MYH9-RD auxiliary screening system and method based on blood cell morphology

Abstract

The invention discloses an MYH9-RD auxiliary screening system and an MYH9-RD auxiliary screening method based on blood cell morphology. Comprising an image acquisition module, an image detection module and an image analysis module, wherein the image acquisition module is responsible for acquiring a Wright stained peripheral blood image of a case to be screened; the image detection module is responsible for acquiring an independent image of each blood cell; the image analysis module is responsible for acquiring blood cell category information, leukocyte form information and platelet size information; the counting and processing module is responsible for generating an analysis result of a case to be screened; the related disease judgment module is responsible for generating an MYH9-RD screening result; and the result output module is responsible for outputting the screening result. According to the method, the morphology of the special cells of the peripheral blood is accurately identified and the number of the cells is accurately calculated through an artificial intelligence algorithm, so that the defects of incomplete visual capture and easy fatigue of human beings are overcome, misdiagnosis and missed diagnosis caused by personal errors or subjective differences are reduced, accurate identification of the cell morphology is realized, and the working efficiency is greatly improved; and early, rapid and efficient screening of the MYH9-RD is facilitated.

Description

Technical Field

[0001] The present invention belongs to the field of medical artificial intelligence technology, and specifically relates to a MYH9-RD auxiliary screening system and method based on blood cell morphology. Background Art

[0002] Nonmuscle myosin heavy chain 9-related diseases (MYH9-RD) are rare autosomal dominant inherited diseases caused by mutations in the MYH9 gene encoding nonmuscle myosin heavy chain ⅡA (NMMHC-ⅡA) located on human chromosome 22q12.3. The typical hematological manifestations of MYH9-RD are the "triad" of thrombocytopenia, large platelets and abnormal granulocyte inclusions, and varying degrees of bleeding, which may be accompanied by extrahematological manifestations such as nephritis, deafness, cataracts and abnormal liver function. Therefore, it is described in pubmed as "granulocyte inclusions and large thrombocytopenia with or without nephritis or sensorineural hearing loss."

[0003] MYH9-RD has a high degree of clinical phenotypic heterogeneity, lacks specific clinical manifestations and screening methods, and is difficult to diagnose clinically. Due to the complexity and diversity of the sites of MYH9 gene mutations, the bleeding symptoms and associated extra-hematological symptoms are different, and clinicians generally lack awareness of this disease, which often leads to misdiagnosis and missed diagnosis. MYH9-RD patients are often misdiagnosed as primary immune or idiopathic thrombocytopenia due to spontaneous bleeding or traumatic hematoma such as "epistaxis, petechiae and ecchymoses", and low platelet count. They are misused with ineffective, inappropriate or even harmful treatments such as hormones, immunoglobulin, or even splenectomy, which increases treatment-related adverse reactions and medical burdens. Therefore, early screening and diagnosis of MYH9-RD are of great significance for precise treatment, improved prognosis and genetic counseling, and avoiding potentially dangerous inappropriate treatments.

[0004] Currently, the definitive diagnosis of MYH9-RD relies on genetic testing. The commonly used chip capture high-throughput sequencing + Sanger verification, the experimental steps are: extract genomic DNA from the patient's peripheral venous blood, first break the DNA and prepare the library, then use the chip to capture and enrich the DNA in the MYH9 gene coding region and adjacent splicing region, and finally use a high-throughput sequencing platform for mutation detection. After determining the mutation site, Sanger verification is performed on the proband and family members. This technology requires high-end instruments and professional bioinformatics analysis technicians, which are not available in general laboratories. It is expensive (about 2,000 yuan per specimen) and the experimental time is long (about 2 weeks).

[0005] Routine blood analysis and blood smear microscopy are the first steps and important screening methods for diagnosing MYH9-RD. At present, conventional laboratories use blood cell analyzers to count and classify blood cells, which cannot identify subtle morphological features such as abnormal granulocyte inclusions. Platelet counting often uses the impedance method, which has many interfering factors. For example, when the platelet size is large, it is easy to misjudge it as red blood cells, and large platelets cannot be accurately identified, resulting in inaccurate platelet count results. Therefore, the morphological identification and counting of cell characteristics such as granulocyte inclusions and large platelets mainly rely on manual microscopy of blood smears. The experimental steps are: after the blood smear is pushed and stained with Wright's stain, an optical microscope is used to observe the platelet size, the proportion of large platelets, the presence or absence of granulocyte inclusions, the shape, number, size, staining intensity and position of the inclusions, and other morphological characteristics (first low-power observation, then high-power counting of 200 white blood cells). This technology relies on professional physicians to observe and identify with the naked eye under a microscope, which is limited by personal experience and subjective judgment, and is time-consuming and inefficient. It is the main constraint for early screening and clinical diagnosis.

[0006] In recent years, with the rapid development of artificial intelligence (AI) technology in the medical field, the combination of AI technology and blood cell morphology is expected to achieve beneficial results in the AI ​​screening technology of MYH9-RD, realize the early, rapid and efficient screening of MYH9-RD, improve the screening accuracy and work efficiency, and reduce the intensity and pressure of clinical work. For example, the patent application with application number CN202310523263.X discloses an intelligent blood cell analysis method and system based on multimodal microscopic imaging, but the AI ​​part of the application is limited to the identification and classification of cells from a morphological perspective, and does not perform AI calculations on the number and proportion of cells, and does not use a method based on the fusion of knowledge reasoning and clinical judgment to generate relevant disease screening results. Summary of the invention

[0007] In response to a series of problems existing in the above-mentioned prior art, the present invention provides a MYH9-RD auxiliary screening system and method based on blood cell morphology, which uses an artificial intelligence algorithm to accurately identify and count "large platelets" and "inclusion body neutrophils" in peripheral blood smear Wright staining to achieve high-accuracy MYH9-RD auxiliary screening.

[0008] In order to solve the above technical problems, the present invention is implemented through the following technical solutions:

[0009] A MYH9-RD auxiliary screening system based on blood cell morphology, comprising:

[0010] The image acquisition module is responsible for obtaining Wright-stained peripheral blood images of the cases to be screened;

[0011] An image detection module, responsible for acquiring a separate image of each blood cell from the Wright-stained peripheral blood image and generating a collection of single blood cell images;

[0012] An image analysis module is responsible for obtaining blood cell category information, white blood cell morphology information and platelet size information from the single blood cell image collection, and generating a blood cell classification result;

[0013] The counting and processing module is responsible for counting and processing the data of various types of blood cells in the blood cell classification results, and generating analysis results of the cases to be screened;

[0014] The relevant disease judgment module is responsible for judging whether the case to be screened is MYH9-RD based on the analysis results of the case to be screened, using a method based on knowledge reasoning and clinical judgment fusion, and generating a MYH9-RD screening result;

[0015] The result output module is responsible for outputting the MYH9-RD screening results.

[0016] Furthermore, the Wright-stained peripheral blood image is a microscopic image of the Wright-stained peripheral blood of the case to be screened, which is collected using a high-power microscope and an oil immersion lens.

[0017] Furthermore, the image detection module is embedded with a deep network detection model. The image detection module is firstly responsible for inputting the acquired Wright-stained peripheral blood image into the trained deep network detection model, and the deep network detection model processes the input Wright-stained peripheral blood image to generate a separate image of each blood cell, and then is responsible for integrating the generated separate image of each blood cell into a collection of single blood cell images.

[0018] Furthermore, the deep network detection model adopts the YOLOv5 model, whose network structure consists of a backbone network, a neck network and three prediction heads; wherein,

[0019] The backbone network, including Conv, CSPDarkNet53 and SPPF modules, is used to extract and integrate features from the input image and generate a feature map;

[0020] The neck network is used to fuse the feature maps generated by the backbone network; it is combined with a feature pyramid network (FPN) and a path aggregation network (PANet) to enhance semantic information and multi-scale positioning capabilities;

[0021] The three prediction heads are used to detect objects of different sizes, generate candidate bounding boxes with different aspect ratios, and use non-maximum suppression (NMS) to output the final bounding box with position, size and confidence score.

[0022] Furthermore, the image analysis module is composed of a blood cell image classification module, a white blood cell image classification module and a platelet image classification module, and the blood cell image classification module is equipped with a deep network classification model, the white blood cell image classification module is equipped with a white blood cell single classification model, and the platelet image classification module is equipped with a platelet size classification model;

[0023] The image analysis module is first responsible for inputting the acquired single blood cell image collection into the trained deep network classification model, and the deep network classification model classifies the single blood cell image collection to distinguish the white blood cell images, red blood cell images and platelet images therein, and generates blood cell category information;

[0024] Then, the set of distinguished white blood cell images is input into the trained white blood cell single classification model, and the white blood cell single classification model classifies the set of white blood cell images to distinguish the neutrophil images containing abnormal inclusion bodies (abbreviated as "inclusion body neutrophil" images) and the neutrophil images without abnormal inclusion bodies (abbreviated as "non-inclusion body neutrophil" images), and generates white blood cell morphology information;

[0025] At the same time, it is responsible for inputting the set of differentiated platelet images into the trained platelet size classification model, and the platelet size classification model performs platelet segmentation processing (traditional image processing method) on the set of platelet images to obtain the size of each platelet therein and generate platelet size information;

[0026] Finally, it is responsible for integrating the generated blood cell category information, white blood cell morphology information and platelet size information into blood cell analysis results.

[0027] Furthermore, the deep network classification model and the single white blood cell classification model both adopt the ResNet-18 model, whose network structure consists of multiple residual blocks with convolutional layers and downsampling layers, average pooling layers and fully connected layers; each residual block contains a shortcut connection for adding the input of a block to its output to achieve gradient flow and simplify the optimization process.

[0028] Furthermore, the counting and processing module will count all red blood cell images, white blood cell images, and platelet images, and will also separately count the "inclusion neutrophil" images and "non-inclusion neutrophil" images in the white blood cell images, and will also separately count the large platelet images and normal-sized platelet images in the platelet images, and calculate the proportion of large-sized platelet images in all platelet images, thereby ultimately generating the analysis results of the cases to be screened through data processing.

[0029] Furthermore, the related disease judgment module includes a feature extraction module and a case classification module; wherein,

[0030] The feature extraction module is responsible for extracting characteristic cell morphologies related to MYH9-RD, including the number of “large platelets”, the proportion of “large platelets” and the number of “inclusion body neutrophils” from the analysis results of the cases to be screened using regular expressions, thereby obtaining the characteristics of MYH9-RD cases;

[0031] The case classification module is responsible for judging whether the case to be screened is MYH9-RD based on the MYH9-RD case characteristics by a method based on knowledge reasoning and clinical judgment fusion, and generating a MYH9-RD screening result, such as a "suspected MYH9-RD" initial screening result or a "non-MYH9-RD" initial screening result.

[0032] A control method of a MYH9-RD auxiliary screening system, which is not intended to directly diagnose MYH9-RD but is only used to assist in the early screening of MYH9-RD, and comprises the following steps:

[0033] Step 1) obtaining a Wright-stained peripheral blood image of the case to be screened by an image acquisition module;

[0034] Step 2) the image detection module inputs the Wright-stained peripheral blood image into a trained deep network detection model for processing to obtain a separate image of each blood cell in the Wright-stained peripheral blood image, and integrates the images into a collection of single blood cell images;

[0035] Step 3) the image analysis module inputs the single blood cell image collection into a trained deep network classification model for classification processing to distinguish the white blood cell images, red blood cell images and platelet images in the single blood cell image collection, and generates blood cell category information;

[0036] Step 4) the image analysis module inputs the collection of the distinguished white blood cell images into the trained white blood cell single classification model for classification processing to distinguish the "inclusion neutrophil" images and the "non-inclusion neutrophil" images in the collection of white blood cell images, and generates white blood cell morphology information;

[0037] Step 5) the image analysis module inputs the set of the differentiated platelet images into a trained platelet size classification model for platelet segmentation processing (traditional image processing method) to obtain the size of each platelet in the set of platelet images and generate platelet size information;

[0038] Step 6) the image analysis module integrates the blood cell category information, the white blood cell morphology information and the platelet size information to generate a blood cell classification result;

[0039] Step 7) The counting and processing module counts and processes the various types of blood cells in the blood cell classification results, and generates analysis results of the cases to be screened;

[0040] Step 8) The relevant disease judgment module uses regular expressions to extract characteristic cell morphologies related to MYH9-RD, including the number of "large platelets", the proportion of "large platelets" and the number of "inclusion body neutrophils", from the analysis results of the cases to be screened, thereby obtaining the characteristics of MYH9-RD cases;

[0041] Step 9) The relevant disease judgment module determines whether the case to be screened is MYH9-RD based on the characteristics of the MYH9-RD case by a method based on knowledge reasoning and clinical judgment fusion, and generates a MYH9-RD screening result;

[0042] Step 10) The MYH9-RD screening result is output by a result output module.

[0043] It should be specially noted that the MYH9-RD auxiliary screening system and control method thereof proposed in the present invention are not diagnostic tools and methods that can directly diagnose MYH9-RD, and their original intention of development is not to directly diagnose MYH9-RD, but to serve as auxiliary screening tools and methods to help professional physicians make early diagnosis of MYH9-RD.

[0044] That is to say, the MYH9-RD screening results obtained by the MYH9-RD auxiliary screening system and control method of the present invention cannot be directly used as the MYH9-RD diagnostic conclusion for the case to be screened, but can only be used as one of the reference bases for professional physicians to diagnose MYH9-RD, or more accurately, as an early screening basis. Only after confirming the MYH9-RD screening results will the professional physician be given an early warning that the test case may be MYH9-RD, and then the professional physician will make a final diagnosis of the test case through the MYH9-RD gene detection method and give a clear diagnosis conclusion.

[0045] A computer device includes: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus, and the memory is used to store at least one executable instruction, wherein the executable instruction enables the processor to execute operations corresponding to the control method of the above-mentioned MYH9-RD auxiliary screening system.

[0046] A computer storage medium, wherein at least one executable instruction is stored in the computer storage medium, and the executable instruction enables a processor to execute operations corresponding to the control method of the above-mentioned MYH9-RD auxiliary screening system.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention uses artificial intelligence algorithms to accurately identify and count the special cell morphologies in the peripheral blood of disease patients (Wright-stained "large platelets" and "inclusion body neutrophils" in peripheral blood smears), and can extract deep features of cell images that are difficult for humans to identify, make up for the shortcomings of incomplete human visual capture and easy fatigue, reduce tiny lesions missed by the naked eye, reduce misdiagnosis and missed diagnosis caused by human errors or subjective differences, achieve accurate identification of cell morphology and greatly improve work efficiency.

[0049] The present invention has the advantages of early, simple, economical and accurate screening, and can greatly improve the accuracy of early screening of MYH9-RD. The screening results can be used as a reference to assist hematologists in making early diagnosis of MYH9-RD cases, which helps to alleviate the problems of shortage of hematologists, low diagnostic level of grassroots personnel, low work efficiency, and uneven distribution of medical resources.

[0050] Compared with the existing MYH9-RD gene diagnosis technology (chip capture high-throughput sequencing + Sanger verification method), the present invention does not require expensive instruments and equipment and specialized bioinformatics analysis technicians, but only requires an ordinary computer, with no additional costs, fast speed (measured in seconds), simple and clear result output, and is suitable for grassroots laboratories.

[0051] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. The specific implementation of the present invention is given in detail by the following embodiments and their accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0053] Figure 1 This is a system framework diagram of the MYH9-RD auxiliary screening system based on blood cell morphology of the present invention;

[0054] Figure 2 This is a schematic diagram of the network structure of the YOLOv5 model used in the present invention;

[0055] Figure 3A set of images of neutrophils containing abnormal inclusion bodies distinguished by the present invention;

[0056] Figure 4 A set of images of neutrophils without abnormal inclusion bodies distinguished by the present invention;

[0057] Figure 5 A set of images of large platelets segmented by the present invention;

[0058] Figure 6 An image set of normal-sized platelets segmented by the present invention;

[0059] Figure 7 This is a schematic diagram of the network structure of the ResNet-18 model used in the present invention;

[0060] Figure 8 The present invention is a flowchart of the steps of the control method of the MYH9-RD auxiliary screening system based on blood cell morphology.

[0061] Fig. 9 This is a diagram of a normal human blood smear test process in an embodiment of the present invention; Fig. 9 a shows the prepared normal human Wright-stained peripheral blood smear. Fig. 9 b shows the whole scan image of normal Wright-stained peripheral blood smear. Fig. 9 c shows the normal human platelet AI test results, Fig. 9 d shows the AI ​​test results of normal human leukocytes.

[0062] Fig.10 This is a screenshot of the auxiliary screening results of the normal human blood smear test in an embodiment of the present invention;

[0063] Fig.11 This is a diagram of the blood smear test process of MYH9-RD patients in an embodiment of the present invention; wherein Fig.11 a shows the prepared Wright-stained peripheral blood smear of MYH9-RD patients. Fig.11 b shows the whole scan image of the Wright-stained peripheral blood smear of a MYH9-RD patient. Fig.11 c shows the platelet AI test results of MYH9-RD patients. Fig.11 d shows the results of leukocyte AI detection in MYH9-RD patients.

[0064] Fig.12 This is a screenshot of the auxiliary screening results of the blood smear test of MYH9-RD patients in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings so that the purpose, features and advantages of the invention can be more clearly understood. It should be understood that the embodiments shown in the accompanying drawings are not intended to limit the scope of the present invention, but are only intended to illustrate the essential spirit of the technical solution of the present invention.

[0066] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the relevant art will recognize that the embodiments may be practiced without one or more of these specific details. In other cases, well-known devices, structures, and techniques associated with the present application may not be shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0067] Unless the context requires otherwise, throughout the specification and claims, the word "comprise" and variations such as "include" and "have" should be construed in an open, inclusive sense, ie, should be interpreted as "including, but not limited to."

[0068] References throughout the specification to "one embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of "in one embodiment" or "in an embodiment" in various places throughout the specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any manner in one or more embodiments.

[0069] As used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It should be noted that the term "or" is generally employed in its sense including "and / or" unless the context clearly dictates otherwise.

[0070] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0071] See also Figure 1 As shown, a MYH9-RD auxiliary screening system based on blood cell morphology mainly includes an image acquisition module, an image detection module, an image analysis module, a counting and processing module, a related disease judgment module and a result output module which are connected in sequence.

[0072] 1. The image acquisition module is responsible for obtaining Wright-stained peripheral blood images of the cases to be screened.

[0073] The Wright-stained peripheral blood image is a microscopic image of the Wright-stained peripheral blood of the case to be screened, which is collected using a high-power microscope and an oil immersion lens.

[0074] 2. The image detection module is responsible for obtaining a separate image of each blood cell from the Wright-stained peripheral blood image and generating a collection of single blood cell images.

[0075] Specifically, the image detection module is embedded with a deep network detection model. The image detection module is firstly responsible for inputting the acquired Wright-stained peripheral blood image into the trained deep network detection model, and the deep network detection model processes the input Wright-stained peripheral blood image to generate a separate image of each blood cell, and then is responsible for integrating the generated separate image of each blood cell into a collection of single blood cell images.

[0076] As a preferred embodiment of the present invention, see Figure 2 As shown, the deep network detection model adopts the YOLOv5 model, and its network structure consists of a backbone network, a neck network and three prediction heads.

[0077] The backbone network, including Conv, CSPDarkNet53 and SPPF modules, is used to extract and integrate features from the input image and generate a feature map.

[0078] The neck network is used to fuse the feature maps generated by the backbone network; it is combined with a feature pyramid network (FPN) and a path aggregation network (PANet) to enhance semantic information and multi-scale positioning capabilities.

[0079] The three prediction heads are used to detect objects of different sizes, generate candidate bounding boxes with different aspect ratios, and use non-maximum suppression (NMS) to output the final bounding box with position, size and confidence score.

[0080] YOLOv5 is an efficient and high-precision target detection model. It is an advanced object detection algorithm based on the YOLO series and can achieve real-time detection. It is used in various scenarios, such as security monitoring, autonomous driving, industrial quality inspection, etc. It is one of the mainstream algorithms in the field of target detection. Compared with other detection networks, YOLOv5 has the following characteristics:

[0081] (1) Better detection accuracy: The mAP indicator has been improved on the PASCAL VOC and COCO datasets.

[0082] (2) Faster detection speed: The detection speed is accelerated without losing accuracy.

[0083] (3) Smaller model size: The model compression method is used to reduce the model size.

[0084] (4) Easier to train: The network structure and loss function are modified to make the model easier to optimize and train.

[0085] (5)Support more tasks.

[0086] (6) Introducing attention mechanism: The attention module is used to enhance the feature expression ability of the model.

[0087] (7) Based on PyTorch: Coding is simpler and community support is wider.

[0088] The training steps of the deep network detection model (YOLOv5) can be summarized as follows:

[0089] First, install the necessary deep learning framework and YOLO library, prepare a dataset containing images and labels, then configure the YOLO model and training parameters, then train the model on the training set and use the validation set to evaluate the performance. After training, test the model performance on the test set and optimize the model based on the results. Finally, deploy the trained model to actual applications and continue to iterate based on feedback.

[0090] 3. An image analysis module is responsible for obtaining blood cell category information from the single blood cell image collection and generating a blood cell classification result.

[0091] Specifically, the image analysis module consists of a blood cell image classification module, a white blood cell image classification module and a platelet image classification module, and the blood cell image classification module is embedded with a deep network classification model, the white blood cell image classification module is embedded with a white blood cell single classification model, and the platelet image classification module is embedded with a platelet size classification model.

[0092] The image analysis module is first responsible for inputting the acquired single blood cell image collection into the trained deep network classification model, and the deep network classification model classifies the single blood cell image collection to distinguish the white blood cell images, red blood cell images and platelet images therein, and generates blood cell category information.

[0093] Then, the collection of the distinguished white blood cell images is input into the trained white blood cell single classification model, see Figure 3-4 As shown, the collection of white blood cell images is classified by the white blood cell single classification model to distinguish the neutrophil images containing abnormal inclusion bodies (abbreviated as: "inclusion body neutrophil" images) and the neutrophil images without abnormal inclusion bodies (abbreviated as: "non-inclusion body neutrophil" images), and generate white blood cell morphological information.

[0094] At the same time, it is responsible for inputting the set of the distinguished platelet images into the trained platelet size classification model, see Figure 5-6As shown, the platelet size classification model performs platelet segmentation processing (traditional image processing method) on the collection of platelet images to obtain the size of each platelet therein and generate platelet size information.

[0095] Generally, platelets with a diameter greater than 3 microns are defined as large platelets, and platelets with a diameter less than 3 microns are defined as normal-sized platelets.

[0096] Finally, it is responsible for integrating the generated blood cell category information, white blood cell morphology information and platelet size information into blood cell classification results.

[0097] As a preferred embodiment of the present invention, see Figure 7 As shown in the figure, the deep network classification model and the single leukocyte classification model both adopt the ResNet-18 model, whose network structure consists of multiple residual blocks with convolutional layers and downsampling layers, average pooling layers and fully connected layers. Each residual block contains a shortcut connection for adding the input of a block to its output to achieve gradient flow and simplify the optimization process.

[0098] After the image is input into the ResNet-18 model, it is processed by a multi-layer convolutional neural network, including a convolutional layer, a maximum pooling layer, multiple residual blocks, and a global average pooling layer. After passing through a fully connected layer and a Softmax output layer, accurate classification of blood cells, white blood cells, and platelets is finally achieved.

[0099] ResNet18 is a concise and efficient image classification network model, known for its jump connections that can alleviate the gradient vanishing problem. It is a typical representative of deep residual networks and the ultimate balance point. It has inspired many subsequent model innovations, and its main features include:

[0100] (1) The residual block structure is introduced, which can directly transfer gradients through identity shortcuts, solving the gradient vanishing problem when the network is deepened.

[0101] (2) A more simplified network design is adopted, which consists of convolutional layers, batch normalization layers, activation layers and residual blocks, greatly reducing the number of model parameters.

[0102] (3) Decompose the network into multiple residual learning units. The overall network can be very deep, but each learning unit only needs to fit a very small residual function.

[0103] (4) A global average pooling layer is used to reduce the parameters of the fully connected layer and avoid overfitting.

[0104] (5) In the ImageNet image classification task, the top-1 accuracy of ResNet18 can reach 69.6%, making it one of the most successful classification models in history.

[0105] (6) Small computational complexity and fast speed make it very suitable for deployment on mobile and embedded devices.

[0106] The training steps of the deep network classification model and the single classification model of white blood cells (ResNet18) include:

[0107] Prepare labeled training data sets, design a suitable neural network architecture, initialize network parameters, use training data to perform forward propagation and loss calculation on the network, update network weights through backpropagation, continuously iterate and optimize until performance requirements are met, and finally evaluate the generalization ability of the model on the validation set and test set.

[0108] The platelet size classification model is a traditional image segmentation model. The platelet segmentation refers to using the threshold segmentation method in the traditional image processing method to divide the platelet image into background and platelets to obtain the area of ​​the platelets (in pixels).

[0109] 4. The counting and processing module is responsible for counting and processing the data of various types of blood cells in the blood cell classification results, and generating analysis results of the cases to be screened.

[0110] Specifically, the counting and processing module will count all red blood cell images, white blood cell images, and platelet images, and will also separately count the "inclusion neutrophil" images and "non-inclusion neutrophil" images in the white blood cell images, and will also separately count the large platelet images and normal-sized platelet images in the platelet images, and calculate the proportion of large-sized platelet images in all platelet images, thereby ultimately generating the analysis results of the cases to be screened through data processing.

[0111] 5. The related disease judgment module is responsible for judging whether the case to be screened is MYH9-RD based on the analysis results of the case to be screened, using a method based on knowledge reasoning and clinical judgment fusion, and generating MYH9-RD screening results.

[0112] Specifically, the relevant disease judgment module includes a feature extraction module and a case classification module.

[0113] The feature extraction module is responsible for extracting characteristic cell morphologies related to MYH9-RD, including the number of "large platelets", the proportion of "large platelets" and the number of "inclusion body neutrophils" from the analysis results of the cases to be screened using regular expressions, thereby obtaining the MYH9-RD case characteristics.

[0114] The case classification module is responsible for judging whether the case to be screened is MYH9-RD based on the MYH9-RD case characteristics by a method based on knowledge reasoning and clinical judgment fusion, and generating a MYH9-RD screening result, such as a "suspected MYH9-RD" initial screening result or a "non-MYH9-RD" initial screening result.

[0115] From a clinical perspective, it is only meaningful when the proportion of large platelets is higher than the normal reference range (normal reference range <7%). Therefore, the two important conditions for judging whether the case to be screened is MYH9-RD in the present invention are whether the proportion of large platelets is greater than 7% and whether "inclusion body neutrophils" are detected. Only when these two conditions are met at the same time can the case to be screened be judged as a "suspected MYH9-RD" initial screening result, otherwise a "non-MYH9-RD" initial screening result is generated.

[0116] 6. A result output module is responsible for outputting the MYH9-RD screening results.

[0117] Specifically, the MYH9-RD screening results can be output in a visual form and displayed on a computer monitor or a mobile terminal screen, for example, displaying the initial screening results such as "suspected MYH9-RD" or "non-MYH9-RD" for intuitive viewing by professional physicians.

[0118] The MYH9-RD auxiliary screening system of the present invention is mainly used to detect and classify large platelets and "inclusion neutrophils" in blood samples, and aims to improve the efficiency and accuracy of clinical testing in an automated manner.

[0119] Before training the model, data collection and preprocessing are required, including: obtaining blood cell image datasets from a variety of clinical and laboratory sources, and then having professionals annotate the collected blood cell image datasets, paying special attention to the characteristics of large platelets and "inclusion neutrophils" to ensure the accuracy and diversity of the training data.

[0120] When training the model, the labeled data set is input into the YOLOv5 model. With the efficient detection performance of the YOLOv5 model and the efficient classification performance of the ResNet18 model, the MYH9-RD auxiliary screening system of the present invention focuses on identifying large platelets and "inclusion neutrophils" in blood cell images.

[0121] When using the MYH9-RD auxiliary screening system of the present invention, the input test image first passes through the deep network detection model (YOLOv5) to identify the cell area in the image. The network uses the Focus structure and the CSP structure to enhance feature extraction; then the identified cell area is analyzed by the three-level corresponding classification network (ResNet18) to accurately distinguish large platelets and "inclusion neutrophils". The classification network architecture combines the FPN+PAN structure to improve the fusion of multi-scale features and classification accuracy.

[0122] After accurately distinguishing large platelets from "inclusion neutrophils", the MYH9-RD auxiliary screening system of the present invention outputs detection results including large platelets and "inclusion neutrophils", provides detailed cell count and characteristic information, and then uses GIOU_Loss as the loss function to further optimize the detection precision and ensure a high level of detection accuracy.

[0123] The MYH9-RD auxiliary screening system of the present invention has the following innovations:

[0124] (1) Focus on large platelets and “inclusion neutrophils”: The system excels in accurately identifying, classifying, and counting these specific cells, providing highly specialized medical analysis capabilities for MYH9-RD screening.

[0125] (2) Advanced network structure: Using modern deep learning architectures (such as Yolov5) combined with data enhancement and optimization algorithms improves the efficiency and reliability of detection.

[0126] (3) Automation and precision: AI technology is used to conduct early, rapid, and efficient screening of the rare disease MYH9-RD, provide immediate clinical judgment and decision-making data support, and improve screening accuracy and work efficiency.

[0127] Based on the above-mentioned MYH9-RD auxiliary screening system, the present invention also provides a control method for the MYH9-RD auxiliary screening system which is not intended for direct diagnosis of MYH9-RD but is only used to assist in the early screening of MYH9-RD, see Figure 8 As shown, the steps include:

[0128] Step 1) The image acquisition module obtains a Wright-stained peripheral blood image of the case to be screened. The Wright-stained peripheral blood image is a microscopic image of the Wright-stained peripheral blood of the case to be screened acquired by using a high-power microscope and an oil immersion lens.

[0129] Step 2) The image detection module inputs the Wright-stained peripheral blood image into a trained deep network detection model for processing to obtain a separate image of each blood cell in the Wright-stained peripheral blood image and integrate them into a collection of single blood cell images.

[0130] Step 3) The image analysis module inputs the single blood cell image collection into a trained deep network classification model for classification processing to distinguish the white blood cell images, red blood cell images and platelet images in the single blood cell image collection and generate blood cell category information.

[0131] Step 4) The image analysis module inputs the collection of the distinguished white blood cell images into the trained white blood cell single classification model for classification processing, see Figure 3-4 As shown, the neutrophil images containing abnormal inclusion bodies (abbreviated as “neutrophil with inclusion bodies” images) and the neutrophil images without abnormal inclusion bodies (abbreviated as “neutrophil without inclusion bodies” images) in the leukocyte image collection are distinguished, and the leukocyte morphology information is generated.

[0132] Step 5) The image analysis module inputs the set of the differentiated platelet images into the trained platelet size classification model for platelet segmentation (traditional image processing method), see Figure 5-6 As shown, the size of each platelet in the platelet image collection is obtained and platelet size information is generated. Generally, platelets with a diameter of more than 3 microns are defined as large platelets, and platelets with a diameter of ≤3 microns are defined as normal-sized platelets.

[0133] Step 6) The image analysis module integrates the blood cell category information, the white blood cell morphology information and the platelet size information to generate a blood cell classification result.

[0134] Step 7) The counting and processing module counts and processes the various types of blood cells in the blood cell classification results, and generates analysis results of the cases to be screened.

[0135] Specifically, the counting and processing module will count all red blood cell images, white blood cell images, and platelet images, and will also separately count the "inclusion neutrophil" images and "non-inclusion neutrophil" images in the white blood cell images, and will also separately count the large platelet images and normal-sized platelet images in the platelet images, and calculate the proportion of large-sized platelet images in all platelet images, thereby ultimately generating the analysis results of the cases to be screened through data processing.

[0136] Step 8) The relevant disease judgment module uses regular expressions to extract characteristic cell morphologies related to MYH9-RD, including the number of "large platelets" and the number of "inclusion body neutrophils", from the analysis results of the cases to be screened, thereby obtaining the MYH9-RD case characteristics.

[0137] Step 9) The relevant disease judgment module determines whether the case to be screened is MYH9-RD based on the MYH9-RD case characteristics by a method based on knowledge reasoning and clinical judgment fusion, and generates a MYH9-RD screening result, such as a "suspected MYH9-RD" initial screening result or a "non-MYH9-RD" initial screening result.

[0138] From a clinical perspective, it is only meaningful when the proportion of large platelets is higher than the normal reference range (normal reference range <7%). Therefore, the two important conditions for judging whether the case to be screened is MYH9-RD in the present invention are whether the proportion of large platelets is greater than 7% and whether "inclusion body neutrophils" are detected. Only when these two conditions are met at the same time can the case to be screened be judged as a "suspected MYH9-RD" initial screening result, otherwise a "non-MYH9-RD" initial screening result is generated.

[0139] Step 10) The MYH9-RD screening result is output by a result output module.

[0140] Specifically, the MYH9-RD screening results can be output in a visual form and displayed on a computer monitor or a mobile terminal screen, for example, displaying the initial screening results such as "suspected MYH9-RD" or "non-MYH9-RD" for intuitive viewing by professional physicians.

[0141] The present invention is tested by using blood smears of MYH9-RD patients and blood smears of normal people, and the test process and results are as follows:

[0142] 1. Blood smear test of normal people;

[0143] See also Fig. 9 As shown in a, firstly, a Wright-stained peripheral blood smear of a normal person is prepared;

[0144] See also Fig. 9 As shown in b, the system of the present invention is then used to perform a full-slide scan on the prepared Wright-stained peripheral blood smear;

[0145] See also Fig. 9 As shown in c, the platelet AI test results show that platelets account for 98.84% of the normal blood smear, and large platelets account for 1.16%;

[0146] See also Fig. 9As shown in d, the results of the WBC AI test showed that neutrophils accounted for 65% of the blood smear of normal people, and there were no inclusion neutrophils;

[0147] See also Fig.10 As shown, the auxiliary initial screening result output by the system of the present invention is "non-MYH9-RD".

[0148] 2. Blood smear test for MYH9-RD patients;

[0149] See also Fig.11 As shown in a, a Wright-stained peripheral blood smear of a MYH9-RD patient was first prepared;

[0150] See also Fig.11 As shown in b, the system of the present invention is then used to perform a full-slide scan on the prepared Wright-stained peripheral blood smear;

[0151] See also Fig.11 As shown in c, the platelet AI test results showed that normal-sized platelets accounted for 59.45% and large platelets accounted for 40.55% in the blood smears of MYH9-RD patients;

[0152] See also Fig.11 As shown in d, the results of leukocyte AI test showed that inclusion neutrophils were identified in the blood smear of MYH9-RD patients;

[0153] See also Fig.12 As shown, the auxiliary initial screening result output by the system of the present invention is "suspected MYH9-RD".

[0154] It needs to be emphasized again here that the MYH9-RD auxiliary screening system and its control method proposed in the present invention are not diagnostic tools and diagnostic methods that can directly diagnose MYH9-RD, and their original intention of development is not to directly diagnose MYH9-RD, but to serve as an auxiliary screening tool and method to help professional physicians make early diagnosis of MYH9-RD.

[0155] That is to say, the MYH9-RD screening results obtained by the MYH9-RD auxiliary screening system and control method of the present invention cannot be directly used as the MYH9-RD diagnostic conclusion for the case to be screened, but can only be used as one of the reference bases for professional physicians to diagnose MYH9-RD, or more accurately, as an early screening basis. Only after confirming the MYH9-RD screening results will the professional physician be given an early warning that the test case may be MYH9-RD, and then the professional physician will make a final diagnosis of the test case through the MYH9-RD gene detection method and give a clear diagnosis conclusion.

[0156] For example, the patients to be screened can first undergo a routine blood test. If the routine blood test reveals a low platelet count (<150×10 9 / L), then Wright staining of the blood smear can be further performed, and the MYH9-RD auxiliary screening system and control method of the present invention can be used to perform MYH9-RD auxiliary screening. If the screening result is suspected MYH9-RD, the MYH9-RD gene needs to be detected for final confirmation.

[0157] The present invention uses artificial intelligence algorithms to accurately identify and count the special cell morphologies in the peripheral blood of disease patients (Wright-stained "large platelets" and "inclusion body neutrophils" in peripheral blood smears), and can extract deep features of cell images that are difficult for humans to identify, make up for the shortcomings of incomplete human visual capture and easy fatigue, reduce tiny lesions missed by the naked eye, reduce misdiagnosis and missed diagnosis caused by human errors or subjective differences, achieve accurate identification of cell morphology and greatly improve work efficiency.

[0158] The present invention has the advantages of early, simple, economical and accurate screening, and can greatly improve the accuracy of early screening of MYH9-RD. The screening results can be used as a reference to assist hematologists in diagnosing MYH9-RD cases, which helps to alleviate the problems of shortage of hematologists, low diagnostic level of grassroots personnel, low work efficiency, and uneven distribution of medical resources.

[0159] Compared with the existing MYH9-RD gene diagnosis technology (chip capture high-throughput sequencing + Sanger verification method), the present invention does not require expensive instruments and equipment and specialized bioinformatics analysis technicians, but only requires an ordinary computer, with no additional costs, fast speed (measured in seconds), simple and clear result output, and is suitable for grassroots laboratories.

[0160] The present invention also provides a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus, and the memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the control method of the above-mentioned MYH9-RD auxiliary screening system.

[0161] The present invention also provides a computer storage medium, in which at least one executable instruction is stored, and the executable instruction enables a processor to execute operations corresponding to the control method of the above-mentioned MYH9-RD auxiliary screening system.

[0162] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A MYH9-RD auxiliary screening system based on blood cell morphology, characterized in that: include: The image acquisition module is responsible for obtaining Wright-stained peripheral blood images of the cases to be screened; An image detection module, responsible for acquiring a separate image of each blood cell from the Wright-stained peripheral blood image and generating a collection of single blood cell images; An image analysis module is responsible for obtaining blood cell category information, white blood cell morphology information and platelet size information from the single blood cell image collection, and generating a blood cell classification result; The counting and processing module is responsible for counting and processing the data of various blood cells in the blood cell analysis results, and generating the analysis results of the cases to be screened; The relevant disease judgment module is responsible for judging whether the case to be screened is MYH9-RD based on the analysis results of the case to be screened, using a method based on knowledge reasoning and clinical judgment fusion, and generating a MYH9-RD screening result; The result output module is responsible for outputting the MYH9-RD screening results.

2. The MYH9-RD auxiliary screening system according to claim 1, characterized in that: The Wright-stained peripheral blood image is a microscopic image of the Wright-stained peripheral blood of the case to be screened, which is collected using a high-power microscope and an oil immersion lens.

3. The MYH9-RD auxiliary screening system according to claim 1, characterized in that: The image detection module is embedded with a deep network detection model. The image detection module is firstly responsible for inputting the acquired Wright-stained peripheral blood image into the trained deep network detection model, and the deep network detection model processes the input Wright-stained peripheral blood image to generate a separate image of each blood cell, and then is responsible for integrating the generated separate image of each blood cell into a collection of single blood cell images.

4. The MYH9-RD auxiliary screening system according to claim 3, characterized in that: The deep network detection model adopts the YOLOv5 model, and its network structure consists of a backbone network, a neck network and three prediction heads; The backbone network, including Conv, CSPDarkNet53 and SPPF modules, is used to extract and integrate features from the input image and generate a feature map; The neck network is used to fuse the feature maps generated by the backbone network; it is combined with a feature pyramid network (FPN) and a path aggregation network (PANet) to enhance semantic information and multi-scale positioning capabilities; The three prediction heads are used to detect objects of different sizes, generate candidate bounding boxes with different aspect ratios, and use non-maximum suppression (NMS) to output the final bounding box with position, size and confidence score.

5. The MYH9-RD auxiliary screening system according to claim 1, characterized in that: The image analysis module is composed of a blood cell image classification module, a white blood cell image classification module and a platelet image classification module, and the blood cell image classification module is equipped with a deep network classification model, the white blood cell image classification module is equipped with a white blood cell single classification model, and the platelet image classification module is equipped with a platelet size classification model; The image analysis module is first responsible for inputting the acquired single blood cell image collection into the trained deep network classification model, and the deep network classification model classifies the single blood cell image collection to distinguish the white blood cell images, red blood cell images and platelet images therein, and generates blood cell category information; Then, the set of the distinguished white blood cell images is input into the trained white blood cell single classification model, and the white blood cell single classification model classifies the set of white blood cell images to distinguish the "inclusion neutrophil" images and the "non-inclusion neutrophil" images, and generates white blood cell morphology information; At the same time, it is responsible for inputting the set of the differentiated platelet images into the trained platelet size classification model, and the platelet size classification model performs platelet segmentation processing on the set of platelet images to obtain the size of each platelet therein and generate platelet size information; Finally, it is responsible for integrating the generated blood cell category information, white blood cell morphology information and platelet size information into blood cell classification results.

6. The MYH9-RD auxiliary screening system according to claim 5, characterized in that: The deep network classification model and the single white blood cell classification model both adopt the ResNet-18 model, whose network structure consists of multiple residual blocks with convolutional layers and downsampling layers, average pooling layers and fully connected layers; each residual block contains a shortcut connection for adding the input of a block to its output to achieve gradient flow and simplify the optimization process.

7. The MYH9-RD auxiliary screening system according to claim 1, characterized in that: The related disease judgment module includes a feature extraction module and a case classification module; wherein, The feature extraction module is responsible for extracting characteristic cell morphologies related to MYH9-RD, including the number of "large platelets", the proportion of "large platelets" and the number of "inclusion body neutrophils", from the analysis results of the cases to be screened using regular expressions, thereby obtaining the characteristics of MYH9-RD cases; The case classification module is responsible for judging whether the case to be screened is MYH9-RD based on the characteristics of the MYH9-RD case by a method based on the fusion of knowledge reasoning and clinical judgment.

8. A control method for the MYH9-RD auxiliary screening system according to any one of claims 1 to 7, characterized in that: This method is not intended to directly diagnose MYH9-RD, but is only used to assist in the early screening of MYH9-RD. The steps include: Step 1) obtaining a Wright-stained peripheral blood image of the case to be screened by an image acquisition module; Step 2) the image detection module inputs the Wright-stained peripheral blood image into a trained deep network detection model for processing to obtain a separate image of each blood cell in the Wright-stained peripheral blood image, and integrates the images into a collection of single blood cell images; Step 3) the image analysis module inputs the single blood cell image collection into a trained deep network classification model for classification processing to distinguish the white blood cell images, red blood cell images and platelet images in the single blood cell image collection, and generates blood cell category information; Step 4) the image analysis module inputs the collection of the distinguished white blood cell images into the trained white blood cell single classification model for classification processing to distinguish the "inclusion neutrophil" image and the "non-inclusion neutrophil" image in the collection of white blood cell images, and generates white blood cell morphology information; Step 5) the image analysis module inputs the set of the differentiated platelet images into a trained platelet size classification model for platelet segmentation processing to obtain the size of each platelet in the set of platelet images and generate platelet size information; Step 6) the image analysis module integrates the blood cell category information, the white blood cell morphology information and the platelet size information to generate a blood cell classification result; Step 7) The counting and processing module counts and processes the various types of blood cells in the blood cell classification results, and generates analysis results of the cases to be screened; Step 8) The relevant disease judgment module uses regular expressions to extract characteristic cell morphologies related to MYH9-RD, including the number of "large platelets", the proportion of "large platelets" and the number of "inclusion body neutrophils", from the analysis results of the cases to be screened, thereby obtaining the characteristics of MYH9-RD cases; Step 9) The relevant disease judgment module determines whether the case to be screened is MYH9-RD based on the characteristics of the MYH9-RD case by a method based on knowledge reasoning and clinical judgment fusion, and generates a MYH9-RD screening result; Step 10) The MYH9-RD screening result is output by a result output module.

9. A computer device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus, and the memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the control method of the MYH9-RD auxiliary screening system as described in claim 8.

10. A computer storage medium, wherein at least one executable instruction is stored in the computer storage medium, wherein the executable instruction enables a processor to perform an operation corresponding to the operation corresponding to the control method of the MYH9-RD auxiliary screening system as described in claim 8.

Images