Nugent analysis method and system for gram-stained slides

Abstract

The present invention provides a method and system for analyzing the overall Nugent of Gram staining films, including: step S1: classifying images collected by a microscope; step S2: using a neural network to classify and train the classified images to obtain a classification model; step S3: using the trained model to identify each image collected by a single sample; step S4: counting the recognition results of a single image, and obtaining the score segment of the entire sample based on the statistical analysis results; step S5: combining the segmentation statistics to give a specific score for the sample. The present invention solves the problem that the recognition results of a single segmentation method and a single image network model cannot represent the whole by adopting a method combining a neural network model with statistical principles; effectively suppresses the serious consequences of a single recognition error on the overall fragment recognition, and achieves the effect of rapid recognition and simulated manual overall evaluation.

Description

Technical Field

[0001] The present invention relates to the field of microecological morphological analysis, and in particular to a Gram staining slide overall Nugent analysis method and system. Background Art

[0002] Bacterial vaginosis is a clinical syndrome characterized by a decrease or disappearance of lactobacilli and an increase in related microorganisms. The Nugent scoring method is a relatively objective and accurate method for diagnosing bacterial vaginosis, and has become the gold standard for the diagnosis of this indicator. Since the target bacterial flora is large in variety and mostly small and fragmented, the traditional segmentation method may cause misjudgment due to problems such as adhesion and staining. The target segmentation method requires a large amount of manual marking, which is cumbersome and has unclear standards. Since the diagnosis is a comprehensive evaluation of multiple fields of view of biological microscope images, the above two methods based on bacterial yin and yang and quantity will be affected by factors such as staining and uneven smearing, and are difficult to adapt to clinical evaluation.

[0003] Therefore, the present invention divides the scoring into modules such as normal, inhibition, BV, and BV intermediate according to the Nugent scoring standard of Gram staining, uses a neural network as the modular result, and combines segmentation statistics to perform scoring, which effectively avoids the effects caused by uneven smearing and staining, and performs well in clinical experiments.

[0004] Patent document CN110895968A (application number: CN201910334631.X) discloses an artificial intelligence medical image automatic diagnosis system and method, which collects medical microscope images and corresponding diagnostic data, annotates the medical microscope images to obtain the annotated data corresponding to the medical microscope images, constructs training sets and test sets based on the diagnostic data and annotated data corresponding to the medical microscope images, and obtains the optimal AI classification model and the optimal AI semantic segmentation model based on deep learning model training, so as to realize the automatic diagnosis of the medical microscope images of the test samples. However, the invention is difficult to label, the network is complex, and the result is a single image result.

[0005] Patent document CN106033540B (application number: CN201610363093.3) discloses a method and system for automatic analysis of vaginal microecological morphology. For low-resolution smear images obtained by conventional equipment, super-resolution amplification and reconstruction are first performed to make them clearer, laying the foundation for subsequent classification, and then neural network technology is used to judge and identify the reconstructed high-resolution images, and the types and densities of microorganisms are output in one step to achieve the purpose of automatic analysis and computer-assisted identification of vaginal microecological morphology, so as to make objective and accurate clinical evaluation of vaginal microecological system. However, the invention is difficult to label, the network is complex, and the result is a single picture result.

[0006] Patent document CN107669698A (application number: CN201711137885.X) discloses a method for treating or preventing bacterial vaginosis, a method for preventing the recurrence of bacterial vaginosis, and a method for alleviating or preventing the symptoms or diagnostic criteria of bacterial vaginosis. The method comprises administering an effective amount of a macromolecule, the macromolecule comprising a polylysine dendrimer, a polyamidoamine dendrimer, a poly(etherhydroxylamine) dendrimer or a poly(propyleneimine) dendrimer and one or more sulfonic acid-containing portions connected thereto. However, when evaluating the invention, the target flora is more diverse and mostly tiny and broken bacteria. The traditional segmentation method will cause misjudgment due to problems such as adhesion and staining. The target segmentation method requires a large amount of manual labeling, which is cumbersome and the standard is unclear. Summary of the invention

[0007] In view of the defects in the prior art, the object of the present invention is to provide a Gram-stained slide whole Nugent analysis method and system.

[0008] A Gram staining slide whole Nugent analysis method provided by the present invention comprises:

[0009] Step S1: classify the images collected by the microscope;

[0010] Step S2: Using a neural network to perform classification training on the classified images to obtain a classification model;

[0011] Step S3: Use the trained model to identify each image collected from a single sample;

[0012] Step S4: Count the recognition results of a single image, and obtain the score segment of the entire sample based on the statistical analysis results;

[0013] Step S5: Combine the segmentation statistics to give a specific score for the sample.

[0014] Preferably, in step S1:

[0015] The images collected by the microscope were classified and labeled according to the segmentation method of manual Nugent scoring;

[0016] After dividing the score segments into modules according to the scoring criteria, the collected images will be manually marked.

[0017] Preferably, in step S2:

[0018] A neural network is used to classify and train the labeled images.

[0019] Preferably, in step S3:

[0020] Use the trained model to identify each image collected from a single sample;

[0021] Each sample contains multiple images that need to be identified, and each image will have a separate scoring result.

[0022] Preferably, in step S4:

[0023] The results of all images collected from the entire sample were summarized and statistically analyzed;

[0024] The recognition result is the model recognition result;

[0025] Statistical analysis results: The number of each category is counted according to the modules, and the overall score range is determined based on the proportion.

[0026] A Gram staining slide whole Nugent analysis system provided by the present invention comprises:

[0027] Module M1: Classify the images collected by the microscope;

[0028] Module M2: Use a neural network to perform classification training on the classified images to obtain a classification model;

[0029] Module M3: Use the trained model to identify each image collected from a single sample;

[0030] Module M4: Count the recognition results of a single image and derive the score range of the entire sample based on the statistical analysis results;

[0031] Module M5: Combined with the segmentation statistics, a specific score for the sample is given.

[0032] Preferably, in the module M1:

[0033] The images collected by the microscope were classified and labeled according to the segmentation method of manual Nugent scoring;

[0034] After dividing the score segments into modules according to the scoring criteria, the collected images will be manually marked.

[0035] Preferably, in the module M2:

[0036] A neural network is used to classify and train the labeled images.

[0037] Preferably, in the module M3:

[0038] Use the trained model to identify each image collected from a single sample;

[0039] Each sample contains multiple images that need to be identified, and each image will have a separate scoring result.

[0040] Preferably, in the module M4:

[0041] The results of all images collected from the entire sample were summarized and statistically analyzed;

[0042] The recognition result is the model recognition result;

[0043] Statistical analysis results: The number of each category is counted according to the modules, and the overall score range is determined based on the proportion.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention solves the problem that the recognition results of a single segmentation method and a single image network model cannot represent the whole by combining a neural network model with statistical principles;

[0046] 2. The present invention effectively suppresses the serious consequences of single-piece recognition errors on overall fragment recognition, achieving the effects of rapid recognition and simulating manual overall evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0048] Figure 1 It is a flow chart of the present invention;

[0049] Figure 2 This is the Nugent scoring standard for Gram staining. DETAILED DESCRIPTION

[0050] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0051] Embodiment 1:

[0052] According to a Gram staining slide overall Nugent analysis method provided by the present invention, Figure 1-2 As shown, including:

[0053] Step S1: classify the images collected by the microscope;

[0054] Step S2: Using a neural network to perform classification training on the classified images to obtain a classification model;

[0055] Step S3: Use the trained model to identify each image collected from a single sample;

[0056] Step S4: Count the recognition results of a single image, and obtain the score segment of the entire sample based on the statistical analysis results;

[0057] Step S5: Combine the segmentation statistics to give a specific score for the sample.

[0058] Specifically, in step S1:

[0059] The images collected by the microscope were classified and labeled according to the segmentation method of manual Nugent scoring;

[0060] After dividing the score segments into modules according to the scoring criteria, the collected images will be manually marked.

[0061] Specifically, in step S2:

[0062] A neural network is used to classify and train the labeled images.

[0063] Specifically, in step S3:

[0064] Use the trained model to identify each image collected from a single sample;

[0065] Each sample contains multiple images that need to be identified, and each image will have a separate scoring result.

[0066] Specifically, in step S4:

[0067] The results of all images collected from the entire sample were summarized and statistically analyzed;

[0068] The recognition result is the model recognition result;

[0069] Statistical analysis results: The number of each category is counted according to the modules, and the overall score range is determined based on the proportion.

[0070] Embodiment 2:

[0071] Embodiment 2 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0072] Those skilled in the art may understand the Gram staining slide whole Nugent analysis method provided by the present invention as a specific implementation of the Gram staining slide whole Nugent analysis system, that is, the Gram staining slide whole Nugent analysis system may be implemented by executing the step flow of the Gram staining slide whole Nugent analysis method.

[0073] A Gram staining slide whole Nugent analysis system provided by the present invention comprises:

[0074] Module M1: Classify the images collected by the microscope;

[0075] Module M2: Use a neural network to perform classification training on the classified images to obtain a classification model;

[0076] Module M3: Use the trained model to identify each image collected from a single sample;

[0077] Module M4: Count the recognition results of a single image and derive the score range of the entire sample based on the statistical analysis results;

[0078] Module M5: Combined with the segmentation statistics, a specific score for the sample is given.

[0079] Specifically, in the module M1:

[0080] The images collected by the microscope were classified and labeled according to the segmentation method of manual Nugent scoring;

[0081] After dividing the score segments into modules according to the scoring criteria, the collected images will be manually marked.

[0082] Specifically, in the module M2:

[0083] A neural network is used to classify and train the labeled images.

[0084] Specifically, in the module M3:

[0085] Use the trained model to identify each image collected from a single sample;

[0086] Each sample contains multiple images that need to be identified, and each image will have a separate scoring result.

[0087] Specifically, in the module M4:

[0088] The results of all images collected from the entire sample were summarized and statistically analyzed;

[0089] The recognition result is the model recognition result;

[0090] Statistical analysis results: The number of each category is counted according to the modules, and the overall score range is determined based on the proportion.

[0091] Embodiment 3:

[0092] Embodiment 3 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0093] A method combining multiple simple neural networks and statistics was used to simulate the manual evaluation process. The images were divided into modules according to the manual evaluation criteria. Then, specific scores were given based on the specific number of bacterial flora to conduct an overall evaluation of the vaginal microecological flora.

[0094] ①Network recognition results in a single image

[0095] ②Summarize the results of each picture collected from each sample (usually about 80 pictures)③According to the module interval based on the proportion

[0096] ④Give a specific score (1-10 points) based on the number of divisions

[0097] Step 1: Classify the images collected by the microscope according to the segmentation method of manual Nugent scoring.

[0098] Manual Nugent scoring refers to the process in which a clinician scores a patient's sample according to the Gram staining Nugent scoring criteria. Figure 1 In the figure, the average number of each type of bacteria observed in every 10 oil immersion fields is calculated and scored. The total score is the sum of the scores of the four bacteria; normal: 1 to 3 points, BV intermediate: 4 to 6 points, BV: 7 to 10 points.

[0099] The classification is divided into modules according to the score segments, using a simple CNN network.

[0100] Step 2: Use a neural network to classify the images using a simple CNN network training.

[0101] Step 3: Count the model classification results of all images scanned from the entire slide.

[0102] The model classification result is the result of using the trained model to identify pictures that the network has never seen. There will be one result for each picture in each sample.

[0103] Step 4: Perform statistical analysis on the classification results and print out the scoring results for the entire slide based on the analysis results.

[0104] Statistical analysis: Due to the uneven smearing of samples, the collection field of view on a sample will be different, and therefore there will be different results. The present invention adopts the method of statistically analyzing the results by modules, and then determines the overall segmentation according to the proportion, and then combines the segmentation statistics to determine the specific score.

[0105] The analysis result is a sample score segment based on the recognition result and proportion statistics;

[0106] The scoring result is a specific score obtained by combining the number of segmentations.

[0107] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0108] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A Gram-stained slide whole Nugent analysis method, characterized in that: include: Step S1: classify the images collected by the microscope; Step S2: Using a neural network to perform classification training on the classified images to obtain a classification model; Step S3: Use the trained model to identify each image collected from a single sample; Step S4: Count the recognition results of a single image, and obtain the score segment of the entire sample based on the statistical analysis results; Step S5: combining the segmentation statistics to give a specific score for the sample; In step S1: The images collected by the microscope were classified and labeled according to the segmentation method of manual Nugent scoring; After the score segments are divided into modules according to the scoring criteria, the collected images are manually marked; In step S2: Use neural network to classify and train the labeled images; In step S3: Use the trained model to identify each image collected from a single sample; Each sample contains multiple images that need to be identified, and each image will have a separate scoring result; In step S4: The results of all images collected from the entire sample were summarized and statistically analyzed; The recognition result is the model recognition result; Statistical analysis results: The number of each category is counted according to the modules, and the overall score range is determined based on the proportion.

2. A Gram-stained slide whole Nugent analysis system, characterized in that: include: Module M1: Classify the images collected by the microscope; Module M2: Use a neural network to perform classification training on the classified images to obtain a classification model; Module M3: Use the trained model to identify each image collected from a single sample; Module M4: Count the recognition results of a single image and derive the score range of the entire sample based on the statistical analysis results; Module M5: Combine the segmentation statistics to give the specific score of the sample; In the module M1: The images collected by the microscope were classified and labeled according to the segmentation method of manual Nugent scoring; After the score segments are divided into modules according to the scoring criteria, the collected images are manually marked; In the module M2: Use neural network to classify and train the labeled images; In the module M3: Use the trained model to identify each image collected from a single sample; Each sample contains multiple images that need to be identified, and each image will have a separate scoring result; In the module M4: The results of all images collected from the entire sample were summarized and statistically analyzed; The recognition result is the model recognition result; Statistical analysis results: The number of each category is counted according to the modules, and the overall score range is determined based on the proportion.

Images