Microscopic image parasitic ovum intelligent detection system

Abstract

The invention discloses a microscopic image parasitic ovum intelligent detection system, and relates to the field of public health monitoring and image recognition analysis. The system comprises three modules: a microscopic magnification guide module, a slide area search guide module and a parasitic ovum automatic detection and identification module. The microscopic magnification guide module utilizes an artificial intelligence technology to realize intelligent guidance of microscope magnification and ensure clear imaging of parasitic ova. And the slide area searching and guiding module realizes full coverage of the slide area through intelligent scanning, so that omission of manual operation is avoided. The parasitic ovum automatic detection and recognition module accurately recognizes parasitic ovum from a complex background by using deep learning and image processing technologies, and performs statistical analysis. According to the method, the microscopic image acquisition and analysis efficiency is remarkably improved, the dependence on professionals is reduced, and powerful technical support is provided for parasite ovum analysis and public health decision making.

Description

Technical Field

[0001] The present invention relates to the intersection of artificial intelligence, image processing, biological testing and automatic control, and mainly relates to an intelligent collection system of parasite egg microscopic images and an intelligent detection system of parasite eggs, and specifically relates to an intelligent detection system of parasite eggs using microscopic images. Background Art

[0002] Parasitic infections seriously endanger people's health and are an important public health issue. Since the founding of the People's Republic of China, the Party and the government have always attached great importance to the prevention and treatment of parasitic diseases and established a daily monitoring system. The infection rate of key parasitic diseases has dropped significantly, playing an extremely important role in protecting the health and life safety of the people and promoting economic and social development.

[0003] Traditional parasite egg detection methods usually rely on manual observation and identification under a microscope, a process that is not only time-consuming and inefficient, but also requires extremely high professional skills from operators. Due to factors such as eye fatigue and experience differences, the accuracy of manual detection is difficult to guarantee, which may lead to missed detection or misjudgment, thus affecting the timeliness and effectiveness of disease control decisions.

[0004] With the development of modern science and technology, especially the progress of artificial intelligence (AI), image processing and precision control technology, automated and intelligent detection methods have gradually emerged. However, there are still deficiencies in the exploration of intelligent acquisition and object recognition of microscopic images, which cannot meet the needs of efficient and accurate automatic detection of various parasite eggs in large-scale sample screening scenarios.

[0005] Therefore, the development of an automatic parasite egg detection system that can achieve automatic adjustment of magnification, comprehensive search of the slide area, and intelligent detection and identification has very important practical significance and broad application prospects for improving the efficiency of disease control work, reducing labor costs, and ensuring the accuracy and completeness of monitoring data. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a technology capable of realizing automatic adjustment of magnification, global search technology of slide area, and intelligent detection and identification technology of parasite eggs.

[0007] In order to solve the above technical problems, the present invention proposes a microscopic image parasite egg intelligent detection system, comprising a microscopic magnification guidance module, a slide area search guidance module and a parasite egg automatic detection and identification module arranged in sequence; wherein,

[0008] The microscopic magnification guidance module uses artificial intelligence technology to achieve intelligent guidance of microscope magnification to ensure clear imaging of parasite eggs, including: statistical analysis of optimal magnification, intelligent decision-making of zooming in and out of the current microscopic image;

[0009] The slide area search and guidance module achieves full coverage of the slide area through intelligent scanning to avoid omissions caused by manual operation. It includes: intelligent recognition of the slide sample boundary area, intelligent decision-making of the stage slide movement direction and step length, and intelligent splicing of split-lens microscopic images.

[0010] The parasite egg automatic detection and identification module uses deep learning and image processing technology to accurately identify parasite eggs from complex backgrounds and perform statistical analysis, including: automatic detection and identification of parasite eggs, classification and counting of parasite eggs, and automatic analysis of the properties of parasite eggs.

[0011] The specific steps are as follows:

[0012] 1) Microscope magnification guidance module:

[0013] (1) Statistical analysis of optimal magnification;

[0014] (2) Intelligent decision-making on the zooming in and out of the current microscopic image;

[0015] 2) Slide area search guidance module:

[0016] (3) Intelligent recognition of the border area of ​​slide samples;

[0017] (4) Intelligent decision-making on the movement direction and step length of the stage slide;

[0018] (5) Intelligent stitching of split-lens microscopic images;

[0019] 3) Parasite egg automatic detection and identification module:

[0020] (6) Automatic detection and identification of parasite eggs;

[0021] (7) parasite egg classification and counting;

[0022] (8) Automatic analysis of parasite egg properties.

[0023] The optimal magnification statistical analysis described in step (1) specifically includes:

[0024] ① Collect a batch of slides for each type of parasite eggs, and manually adjust each slide to the optimal magnification so that the eggs are clearly visible;

[0025] ② Calculate the mean and variance of the optimal magnification for the entire batch of slides.

[0026] The intelligent decision-making of the zooming-in and zooming-out actions of the current microscopic image described in step (2) specifically includes:

[0027] ① Taking the mean of the optimal magnification calculated in step (1) as the center, enumerate multiple magnifications from the two dimensions of magnification and reduction;

[0028] ② Collect several microscopic images at each magnification;

[0029] ③Mark each microscopic image, and the marking value can reflect the scale of the distance to the optimal magnification;

[0030] ④Training the zoom action command predictor based on the deep learning model;

[0031] ⑤ Use the mean of the optimal magnifications calculated in step (1) as the empirical optimal magnification to initialize the microscope magnification, call the zoom action command predictor trained in the previous step, obtain and execute the zoom action commands, and continuously fine-tune to the optimal magnification that suits the current slide.

[0032] The intelligent identification of the border area of ​​the slide sample described in step (3) specifically includes:

[0033] ① Training sample collection: Based on a batch of slides in step (1), the magnification is manually adjusted to the optimal magnification, and microscopic images of the sample-rich area of ​​the slide are collected as positive data, and microscopic images of the edge area are collected as negative data;

[0034] ②Use a deep learning model to train a two-class classifier that can distinguish between sample-rich areas and edge areas;

[0035] The intelligent decision-making of the moving direction and step length of the stage slide described in step (4) specifically includes:

[0036] ① Move a small step in any direction and obtain two microscopic images before and after the movement;

[0037] ② Calculate the size of the overlapping area of ​​the two images, thereby obtaining the proportional relationship between the size of the overlapping area and the moving step length;

[0038] ③ According to the preset overlap ratio, convert it into the corresponding moving step length;

[0039] ④ During the movement process, based on the classifier obtained in step (3), determine whether each microscopic image is a boundary area;

[0040] ⑤ When encountering a boundary area, move one step in the vertical direction, then turn back to the original direction and continue moving to complete the microscopic image acquisition in the entire space.

[0041] The intelligent stitching of the split-lens microscopic images described in step (5) specifically includes:

[0042] ① Determine the overlapping area of ​​two microscopic images taken at adjacent steps;

[0043] ② Overlay the overlapping areas to complete the splicing;

[0044] ③ Based on the spatial search method described in step (4), each step is stitched to finally obtain a panoramic image of the glass slide.

[0045] The automatic detection and identification of parasite eggs described in step (6) specifically includes:

[0046] ① Collect a large number of microscopic images at the optimal magnification;

[0047] ②Mark the location frame and category of parasite eggs in the microscopic image;

[0048] ③Training a parasite egg detector based on a deep learning model.

[0049] The parasite egg classification and counting described in step (7) specifically includes:

[0050] ① On the glass slide panoramic image obtained in step (6), the parasite egg detector obtained in step (6) is called to realize automatic detection and identification of the parasite eggs on the glass slide to be inspected;

[0051] ② Based on the test results, count the number of parasite eggs appearing on the slide according to the type of parasite eggs.

[0052] The automatic analysis of the properties of the parasite eggs described in step (8) specifically includes:

[0053] ① On the glass slide panorama in step (7), intercept the egg image block according to the marked box of the parasite egg position;

[0054] ② Label the properties of each egg, including but not limited to: fertile eggs, non-fertile eggs, and growth stage;

[0055] ③Training attribute identifiers based on deep learning models;

[0056] ④ For any test image, call the parasite egg detector obtained in step (6), obtain the parasite egg position frame, and intercept the parasite egg image block, input it into the attribute identifier obtained in the previous step, and obtain the attribute analysis result.

[0057] The beneficial effects of the present invention are:

[0058] 1. In the microscopic image acquisition stage, the present invention realizes automatic guidance of magnification and automatic search of the slide space area, and can automatically obtain a full-area microscopic image of the slide with clear imaging of parasite eggs, providing a good data basis for the next step of automatic detection and identification of parasite eggs. Thus, the reliance on expert manpower in microscopic image acquisition is solved.

[0059] 2. In the stage of automatic detection and identification of parasite eggs, the intelligent prediction of the location of parasite eggs in the microscopic image is realized, and the number of detected eggs is counted and the attribute analysis is performed according to the category. This solves the problem of time-consuming and labor-intensive screening and empirical errors in large-scale sample screening, and can automatically statistically analyze the corresponding indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0061] Figure 1 A schematic diagram of the framework of a microscopic image parasite egg intelligent detection system;

[0062] Figure 2 Microscopic images of parasite eggs clearly imaged at optimal magnification, where the left image is a microscopic image under foreign body contamination, and the right image is a microscopic image under changing lighting conditions.

[0063] Figure 3 Enumerate the microscopic images under various magnifications, where the left image is a microscopic image under a smaller magnification, and the right image is a microscopic image under the optimal magnification;

[0064] Figure 4 Prediction and execution of zoom-in / zoom-out action commands. The left image is an image that needs to be further magnified, and the right image is an image that has reached the optimal magnification.

[0065] Figure 5 An example of a slide edge region identifier training sample, where the left image is a left and right boundary image sample, and the right image is an upper and lower boundary image sample;

[0066] Figure 6 An example of overlapping region recognition based on interest point detection and matching, where the left image is a microscopic image before moving a small step size S, and the right image is a microscopic image after moving a small step size S;

[0067] Figure 7 An example of panorama stitching, where the top image is a local micrograph and the bottom image is a panorama stitching image;

[0068] Figure 8 Example of parasite egg detection and identification: blue boxes are manually annotated and the model predicts red boxes. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0070] In the present invention, a microscopic image parasite egg intelligent detection system framework is as follows Figure 1 As shown, the following steps are included:

[0071] 1) Microscope magnification guide:

[0072] (1) Statistical analysis of optimal magnification;

[0073] (2) Intelligent decision-making on the zooming in and out of the current microscopic image;

[0074] 2) Slide area search guide:

[0075] (3) Intelligent recognition of the border area of ​​slide samples;

[0076] (4) Intelligent decision-making on the movement direction and step length of the stage slide;

[0077] (5) Intelligent stitching of split-lens microscopic images;

[0078] 3) Automatic detection and identification of parasite eggs:

[0079] (6) Automatic detection and identification of parasite eggs;

[0080] (7) parasite egg classification and counting;

[0081] (8) Automatic analysis of parasite egg properties.

[0082] 1. Statistical analysis of the optimal magnification. The specific steps are as follows:

[0083] 1) Collect a batch of slides for each parasite egg {B cn}, where c = 1, 2, ..., C represents the parasite species, n = 1, 2, ..., N represents the number of slides, and each slide is manually adjusted to the optimal magnification Make the eggs clearly visible, such as Figure 2 As shown;

[0084] 2) Calculate the mean μ and variance σ of the optimal magnification on the entire batch of slides 2 ,Right now, This mean μ can be regarded as the empirically optimal magnification.

[0085] 2. Intelligent decision-making of the zoom in and zoom out actions of the current microscopic image. The specific steps are as follows:

[0086] 1) Taking the mean μ of the optimal magnification calculated in step (1) as the center, enumerate S magnifications {μ+s i},s i ∈[-3σ, 3σ], i=1, 2,…,S; where s i Refers to the zoom value based on the average optimal magnification μ.

[0087] 2) Collect several microscopic images at each magnification m refers to the number of images collected at each zoom factor, such as Figure 3 As shown;

[0088] 3) Mark the relative zoom value of each microscopic image The marker value can reflect the size of the distance to the optimal magnification, for example, The marked value here has a positive or negative sign, indicating that an enlargement or reduction operation is performed relative to the empirical optimal magnification μ. The absolute value of the marked value indicates the relative scale of the enlargement or reduction. Here, the standard deviation σ calculated in step (1) is used as a reference. From the perspective of Gaussian distribution, the variation amplitude of 6σ can basically cover the possible variation range.

[0089] 4) Training phase: training the zoom action command predictor based on the deep learning model This task can be modeled as a regression task in machine learning, that is, given the imaging features of the microscopic image Predict the relative scaling value of its magnification Based on the microscopic images collected in step 3) and the annotated relative zoom values A training sample data pair can be constructed. Then, to build a deep neural network model, a relatively mature skeleton network can be used, such as a residual network, and the last network layer can be designed as a regression task head to fit the corresponding relative scaling values ​​in the training samples. In order to better utilize the continuity and gradualness of the magnification scaling space, a sequence model can be selected, such as a long short-term memory network (LSTM), to iteratively predict the relative scaling value in a sequence prediction manner.

[0090] 5) Testing phase: Initialize the microscope magnification with the empirical optimal magnification μ obtained in step (1), and call the zoom action command predictor trained in the previous step. Get and execute zoom in and zoom out commands like Figure 4 As shown, the magnification factor is constantly adjusted to the optimal magnification for the current slide.

[0091] 3. The specific steps of intelligent recognition of the boundary area of ​​the slide sample described are as follows:

[0092] 1) Training sample collection: Based on a batch of slides in step (1), the magnification is manually adjusted to the optimal magnification. Collect microscopic images of sample-rich areas of the slide I i+ As the positive sample set {I i+}, microscopic image I on the edge area i- As the negative sample set {I i-}, which includes negative sample examples in the four boundary directions of up, down, left, and right, such as Figure 5 As shown;

[0093] 2) Use a deep learning model to train a neural network that can distinguish sample-rich regions {I i+} and the edge area {I i-} of the two-class classifier Y = F(I i )={0,1}, where 1 represents the sample-rich area and 0 represents the edge area of ​​the slide. From the perspective of machine learning, the classifier is a two-class classifier and can be trained using a common deep neural network model.

[0094] 4. Intelligent decision-making of the moving direction and step length of the stage slide. The specific steps are as follows:

[0095] 1) Move a small step size S in any direction and obtain two microscopic images I before and after the movement 1 , I 2 ;

[0096] 2) Calculate the size of the overlapping area of ​​the two images O (pixels), and then obtain the proportional relationship S: O between the size of the overlapping area and the moving step size. The calculation of the overlapping area here can be achieved based on the method of interest point detection and matching, such as Figure 6 As shown;

[0097] 3) According to the preset overlap ratio R, convert it into the corresponding moving step length

[0098] 4) During the movement, based on the classifier Y obtained in step (3), i ), determine whether each microscopic image is a boundary area;

[0099] 5) When encountering a boundary area, move a step length S in the vertical direction * , then turn back to the original direction and continue moving and searching to complete the microscopic image acquisition in the entire space.

[0100] 5. Intelligent stitching of split-lens microscopic images. The specific steps are as follows:

[0101] 1) For two microscopic images taken at adjacent steps, determine their overlapping areas according to the interest point detection and matching method described in step 4;

[0102] 2) Overlapping the overlapping areas to complete the splicing;

[0103] 3) Based on the spatial search method described in step 4, each step is stitched to finally obtain the glass slide panorama I full ,like Figure 7 shown.

[0104] 6. Automatic detection and identification of parasite eggs. The specific steps are as follows:

[0105] 1) Collect a large number of microscopic images at the optimal magnification. For each slide, the optimal magnification here includes two types: the optimal magnification D obtained manually in step 1 * In step 2, the algorithm automatically finds the optimal magnification The latter can ensure that the parasite egg detection model has good robustness and can better adapt to the possible errors in the process of optimizing the amplification factor in real test situations;

[0106] 2) Annotate the location boxes and categories of parasite eggs in the microscopic image, which is done manually;

[0107] 3) Training the parasite egg detector Det(I) based on the deep learning model. Based on the above collected training data, any commonly used object detection model framework can be selected for training, such as the YOLO detector.

[0108] 7. Classification and counting of parasite eggs. The specific steps are as follows:

[0109] 1) Slide Panorama I obtained in step 5 full The parasite egg detector Det(I) obtained in step 6 is called to realize automatic detection and identification of the parasite eggs on the glass slide to be tested, and obtain the position marking box BB eggs and category Cls eggs Tags, such as Figure 8 As shown;

[0110] 2) According to the test results, count the number of parasite eggs appearing on the slide according to the type of parasite eggs.

[0111] 8. Automatic analysis of the properties of parasite eggs. The specific steps are as follows:

[0112] 1) Slide Panorama I in step 7 full On the top, the marked box BB according to the location of the parasite eggseggs , intercept the egg image block I eggs ;

[0113] 2) Label the attributes of each egg eggs , including but not limited to: pregnant eggs, non-pregnant eggs, and growth stages;

[0114] 3) Training an attribute identifier based on a deep learning model. This task can be modeled as a classification task in machine learning. Select any deep neural network model, such as the commonly used skeleton network Resnet, and set the last layer as the attribute classification task head. Based on the large amount of insect egg data collected above, {(I eggs , Attr eggs )}, perform model training and obtain the attribute identifier F Attr (I eggs );

[0115] 4) For any test image I eggs , call the parasite egg detector Det(I) obtained in step 7 to obtain the parasite egg position frame BB eggs , and intercept the parasite egg image block I eggs , input the attribute identifier F obtained in the previous step Attr (I eggs ) to obtain the attribute analysis results.

Claims

1. A microscopic image parasite egg intelligent detection system, characterized in that: It includes a microscopic magnification guidance module, a slide area search guidance module and a parasite egg automatic detection and identification module which are arranged in sequence; wherein the microscopic magnification guidance module includes in sequence: (1) Statistical analysis of optimal magnification; (2) Intelligent decision-making on the zooming in and out of the current microscopic image; The slide area search guidance module includes: (3) Intelligent recognition of the border area of ​​slide samples; (4) Intelligent decision-making on the movement direction and step length of the stage slide; (5) Intelligent stitching of split-lens microscopic images; The parasite egg automatic detection and identification module includes: (6) Automatic detection and identification of parasite eggs; (7) parasite egg classification and counting; (8) Automatic analysis of parasite egg properties.

2. The microscopic image parasite egg intelligent detection system as claimed in claim 1, characterized in that: The optimal magnification statistical analysis of step (1) specifically includes: (11) collecting a batch of slides for each type of parasite eggs, and manually adjusting each slide to an optimal magnification so that the eggs are clearly visible; (12) Calculate the mean and variance of the optimal magnification for the entire batch of slides.

3. The microscopic image parasite egg intelligent detection system as claimed in claim 2, characterized in that: The step (2) of intelligently deciding the zooming in and zooming out of the current microscopic image specifically includes: (21) Centering on the mean of the optimal magnification in step (11), enumerate multiple magnifications from the two dimensions of magnification and reduction, and collect a number of microscopic images at each magnification; and mark each microscopic image, and the mark value can reflect the size of the distance from the optimal magnification; (22) Training a zoom-in and zoom-out action command predictor based on a deep learning model; (23) Acquire a microscopic image at the optimal magnification of step (11), and obtain and execute the zoom action instructions through the zoom action instruction predictor of step (22) to fine-tune to the optimal magnification suitable for the current slide.

4. The microscopic image parasite egg intelligent detection system as claimed in claim 2, characterized in that: The step (3) of intelligently identifying the boundary area of ​​the slide sample specifically includes: (31) Based on the batch of slides in step (11), under manual adjustment to the optimal magnification, microscopic images of sample-rich areas of the slides are collected as positive data, and microscopic images of edge areas are collected as negative data; (32) A deep learning model is used to train a two-class classifier that can distinguish between sample-rich areas and edge areas.

5. The microscopic image parasite egg intelligent detection system as claimed in claim 4, characterized in that: The step (4) of intelligently determining the moving direction and step length of the stage slide specifically includes: (41) Move a small step along any direction and obtain two microscopic images before and after the movement; (42) calculating the size of the overlapping area of ​​the two images, thereby obtaining the proportional relationship between the size of the overlapping area and the moving step length; (43) converting into a corresponding moving step length according to a preset overlap ratio; (44) During the moving process, based on the two-class classifier in step (32), determining whether each microscopic image is a boundary area; (45) When encountering a boundary area, it moves one step in the vertical direction, then turns back to the original direction and continues to move to complete the microscopic image acquisition in the entire space.

6. The microscopic image parasite egg intelligent detection system as claimed in claim 1, characterized in that: The step (5) of intelligently stitching the split-lens microscopic images specifically includes: (51) For two microscopic images taken at adjacent steps, determine their overlapping area; (52) Overlapping the overlapping areas to complete the splicing; (53) Based on step 4, each step is stitched to finally obtain a panoramic image of the slide.

7. The microscopic image parasite egg intelligent detection system as claimed in claim 6, characterized in that: The step (6) of automatically detecting and identifying parasite eggs specifically includes: (61) Collect a large number of microscopic images at optimal magnification; (62) Annotate the location box and category of parasite eggs in microscopic images; (63) Training a parasite egg detector based on a deep learning model.

8. The microscopic image parasite egg intelligent detection system as claimed in claim 7, characterized in that: The step (7) of classifying and counting parasite eggs specifically comprises: (71) running the parasite egg detector of step (63) on the glass slide panoramic image of step (53) to achieve automatic detection and identification of the parasite eggs on the glass slide to be inspected; (72) Based on the test results, the number of parasite eggs appearing on the slide was counted according to the type of parasite eggs.

9. The microscopic image parasite egg intelligent detection system as claimed in claim 8, characterized in that: The step (8) of automatically analyzing the properties of parasite eggs specifically includes: (81) on the glass slide panorama of step (71), intercepting the egg image block according to the marked box of the parasite egg position; (82) Label the attributes of each egg, including but not limited to: fertile egg, non-fertile egg, and growth stage; (83) Training attribute identifiers based on deep learning models; (84) Based on the detection result of step (71), the parasite egg image block is intercepted and input into the attribute identifier of step (83) to obtain the attribute analysis result.

Images