell proportion is greater than the set unsorted threshold, continue step C1 until the final unsorted cell proportion is less than The set unclassified threshold.

The use process of the present invention is as follows:

Use the data acquisition unit to read the information of the bone marrow cell image, enter the clinical information of the patient, and enter the genetic information of the patient; then use the data noise reduction module to denoise and normalize the data collected by the image acquisition module. , find the lesions corresponding to the pathological results and mark the lesions on the images; then automatically extract the morphology of the bone marrow cells, and the cell morphology comparison module compares the extracted information of the bone marrow cell images with the cell morphology standard module, and the cell morphology automatic labeling module compares The identified bone marrow cell morphology is automatically marked; and the cell counting module divides the bone marrow cells into: erythroblasts, metaplastic erythrocytes, other erythroid cells, blast cells, mature lymphocytes, other lymphoid cells, and monocytic cells , promyelocytes, myelocytes, metamyelocytes, rod-nucleated cells, lobulated nucleated cells and other myeloid cells; calculate the ratio of nucleated cells to mature red blood cells, and the cell proportion analysis module calculates each The proportion of the number of cells in the total number of bone marrow cells; finally, the image-aided grading diagnosis unit divides the bone marrow image into: first, second, third, and fourth. Reduce the workload of the radiologist.

The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced. Without departing from the spirit and scope of the technical solutions of the present invention, all of them should be included in the scope of the claims of the present invention. The technology, shape, and structural part that are not described in detail in the present invention are all well-known technologies.

Claims (9)

1. The artificial intelligence auxiliary grading diagnosis system for the bone marrow cell image is characterized in that: the cell data analysis and classification system comprises a data acquisition unit, an automatic identification and labeling unit, a manual labeling unit, a cell data statistics unit, a data analysis and classification unit, an image auxiliary classification diagnosis unit, a display, a storage module and a processor, wherein the data acquisition unit, the automatic identification and labeling unit, the manual labeling unit, the cell data statistics unit, the data analysis and classification unit, the image auxiliary classification diagnosis unit, the storage module and the display are all in communication connection with the processor;

the data acquisition unit comprises an image acquisition module for reading information of bone marrow cell images, a clinical information acquisition module for inputting clinical information of patients and a gene information acquisition module for inputting gene information of patients;

the automatic identification and marking unit comprises a cell form standard module for expressing standard cell forms, a bone marrow cell form extraction module for automatically extracting the forms of bone marrow cells, a cell form comparison module for comparing the extracted information of bone marrow cell images with the cell form standard module, and a cell form automatic marking module for automatically marking the identified bone marrow cell forms;

the manual labeling unit comprises a cell morphology manual labeling module and a labeling storage module, wherein the cell morphology manual labeling module is used for manually labeling the bone marrow cell images by a doctor, and the labeling storage module is used for manually storing labeling information of the doctor;

the data analysis unit comprises a granulocyte ratio analysis module for calculating the ratio of nucleated cells to mature red blood cells, and a cell ratio analysis module for calculating the ratio of the number of each cell to the total number of the whole bone marrow cells.

2. The system of claim 1, wherein the image-assisted hierarchical diagnosis unit comprises a primary stage, a secondary stage, a tertiary stage, and a quaternary stage, wherein the primary stage comprises normal myeloid elephants and abnormal myeloid elephants, the secondary stage is a large class of diseases comprising anemic myeloid elephants and proliferative anemic myeloid elephants, the tertiary stage is a disease comprising IDA myeloid elephants, C L myeloid elephants and A L myeloid elephants, and the quaternary stage is a disease containing specific genes comprising AP L myeloid elephants and CM L myeloid elephants.

3. The system of claim 2, wherein the bone marrow cell image artificial intelligence aided grading diagnosis system comprises: the data preprocessing unit comprises a data denoising module for denoising and data normalization processing of the data acquired by the image acquisition module, a lesion labeling module for searching lesions corresponding to pathological results and labeling the images, and a lesion segmentation module.

4. The system of claim 3, wherein the bone marrow cell image artificial intelligence aided grading diagnosis system comprises: the cell data statistical unit comprises a cell classification module and a cell counting module.

5. The system of claim 4, wherein the bone marrow cell image artificial intelligence aided grading diagnosis system comprises: the cell counting module divides the bone marrow cells into: mesoblast, metablast, other erythroid cells, blasts, mature lymphocytes, other lymphoid cells, monocytic cells, promyelocytes, mesogranulocytes, metagranulocytes, rhabdocytes, and other granulocytic cells.

6. The system of claim 5, wherein the bone marrow cell image artificial intelligence aided grading diagnosis system comprises: the cell data analysis unit also comprises an error calibration module which is used for eliminating useless cell interference and accurately and efficiently classifying the bone marrow cells.

7. The method of claim 6, wherein the step of performing artificial intelligence-aided hierarchical diagnosis on the bone marrow cell image comprises: the method comprises the following steps:

a1, reading information of a bone marrow cell image by an image acquisition module, inputting clinical information of a patient by a clinical information acquisition module, and inputting gene information of the patient by a gene information acquisition module;

a2, the data noise reduction module carries out noise reduction and data normalization processing on the data acquired by the image acquisition module, searches for lesions corresponding to pathological results and carries out lesion marking on the images;

a3, automatically extracting the morphology of the bone marrow cells by a cell morphology extraction module, comparing the extraction information of the bone marrow cell images with a cell morphology standard module by a cell morphology comparison module, and automatically labeling the recognized bone marrow cell morphology by a cell morphology automatic labeling module;

a4, manually operating by a doctor, manually labeling the bone marrow cell image by using a manual labeling module, and manually storing labeling information of the doctor by using a labeling storage module;

a5, an error calibration module discharges useless cell interference, and bone marrow cells are accurately and efficiently classified;

a6, cell counting module divides bone marrow cells into: mesoblast, metablast, other erythroid cells, blasts, mature lymphocytes, other lymphoid cells, monocytic cells, promyelocytes, mesogranulocytes, metagranulocytes, rhabdocytes, subtenocytes, and other myeloid cells;

a7, calculating the ratio of nucleated cells to mature red blood cells by a granulocyte ratio analysis module, and calculating the ratio of the number of each cell in the total number of the whole bone marrow cells by a cell ratio analysis module;

a8, image-aided grading diagnosis unit divides the bone marrow image into primary, secondary, tertiary and quaternary bone marrow images, wherein the primary bone marrow image comprises normal bone marrow image and abnormal bone marrow image, the secondary bone marrow image comprises anemia bone marrow image and proliferative anemia bone marrow image, the tertiary bone marrow image comprises IDA bone marrow image, C L bone marrow image and A L bone marrow image, and the quaternary bone marrow image comprises AP L bone marrow and CM L bone marrow image.

8. The method of claim 7, wherein the step of performing artificial intelligence-aided hierarchical diagnosis on the bone marrow cell image comprises: the error calibration method of the error calibration module comprises the following steps:

b1, obtaining a stained bone marrow cell image, mapping the bone marrow cell image to an HSV space, and separating an S channel;

b2, drawing a histogram of the S channel image, binarizing the S channel image according to a threshold range to obtain a binary image of the bone marrow cells, and performing morphological processing on the binary image;

b3, extracting edge pixel points of the bone marrow cell image after morphological processing by using a connected domain method, finding edge pixel points on the upper side, the lower side, the left side and the right side of the bone marrow cell, and then dividing the bone marrow cell;

b4, selecting the images of the segmented bone marrow cells, inputting the cell images with obvious characteristics in each class as training cells into a depth residual error network, and training the network;

b5, taking the remaining cells after selection as test cells, and grading the test cells by using a softmax classifier; if the maximum score is greater than or equal to a set threshold, classifying the score into a certain class; if the maximum score is smaller than a set threshold, classifying the score into a sub-classification;

b6, taking any pixel point at the edge of the training cell and the cell in the sub-classification in the step S5 as a pole, establishing a polar coordinate system, and mapping all pixel points into a rectangular coordinate system one by using polar coordinate transformation;

b7, traversing edge pixel points of the training cells, wherein each pixel point generates a converted image, and each image has n pixel points, namely n times of conversion; for cells in the sub-classification, each image is transformed only once;

b8, the image after the training cell transformation is used as the input of the sub-classification network, the deep residual error network is retrained, and the network parameters are stored; taking the image after cell transformation in the sub-classification as test data, and scoring again by using a softmax classifier: if the maximum score of the cell image is greater than or equal to a set threshold, classifying the cell image into a certain subcategory; if the maximum score is less than a set threshold, the cell image is classified as an unclassified cell.

9. The method of claim 8, wherein the step of performing artificial intelligence-aided hierarchical diagnosis on the bone marrow cell image comprises: if the proportion of the unclassified cells is greater than the set unclassified threshold, then the unclassified cells are segmented and classified, and the method comprises the following steps of C1, taking edge pixel points of a bone marrow cell image after morphological processing for the classified cells, finding edge pixel points above, below, left and right the bone marrow cells, and segmenting the bone marrow cells;

c2, inputting the segmented marrow cell image as a training cell into a depth residual error network, and training the network;

c3, scoring the test cells by using a softmax classifier, and if the maximum score is greater than or equal to a set threshold value, classifying the test cells into a certain class; if the maximum fraction is smaller than a set threshold value, classifying the cells into finally unclassified cells;

and C4, if the ratio of the final unclassified cells is less than the set unclassified threshold value, stopping, and if the ratio of the final unclassified cells is more than the set unclassified threshold value, continuing the step C1 until the ratio of the final unclassified cells is less than the set unclassified threshold value.

Images