Bone marrow cell detection and recognition system

Abstract

The invention relates to a cell detection technology, in particular to a bone marrow cell detection and recognition system which integrates a slice image acquisition module, an image classification module, a real-time detection module, a cell data analysis module and the like. According to the scheme, automatic analysis of cell density, completeness and staining quality is carried out on the bone marrow smear image through an image processing technology, target detection is carried out by introducing the YOLO model, accurate identification and counting of cells in the image with complete cell morphology and uniform staining are realized, an alarm is given out in time when the cell number exceeds the preset threshold value, and the accuracy of cell detection is improved. While the automation level of bone marrow cell detection is improved, the reliability of a detection result and the clinical application value are enhanced.

Description

Technical Field

[0001] The present invention relates to cell detection technology, and specifically to a bone marrow cell detection and recognition system. Background Art

[0002] Bone marrow cell detection plays a crucial role in medical diagnosis and treatment, especially in the diagnosis and treatment of blood diseases. Traditional bone marrow cell detection methods mainly rely on manual observation and analysis under a microscope. This method is not only time-consuming and laborious, but also easily affected by the operator's experience and subjective judgment, making it difficult to ensure the accuracy and consistency of detection results. With the development of computer vision and deep learning technologies, automated bone marrow cell detection systems based on image processing and machine learning have gradually become a research hotspot. These systems can greatly improve the detection efficiency and accuracy by automatically collecting, classifying, and analyzing bone marrow smear images, and reducing human errors. However, existing automated detection systems still have certain limitations in aspects such as cell density analysis, cell integrity assessment, staining quality analysis, and accurate identification of target cells. Therefore, there is an urgent need to develop a bone marrow cell detection and recognition system that integrates multiple modules such as slice image acquisition, image classification, real-time detection, and cell data analysis. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a bone marrow cell detection and recognition system to solve the technical problem that the existing bone marrow cell detection is affected by the operator's experience and subjective judgment, resulting in difficult to ensure the accuracy and consistency of detection results.

[0004] The basic solution provided by the present invention: A bone marrow cell detection and recognition system, including a slice image acquisition module, an image classification module, a real-time detection module, and a cell data analysis module;

[0005] The slice image acquisition module is used to collect bone marrow smear images under a microscope;

[0006] The image classification module includes a cell density analysis module, a cell integrity analysis module, and a staining quality analysis module; The cell density analysis module divides the bone marrow smear image into a cell evenly distributed picture and a cell unevenly distributed picture according to a preset density interval threshold; The cell integrity analysis module evaluates the clarity of the cell contour through image processing technology, and divides the cell evenly distributed picture into a cell morphologically complete picture and a cell morphologically incomplete picture according to a preset clarity threshold; The staining quality analysis module is used to calculate the color mean and standard deviation of the cell morphologically complete picture, and divides the cell morphologically complete picture into a cell evenly stained picture and a cell unevenly stained picture according to a preset color uniformity threshold;

[0007] The real-time detection module is used to perform object detection on cell staining uniform pictures through the YOLO model, identify the cells in the target pictures, and generate single-cell pictures; the cell data analysis module is used to count and statistically analyze the identified various types of cells, and send corresponding cell quantity abnormality warnings when the quantity of various types of cells exceeds the preset quantity threshold range.

[0008] Furthermore, the network architecture of the YOLO model includes an input end, a baseline network, a neck network, and an output layer;

[0009] The input end adopts the Mosaic data augmentation technology to enhance the picture background and improve the detection effect of small targets by randomly cropping, scaling, and merging four pictures;

[0010] The baseline network adopts the CBS structure, C3 structure, and SPPF structure to extract object information in the image;

[0011] The neck network adopts the Feature Pyramid Network and Path Aggregation Network structures to enhance the network robustness;

[0012] The output layer uses CIOU Loss as the bounding box loss function and uses DIOU nms for non-maximum suppression to optimize the selection of detection boxes.

[0013] Furthermore, it also includes a model training module, which is used to train the YOLO model on the labeled cell picture dataset.

[0014] Furthermore, the slice image acquisition module obtains bone marrow smear images through a microscope camera.

[0015] Furthermore, it also includes a preprocessing module, which is used to perform normalization processing and image enhancement on the collected bone marrow smear images.

[0016] Furthermore, the cell data analysis module includes a cell counting module, a cell statistical report module, an anomaly detection module, and a warning module;

[0017] The cell counting module is used to traverse the cell classification results and count the number of cells in each category; the cell statistical report module is used to generate a classification statistical report of each type of cell according to the statistical results, including the proportion and quantity information of each type of cell;

[0018] The anomaly detection module is used to monitor whether the cell quantity is within the preset quantity threshold range of each type of cell. If it exceeds the quantity threshold range, a warning message will be sent to the user through the warning module.

[0019] Furthermore, it also includes a sample acquisition module, and the sample acquisition module includes a sample collection module, an image acquisition module, and a sample annotation module;

[0020] The sample collection module is used to obtain blood samples and record the corresponding personal information;

[0021] The image acquisition module scans and takes pictures of the blood sample through an existing automated blood cell imaging system to obtain blood cell pictures;

[0022] The sample annotation module is used to annotate the blood cell pictures, distinguish different types of blood cells and abnormal cell morphologies, and obtain an annotated cell picture data set.

[0023] Furthermore, the sample annotation module further includes a desensitization module, which is used to perform data desensitization processing on the blood cell pictures and remove or replace personal information including the name and ID number of the sampled person.

[0024] The principle and advantages of the present invention are as follows: The bone marrow cell detection and recognition system proposed by the present invention significantly improves the efficiency and accuracy of bone marrow cell detection by integrating multiple modules such as section image acquisition, image classification, real-time detection, and cell data analysis. The system uses image processing technology to automatically analyze the cell density, integrity, and staining quality of bone marrow smear images, effectively reducing human errors and improving the consistency of detection. By introducing the YOLO model for target detection, accurate identification and counting of cells in pictures with complete cell morphology and uniform staining are achieved, further optimizing the detection process. In addition, the system has an abnormal detection function, which can issue a warning in a timely manner when the number of cells exceeds a preset threshold, providing an important reference for clinical diagnosis. Overall, the system not only improves the automation level of bone marrow cell detection, but also enhances the reliability of the detection results and the clinical application value. Brief Description of the Drawings

[0025] Figure 1 It is a logic block diagram of an embodiment of a bone marrow cell detection and recognition system of the present invention. Detailed Description of the Invention

[0026] The following is a further detailed description through specific embodiments:

[0027] The specific implementation process is as follows:

[0028] Embodiment 1

[0029] Embodiment 1 is basically as shown in the appendix Figure 1As shown in the figure, a bone marrow cell detection and recognition system includes a section image acquisition module, a preprocessing module, a model training module, an image classification module, a real-time detection module, and a cell data analysis module. Through image processing technology, automatic analysis of cell density, integrity, and staining quality of bone marrow smear images is carried out. Then, by introducing the YOLO model for object detection, accurate recognition and counting of cells in images with complete cell morphology and uniform staining are achieved, and a warning is issued in a timely manner when the number of cells exceeds a preset threshold, improving the automation level of bone marrow cell detection while enhancing the reliability and clinical application value of the detection results.

[0030] Specifically, the section image acquisition module in this embodiment uses a high-resolution microscope camera to acquire bone marrow smear images, and is also equipped with an LED light source with adjustable brightness and color temperature to meet the requirements of different types of staining and observation.

[0031] The preprocessing module includes an image normalization module, an image enhancement module, a noise removal module, and a size normalization module. The image normalization module is used to scale the image pixel values from the original range (usually 0 - 255) to a more appropriate range, such as between 0 and 1. This helps to reduce the impact of illumination changes on image quality. The image enhancement module is used to enhance the image contrast through methods such as histogram equalization, making the cell structure more clearly visible. The noise removal module uses methods such as Gaussian filtering and median filtering to remove random noise in the image while keeping the edges of the cell structure from being blurred. The size normalization module is used to crop off unimportant parts of the image edges as needed and scale the image to a unified size to meet the requirements of subsequent processing or model input.

[0032] The image classification module includes a cell density analysis module, a cell integrity analysis module, and a staining quality analysis module; the cell density analysis module divides the bone marrow smear image into pictures with uniform cell distribution and pictures with non-uniform cell distribution according to the preset density interval threshold; the cell integrity analysis module evaluates the clarity of the cell contour through image processing technology and divides the pictures with uniform cell distribution into pictures with complete cell morphology and pictures with incomplete cell morphology according to the preset clarity threshold; the staining quality analysis module is used to calculate the color mean and standard deviation of the pictures with complete cell morphology and divides the pictures with complete cell morphology into pictures with uniform cell staining and pictures with non-uniform cell staining according to the preset color uniformity threshold.

[0033] Specifically, in this embodiment, the cell density analysis module uses a threshold-based segmentation method (such as the Otsu method) to divide the bone marrow smear image into multiple small blocks. In this embodiment, the cells in each small block are counted and classified according to a preset density interval threshold. For example, the following thresholds are set in this embodiment: low density: the number of cells per small block is less than 10; medium density: the number of cells per small block is between 10 and 50; high density: the number of cells per small block is greater than 50. Each small block is marked as low, medium, or high density. In this embodiment, the blocks with medium density can be classified as pictures with uniform cell distribution; thus, the corresponding classification images are generated.

[0034] In the cell integrity analysis module, an edge detection algorithm is applied to extract the cell contour. In this embodiment, the Canny edge detector is used. The integrity of the cell morphology is evaluated by calculating the clarity score of the contour. In this embodiment, the ratio of the length to the area of the contour is used as the clarity index. A clarity threshold of 0.7 is set. When the clarity score is lower than this threshold, the cell morphology is considered incomplete. The cells with complete morphology and the cells with incomplete morphology are marked respectively, and the pictures of cells with complete morphology and the pictures of cells with incomplete morphology are generated.

[0035] In the staining quality analysis module, the pictures of cells with complete morphology are converted from the RGB color space to the HSV color space to better analyze the color information, and then their color means and standard deviations are calculated. Whether the staining is uniform is judged according to the color uniformity threshold. For example, in this embodiment, the staining is uniform when the color standard deviation is less than 20, and the staining is non-uniform when the color standard deviation is greater than or equal to 20. Thus, the pictures of cells with complete morphology are divided into the pictures of cells with uniform staining and the pictures of cells with non-uniform staining.

[0036] After obtaining the pictures of cells with uniform staining, the real-time detection module inputs the pictures of cells with uniform staining into the pre-trained YOLO model to perform object detection on the pictures of cells with uniform staining, identifies the cells in the target pictures, and generates single-cell pictures. The YOLO model is obtained by training on the labeled cell picture dataset through the model training module.

[0037] Specifically, the network architecture of the YOLO model in this embodiment includes an input end, a baseline network, a neck network, and an output layer. The input end adopts the Mosaic data augmentation technique to enhance the image background and improve the detection effect of small targets by randomly cropping, scaling, and merging four pictures. In addition, the adaptive anchor box calculation and adaptive image scaling techniques are applied to optimize the training and detection performance of the model. The baseline network adopts the CBS structure, C3 structure, and SPPF structure to extract object information in the image. These structures help to extract object information in the image and provide effective features for subsequent object detection. The neck network adopts the Feature Pyramid Network and Path Aggregation Network structures to enhance the network robustness. The output layer uses CIOU Loss as the bounding box loss function and uses DIOU nms for non-maximum suppression to optimize the selection of detection boxes.

[0038] The cell data analysis module includes a cell counting module, a cell statistical report module, an anomaly detection module, and a warning module. The cell counting module is used to traverse the cell classification results and count the number of cells in each category; the cell statistical report module is used to generate a classification statistical report of each type of cell according to the statistical results, including the proportion and quantity information of each type of cell. The anomaly detection module is used to monitor whether the cell quantity is within the preset quantity threshold range of each type of cell. If it exceeds the quantity threshold range, a warning message is sent to the user through the warning module.

[0039] In addition, a sample acquisition module is also included in this embodiment. The sample acquisition module includes a sample collection module, an image acquisition module, and a sample annotation module. The sample collection module is used to collect blood samples from a number of sampled persons and record the personal information including age, gender, and health status for subsequent data analysis and clinical correlation. The image acquisition module is used to scan and photograph the collected blood samples through a high-resolution microscope or an existing automated blood cell imaging system to obtain high-quality blood cell pictures. The sample annotation module in this embodiment annotates the obtained blood cell pictures by experienced hematologists or pathologists to clearly distinguish different types of blood cells (such as red blood cells, white blood cells, platelets, etc.) and abnormal cell morphologies (such as abnormal cells, dysplastic hematopoietic cells, etc.). In addition, the sample annotation module also includes a desensitization module, which is used to perform data desensitization processing on the blood cell pictures to remove or replace the personal information including the name and ID number of the sampled person.

[0040] Specifically, the labeling method includes selecting at least two groups of blood cell identification personnel, with at least two people in each group for blood cell identification. When the recognition consistency rate of each collected blood cell information is greater than the set threshold, it is determined that the manual recognition result passes. When the recognition consistency rate of each collected blood cell information is less than the set threshold, it is determined that the manual recognition result fails, and another group of blood cell identification personnel is required to conduct re-identification. When the recognition consistency rate is greater than the set threshold, it is determined that the manual recognition result passes. When the recognition consistency rate is less than the set threshold, the blood cell information is determined to be invalid information. Thus, the labeled cell picture dataset is obtained.

[0041] In summary, this solution automatically analyzes the cell density, integrity, and staining quality of bone marrow smear images through image processing technology, and then introduces the YOLO model for target detection, realizing the accurate recognition and counting of cells in pictures with complete cell morphology and uniform staining, and promptly issuing a warning when the cell quantity exceeds the preset threshold. While improving the automation level of bone marrow cell detection, it also enhances the reliability and clinical application value of the detection results.

[0042] The above are only the embodiments of the present invention. Common knowledge such as the specific structures and characteristics in the solution is not described in detail here. Those of ordinary skill in the art know all the common technical knowledge in the technical field to which the invention belongs before the application date or the priority date, can know all the existing technologies in this field, and have the ability to apply the conventional experimental means before this date. Those of ordinary skill in the art can, under the inspiration given in this application, combine their own abilities to improve and implement this solution. Some typical well-known structures or well-known methods should not become an obstacle for those of ordinary skill in the art to implement this application. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can still be made, and these should also be regarded as the protection scope of the present invention, which will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be based on the content of its claims, and the specific implementation manners described in the specification can be used to interpret the content of the claims.

Claims

1. A bone marrow cell detection and identification system, characterized in that: It includes a slice image acquisition module, an image classification module, a real-time detection module and a cell data analysis module; The slice image acquisition module is used to acquire bone marrow smear images under a microscope; The image classification module includes a cell density analysis module, a cell integrity analysis module and a staining quality analysis module; the cell density analysis module divides the bone marrow smear image into a cell uniform distribution picture and a cell uneven distribution picture according to a preset density interval threshold; the cell integrity analysis module evaluates the cell contour clarity through image processing technology, and divides the cell uniform distribution picture into a cell morphology complete picture and a cell morphology incomplete picture according to a preset clarity threshold; The staining quality analysis module is used to calculate the color mean and standard deviation of the complete cell morphology picture, and divide the complete cell morphology picture into a cell staining uniform picture and a cell staining uneven picture according to a preset color uniformity threshold; The real-time detection module is used to perform target detection on the cell uniformly stained picture through the YOLO model, identify the cells in the target picture and generate a single cell picture; the cell data analysis module is used to count and count the identified cells of each type, and issue a corresponding cell quantity abnormality warning when the number of cells of each type exceeds the preset quantity threshold range.

2. A bone marrow cell detection and identification system according to claim 1, characterized in that: The network architecture of the YOLO model includes an input end, a reference network, a neck network, and an output layer; The input end uses Mosaic data enhancement technology to enhance the image background and improve the small target detection effect by randomly cropping, scaling and merging four images; The benchmark network uses CBS structure, C3 structure and SPPF structure to extract object information in the image; The neck network adopts a feature pyramid network and a path aggregation network structure to enhance network robustness; The output layer uses CIOU Loss as the bounding box loss function and uses DIOU nms for non-maximum suppression to optimize the selection of the detection box.

3. A bone marrow cell detection and identification system according to claim 2, characterized in that: It also includes a model training module, which is used to train the YOLO model on the labeled cell image data set.

4. A bone marrow cell detection and identification system according to claim 3, characterized in that: The slice image acquisition module acquires the bone marrow smear image through a microscope camera.

5. A bone marrow cell detection and identification system according to claim 4, characterized in that: It also includes a preprocessing module, which is used to perform normalization processing and image enhancement on the collected bone marrow smear image.

6. A bone marrow cell detection and identification system according to claim 5, characterized in that: The cell data analysis module includes a cell measurement module, a cell statistics reporting module, an abnormality detection module and a warning module; The cell measurement module is used to traverse the cell classification results and count the number of cells of each category; the cell statistical report module is used to generate a classification statistical report of each type of cells according to the statistical results, including the proportion and quantity information of each type of cells; The abnormality detection module is used to monitor whether the number of cells is within a preset number threshold range of each type of cells. If the number threshold range is exceeded, a warning message is sent to the user through the warning module.

7. A bone marrow cell detection and identification system according to claim 3, characterized in that: It also includes a sample acquisition module, which includes a sample collection module, an image acquisition module and a sample annotation module; The sample collection module is used to obtain blood samples and record corresponding personnel information; The image acquisition module scans and photographs the blood sample through an existing automated blood cell imaging system to obtain a blood cell image; The sample annotation module is used to annotate blood cell images, distinguish different types of blood cells and abnormal cell morphology, and obtain annotated cell image data sets.

8. A bone marrow cell detection and identification system according to claim 7, characterized in that: The sample labeling module also includes a desensitizing module, which is used to perform data desensitization processing on the blood cell image, removing or replacing personal information including the name and ID number of the sampled person.

Images