Intelligent identification system and method for vaginal endoscope

Abstract

The invention discloses an intelligent recognition system and method for vaginal endoscopy, and relates to the technical field of medical instruments, and the system comprises a vaginal dilator, a camera, a processor and a display; the camera is arranged in the head area of the vaginal dilator, and the processor is connected with the camera and the displayer. The camera is used for collecting image data in the vagina; the processor is used for identifying the vagina health condition of the patient based on the image data and the multi-dimensional health record information of the patient to obtain an identification result; and the display is used for displaying the identification result in real time. Based on the intelligent recognition system, image data in the vagina of the patient and multi-dimensional health record information of the patient are combined, the health condition of the vagina of the patient is recognized, the basic data dimension is enriched, the personal physique difference of the patient is also considered, and the accuracy of the final recognition result is improved.

Description

Technical Field

[0001] The present application relates to the field of medical device technology, and in particular to an intelligent recognition system and method for vaginal endoscopy. Background Art

[0002] In the medical field, traditional vaginal endoscopy mainly uses an endoscope to observe the internal conditions of the vagina, and relies on the doctor's experience to judge the health of the vagina. With the development of technology, image recognition technology has been widely used. However, most existing technologies focus on the image data itself. For example, to improve the accuracy of image recognition, this method often isolates image data from other types of data. Relying only on image data, the final recognition result lacks the support of multi-dimensional data. Therefore, the reliability needs to be improved. Summary of the invention

[0003] The present application discloses an intelligent recognition system and method for vaginal endoscopy, which is used to identify the health status of the patient's vagina by combining image data of the inside of the patient's vagina and the patient's multi-dimensional health record information. This not only enriches the basic data dimensions, but also takes into account the individual physical differences of the patients, thereby improving the accuracy of the final recognition results.

[0004] On the one hand, an embodiment of the present application provides an intelligent recognition system for vaginal endoscopy, including a vaginal dilator, a camera, a processor, and a display;

[0005] A camera, used for collecting image data inside the vagina when the vaginal dilator is inserted into the vagina;

[0006] A processor, used for reshaping the multi-dimensional fusion feature vector corresponding to the image data and the multi-dimensional health record information of the patient into a two-dimensional feature map, performing data manifold transformation on the two-dimensional feature map to obtain a one-dimensional flattened vector, and obtaining an intravaginal recognition result based on the one-dimensional flattened vector;

[0007] The display is used to display the recognition results in real time.

[0008] In a possible embodiment, the vaginal dilator is designed in a bouquet mirror type, including a plurality of blades, and the opening angle of the blades is adjustable to meet the needs of different patients.

[0009] In a possible embodiment, the system further includes a lighting device and a controller;

[0010] A lighting device is disposed in the head region of the vaginal dilator for enhancing visibility within the vagina;

[0011] The controller is connected to the lighting device and is used to adjust the light intensity of the lighting device.

[0012] In a possible embodiment, the processor includes a feature extraction module;

[0013] The feature extraction module is used to extract feature vectors of multiple different dimensions corresponding to image data and multidimensional health record information, and to concatenate the feature vectors of multiple different dimensions on the feature dimension to obtain a fused feature vector; wherein the dimension of the fused feature vector is equal to the sum of the corresponding dimensions of the feature vectors of multiple different dimensions respectively. The multidimensional health record information includes the patient's biological characteristics, clinical information, living habit data, living environment information, and psychosocial data.

[0014] In a possible embodiment, the processor further includes a trained multi-layer perceptron MLP model;

[0015] The MLP model is used to reshape the fused feature vector into a two-dimensional feature map and perform data manifold transformation on the two-dimensional feature map to obtain a one-dimensional flattened vector; the data manifold transformation includes convolution calculation, activation function calculation, and pooling calculation;

[0016] The recognition result inside the vagina is obtained based on the one-dimensional flattened vector.

[0017] In a possible embodiment, the processor further includes a preprocessing module, and the preprocessing module is used to:

[0018] Normalizing, resizing, and enhancing the image data to obtain first data;

[0019] Numerical processing is performed on the multi-dimensional health record information to obtain second data;

[0020] The first data and the second data are standardized to obtain standardized data that matches the input of the feature extraction module.

[0021] In a possible embodiment, the feature extraction module includes a ResNet unit and a splicing unit;

[0022] ResNet unit is used to extract features from multi-dimensional health record information and obtain feature vectors of multiple different dimensions;

[0023] The splicing unit is used to set corresponding weight values ​​for feature vectors of multiple different dimensions based on the correlation between the patient's chief complaint information and the multi-dimensional health record information, and to splice the feature vectors after the weight values ​​are set in the feature dimension to obtain a fused feature vector.

[0024] In one possible embodiment, the MLP model includes an input layer, a hidden layer, and an output layer:

[0025] The input layer is used to receive the fused feature vector and pass it to the hidden layer;

[0026] Based on the hidden layer, it is used to reshape the fused feature vector into a two-dimensional feature map, the size of which is equal to the total dimension of the fused feature vector; convolution calculation, activation function calculation, and pooling calculation are performed on the two-dimensional feature map in sequence, and the pooling calculation result is flattened to obtain a one-dimensional flattened vector; the one-dimensional flattened vector is fully connected and the calculation result is transmitted to the output layer;

[0027] Based on the output layer, the one-dimensional flattened vector is converted into the probability distribution of each classification result, and the probability distribution is used as the recognition result.

[0028] Optionally, the convolution calculation includes:

[0029] For each coordinate position of the two-dimensional feature map, perform the following operations:

[0030] Slide the convolution kernel to a coordinate position on the two-dimensional feature map so that each element of the convolution kernel corresponds to a local area of ​​the two-dimensional feature map;

[0031] Multiply multiple weights in the convolution kernel by the feature vector elements corresponding to the multiple weights at a local position to obtain multiple products;

[0032] Sum multiple products to get a temporary accumulated value, and add the accumulated value to the corresponding bias term to get the element of the output feature map.

[0033] Optionally, the activation function calculation includes:

[0034] Determine whether the elements of the output feature map are positive;

[0035] If so, the elements of the output feature map are used as the calculation results of the activation function;

[0036] If not, the result of the activation function calculation is recorded as 0.

[0037] On the one hand, an embodiment of the present application provides an intelligent recognition method for vaginal endoscopy, comprising:

[0038] collecting image data of the interior of the patient's vagina;

[0039] Based on the image data and the patient's multi-dimensional health record information, the patient's vaginal health condition is identified to obtain an identification result; the identification result is visualized to observe the internal condition of the vagina in real time.

[0040] The beneficial effects of this application are as follows:

[0041] The embodiment of the present application provides an intelligent recognition system and method for vaginal endoscopy. Based on the intelligent recognition system, the health status of the patient's vagina is identified in combination with the image data of the patient's vagina and the patient's multi-dimensional health file information. This not only enriches the basic data dimensions, but also takes into account the individual physical differences of the patient, thereby improving the accuracy of the final recognition result. In addition, when processing multi-dimensional feature vectors, the intelligent recognition system can convert complex multi-dimensional data into one-dimensional flattened vectors through data manifold transformation technology, which is convenient for machine learning model processing, thereby improving recognition speed and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is one of the structural schematic diagrams of an intelligent recognition system for vaginal endoscopy provided in this application;

[0043] Figure 2 The second structural diagram of an intelligent recognition system for vaginal endoscopy provided by the present application;

[0044] Figure 3 The third structural diagram of an intelligent recognition system for vaginal endoscopy provided in this application;

[0045] Figure 4 The fourth structural diagram of an intelligent recognition system for vaginal endoscopy provided in this application;

[0046] Figure 5 The fifth structural diagram of an intelligent recognition system for vaginal endoscopy provided in this application;

[0047] Figure 6 A flowchart of the steps of an intelligent identification method for vaginal endoscopy provided in this application.

[0048] 101. Vaginal dilator, 102. Camera, 103. Processor, 104. Display, 201. Lighting device, 202. Controller, 301. Extraction module, 302. Multi-layer perceptron model, 302A. Input layer, 302B. Hidden layer, 302C. Output layer, 401. Preprocessing module, 501. ResNet unit, 502. Splicing unit. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The specific operating methods in the method embodiments can also be applied to device embodiments or system embodiments. It should be noted that in the description of the present application, "multiple" is understood as "at least two". "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A is connected to B, which can represent: A is directly connected to B and A is connected to B through C. In addition, in the description of the present application, words such as "first" and "second" are only used to distinguish the purpose of description, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order.

[0050] The embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0051] In the medical field, traditional vaginal endoscopy mainly uses an endoscope to observe the internal conditions of the vagina, and relies on the doctor's experience to judge the health of the vagina. With the development of technology, image recognition technology has been widely used. However, most existing technologies focus on the image data itself. For example, to improve the accuracy of image recognition, this method often isolates image data from other types of data. Relying only on image data, the final recognition result lacks the support of multi-dimensional data. Therefore, the reliability needs to be improved.

[0052] In order to solve the above problems, on the one hand, the embodiments of the present application provide an intelligent recognition system for vaginal endoscopy, such as Figure 1 The figure shows a schematic diagram of the structure of an intelligent recognition system. Figure 1 , an intelligent recognition system, including a vaginal dilator 101, a camera 102, and an endoscope image processor, the endoscope image processor including a processor 103 and a display 104; the camera 102 is disposed in the head area of ​​the vaginal dilator 101 ( Figure 1 The black area in the figure), the processor 103 is connected to the camera 102 and the display 104 respectively; wherein the camera 102 is used to collect image data in the vagina when the vaginal dilator 101 is inserted into the vagina; the processor 103 is used to identify the vaginal health status of the patient based on the image data and the multi-dimensional health file information of the patient to obtain an identification result; the display 104 is used to display the identification result in real time.

[0053] Among them, multidimensional health record information includes the patient's biological characteristics, clinical information, lifestyle data, living environment information, and psychosocial data; biological characteristic information includes pathology reports, laboratory test results (blood tests, urine tests), genetic data, etc., clinical information includes medical history data, physiological measurement data, etc., lifestyle data includes eating habits, exercise habits, smoking and drinking conditions, etc., and living environment information includes environmental factors such as chemicals and radiation.

[0054] The specific recognition method of processor 103 includes: reshaping the multi-dimensional fusion feature vector corresponding to the image data and the patient's multi-dimensional health record information into a two-dimensional feature map, and then performing data manifold transformation on the two-dimensional feature map to obtain a one-dimensional flattened vector, and obtaining the recognition result inside the vagina based on the one-dimensional flattened vector.

[0055] Based on this intelligent recognition system, the health status of the patient's vagina is identified by combining the image data of the patient's vagina and the patient's multi-dimensional health record information. This not only enriches the basic data dimensions, but also takes into account the individual physical differences of the patient, improving the accuracy of the final recognition result. In addition, when processing multi-dimensional feature vectors, the intelligent recognition system can convert complex multi-dimensional data into one-dimensional flattened vectors through data manifold transformation technology, which is convenient for machine learning model processing, thereby improving recognition speed and accuracy.

[0056] Optionally, the vaginal dilator has a bouquet mirror design, including multiple blades, the opening angle of the blades can be adjusted to suit the needs of different patients.

[0057] In a possible embodiment, Figure 2 The figure shows a schematic diagram of the structure of an intelligent recognition system. Figure 2 In the figure, the intelligent identification system also includes a lighting device 201 and a controller 202, wherein the lighting device 201 is arranged in the head area of ​​the vaginal dilator 101, is in the shape of a ring belt, and is embedded. The circles inside the belt represent light beads, which are used to enhance the visibility in the vagina. Of course, the shape and installation method of the lighting device 201 here are only examples and are not specifically limited; the controller 202 is connected to the lighting device 201 and is used to adjust the light intensity of the lighting device 202.

[0058] In a possible embodiment, Figure 3 The figure shows a schematic diagram of the structure of an intelligent recognition system. Figure 3 In the embodiment, the processor 103 includes a feature extraction module 301,

[0059] The feature extraction model 301 is used to extract feature vectors of multiple different dimensions corresponding to the image data and the multi-dimensional health record information.

[0060] For example, the feature vector F corresponding to the extracted image data image , the dimension is [batch_size, channels_image, height, width]; the feature vector F corresponding to the clinical information clinical , the dimension is [batch_size,num_clinical_features]; the feature vector F corresponding to the living habit data lifestyle , dimension is [batch_size,num_lifestyle_features]; the feature vector F corresponding to the living environment information environment , dimension is [batch_size,num_environment_features]; the feature vector F corresponding to the psychosocial data psychosocial , with a dimension of [batch_size,num_psychosocial_features].

[0061] Then, it is used to concatenate multiple feature vectors of different dimensions on the feature dimension to obtain a fused feature vector; wherein the dimension of the fused feature vector is equal to the sum of the corresponding dimensions of the feature vectors of multiple different dimensions, wherein the dimension calculation formula of the fused feature vector is:

[0062] D combined =D image +D clinical +D biological +D lifestyle +D environment +D psy

[0063] chosocial (1)

[0064] In formula (1), D combined is the total dimension of the fused feature vector, D image is the feature vector dimension corresponding to the image data, D biological is the feature vector dimension corresponding to the biological feature, D clinical is the feature vector dimension corresponding to clinical information, D lifestyle is the feature vector dimension corresponding to the living habit data, D environment is the feature vector dimension corresponding to the residential environment information, D psychosocial is the feature vector dimension corresponding to psychosocial data.

[0065] Optional, reference Figure 3 The processor 103 also includes a trained multilayer perceptron (MLP) model 302.

[0066] The MLP model 302 is used to identify the fused feature vector to obtain a recognition result; wherein the MLP model 302 includes an input layer 302A, a hidden layer 302B and an output layer 302C, the dimension of the input layer 302A is the same as the total dimension of the fused feature vector, and the hidden layer 302B and the output layer both have at least one neuron.

[0067] In this embodiment, the input layer 302A receives the fused feature vectors transmitted by the extraction module 301, and then transmits these feature vectors to the hidden layer 302B for processing, including convolution operations, nonlinearization, pooling operations, full connection, etc., and transmits the processing results to the output layer 302C. The output layer 302C obtains the recognition results of the internal vaginal data based on the processing results of 302B, including: abnormal or normal.

[0068] In a possible embodiment, Figure 4 The figure shows a schematic diagram of the structure of an intelligent recognition system. Figure 4 In the embodiment, the processor 103 also includes a preprocessing module 401, which is used to: normalize, resize and enhance the image data to obtain the first data; perform numerical processing on the multi-dimensional health record information to obtain the second data; and standardize the first data and the second data to obtain standardized data that matches the input of the feature extraction module.

[0069] In a possible embodiment, Figure 5 The figure shows a schematic diagram of the structure of an intelligent recognition system. Figure 5 In the figure, the feature extraction module 301 includes a ResNet unit 501 and a splicing unit 502.

[0070] ResNet unit 501 is used to extract features from multi-dimensional health record information, including image data, biological features, clinical information, lifestyle data, living environment information, and psychosocial data, to obtain feature vectors of multiple different dimensions.

[0071] The splicing unit 502 is used to set corresponding weight values ​​for feature vectors of multiple different dimensions based on the correlation between the patient's chief complaint information and the multi-dimensional health record information, and to splice the feature vectors after the weight values ​​are set in the feature dimension to obtain a fused feature vector.

[0072] Among them, the patient's chief complaint information mainly includes the patient's current symptom information and the doctor's advice. If a certain dimension of information in the multidimensional health record information is strongly related to the patient's chief complaint information, the weight value of the dimension information will be relatively large, otherwise, the weight value of the dimension information will be reduced.

[0073] The weighted feature vectors of multiple different dimensions are fused based on the splicing model to obtain a fused feature vector. The specific formula of the splicing model is:

[0074] F combined =Concat(a*F image ,b*F biological ,c*F clinical ,d*F lifestyle ,e*F environment ,

[0075] f*F psychosocial )(2)

[0076] In formula (2), Concat is a concatenation function, and a, b, c, d, e, and f are weight values. Before concatenation, each feature vector can be flattened to make the dimensions of each feature vector consistent, thereby facilitating concatenation.

[0077] In this technical solution, different weights are set for feature vectors of different dimensions according to the patient's chief complaint information, which helps doctors provide patients with more personalized treatment plans.

[0078] In a possible embodiment, the MLP model is specifically used to: after the input layer obtains the fused feature vector, pass the fused feature vector to the hidden layer; based on the hidden layer, reshape the fused feature vector into a two-dimensional feature map, the size of the two-dimensional feature map is equal to the total dimension of the fused feature vector. For example, if the dimension of the fused feature vector is D combined , which can be reshaped into a two-dimensional feature map of H×W, where H×W×C=D combined , H and W are the height and width of the feature map respectively, C is the number of channels, that is, the depth of the feature map; perform convolution calculation, activation function calculation, pooling calculation on the two-dimensional feature map in turn, and flatten the final calculation result to obtain a one-dimensional flattened vector; perform full connection calculation on the one-dimensional flattened vector, and transmit the calculation result to the output layer; based on the output layer, convert the one-dimensional flattened vector into the probability distribution of each classification result, and use the probability distribution as the recognition result.

[0079] Optionally, the convolution calculation method includes: for each coordinate position of the two-dimensional feature map, performing the following operations respectively:

[0080] Slide the convolution kernel to a coordinate position on the two-dimensional feature map so that each element of the convolution kernel corresponds to a local area of ​​the two-dimensional feature map;

[0081] Multiply multiple weights in the convolution kernel by the feature vector elements corresponding to the multiple weights at a local position to obtain multiple products;

[0082] Sum multiple products to get a temporary accumulated value, and add the accumulated value to the corresponding bias term to get the element of the output feature map.

[0083] Optionally, the size of the convolution kernel is K×K×C, and the convolution calculation formula is:

[0084]

[0085] In formula (3), F out (i, j, k) represents the element of the output feature map after convolution calculation, K is the height and width of the convolution kernel, C is the depth of the convolution kernel, P m,n,c is the weight of the convolution kernel, X i+m,j+n,c is the feature vector element corresponding to the two-dimensional feature image, b k is the bias term, i, j are the position coordinates of the output feature map, and k is the index of the output channel.

[0086] Optionally, the calculation method of the activation function calculation includes: determining whether an element of the output feature map is a positive value; if so, using the element of the output feature map as a calculation result of the activation function calculation; if not, recording the calculation result of the activation function calculation as 0.

[0087] Specifically, the calculation formula for the activation function is:

[0088] A out (i, j, k) = max(0, F out (i, j, k)) (4)

[0089] Optionally, the calculation formula for pooling calculation is:

[0090] P out (i, j, k) = max m∈M max n∈N A out (i·s+m,j·s+n,k) (5)

[0091] In formula (5), s is the step size, and M and N are the sizes of the pooling windows.

[0092] Optionally, the calculation formula for full connection calculation is:

[0093] O=P′·F flat +b′ (6)

[0094] In formula (6), P' is the weight of the fully connected layer, F flat is the one-dimensional flattened vector, and b' is the bias term.

[0095] Optionally, the conversion formula for the probability distribution is:

[0096]

[0097] In formula (7), y z represents the probability of the zth result category, T is the number of result categories, O z is the output score of the zth neuron.

[0098] In the embodiment of the present application, the output result categories are 2 categories, abnormal and normal; of course, in the case of abnormality, it can also be subdivided to identify the specific type of disease.

[0099] On the other hand, the embodiment of the present application provides an intelligent recognition method for vaginal endoscopy, based on any of the above systems, such as Figure 6 FIG. 1 is a flowchart of a method for intelligent identification of vaginal endoscopy, which specifically includes:

[0100] S601, collecting image data of the interior of the patient's vagina;

[0101] S602, based on the image data and the patient's multi-dimensional health file information, identify the patient's vaginal health status to obtain an identification result; wherein the multi-dimensional health file information includes the patient's biological characteristics, clinical information, living habit data, living environment information, and psychosocial data;

[0102] S603, visually displaying the recognition results to observe the internal conditions of the vagina in real time.

[0103] In an embodiment of the present application, when the vaginal dilator is inserted into the patient's vagina, the camera provided on the head of the vaginal dilator is used to collect image data of the patient's vagina. When the patient has health problems in the vagina, these image data may contain some lesion information, so that the health condition of the patient's vagina can be identified through the image data.

[0104] In order to improve the accuracy of recognition, after obtaining the image data of the patient's vagina, the embodiment of the present application further obtains the patient's multi-dimensional health record information, which includes the patient's biological characteristics, clinical information, lifestyle data, living environment information, and psychosocial data; the biological characteristic information includes pathology reports, laboratory test results (blood tests, urine tests), genetic data, etc., the clinical information includes medical history data, physiological measurement data, etc., the lifestyle data includes eating habits, exercise habits, smoking and drinking conditions, etc., and the living environment information includes the exposure history to environmental factors such as chemicals and radiation.

[0105] Furthermore, the image data and multi-dimensional health record data are combined as the basis for identifying the patient's genital health status. Specifically:

[0106] First, the image data and multidimensional health record data are preprocessed, including: normalizing, resizing and data enhancement operations on the image data to obtain the first data; numerically processing the multidimensional health record information to obtain the second data; and standardizing the first data and the second data to obtain standardized data that matches the input of the feature extraction module.

[0107] Secondly, feature extraction is performed on the preprocessed image data and multi-dimensional health record information to obtain feature vectors of multiple different dimensions; the feature vectors of multiple different dimensions are concatenated on the feature dimension to obtain a fused feature vector.

[0108] Specifically, the ResNet unit is used to extract features of image data, biological features, clinical information, living habit data, living environment information, and psychosocial data, respectively, to obtain feature vectors of multiple different dimensions; the preset splicing model is used to fuse the feature vectors of multiple different dimensions based on the splicing model to obtain a fused feature vector, wherein the specific formula of the splicing model refers to the above formula (2), which will not be repeated here.

[0109] Again, the MLP model is used to identify the fused feature vector to obtain a recognition result; wherein the MLP model includes an input layer, a hidden layer and an output layer, the dimension of the input layer is the same as the total dimension of the fused feature vector, the hidden layer has at least one neuron, and the output layer has at least one neuron.

[0110] Specifically, after the input layer obtains the fused feature vector, the fused feature vector is passed to the hidden layer; based on the hidden layer, the fused feature vector is reshaped into a two-dimensional feature map, and the size of the two-dimensional feature map is equal to the total dimension of the fused feature vector; convolution calculation, activation function calculation, pooling calculation are performed on the two-dimensional feature map in sequence, and the final calculation result is flattened to obtain a one-dimensional flattened vector; the one-dimensional flattened vector is fully connected, and the calculation result is transmitted to the output layer.

[0111] Among them, the specific calculation process of convolution calculation, activation function calculation, and pooling calculation can be found in the above system part and will not be repeated here.

[0112] Finally, based on the output layer, the one-dimensional flattened vector is converted into the probability distribution of each classification result, and the probability distribution is used as the recognition result.

[0113] Based on this intelligent recognition system, the health condition of the patient's vagina is identified by combining the image data of the patient's vagina and the patient's multi-dimensional health record information. This not only enriches the basic data dimensions, but also takes into account the individual physical differences of the patient, thereby improving the accuracy of the final recognition results.

[0114] In particular, the preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application, and the embodiments of the present application and the features in the embodiments may be combined with each other if there is no conflict.

[0115] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. An intelligent recognition system for vaginal endoscopy, characterized in that: Includes vaginal dilator, camera, processor and display; The camera is used to collect image data inside the vagina when the vaginal dilator is inserted into the vagina; The processor is used to reshape the multi-dimensional fusion feature vector corresponding to the image data and the multi-dimensional health record information of the patient into a two-dimensional feature map, perform data manifold transformation on the two-dimensional feature map to obtain a one-dimensional flattened vector, and obtain the recognition result of the vagina based on the one-dimensional flattened vector; The display is used to display the recognition result in real time.

2. The system according to claim 1, characterized in that: The vaginal dilator has a bouquet mirror-shaped design and comprises a plurality of blades, the opening angles of which are adjustable to meet the needs of different patients.

3. The system according to claim 1, characterized in that: The system also includes a lighting device and a controller; The lighting device is disposed in the head region of the vaginal dilator to enhance visibility in the vagina; The controller is connected to the lighting device and is used to adjust the light intensity of the lighting device.

4. The system according to claim 1, characterized in that: The processor includes a feature extraction module; The feature extraction module is used to extract feature vectors of multiple different dimensions corresponding to the image data and the multidimensional health record information, and to splice the feature vectors of multiple different dimensions on the feature dimension to obtain the fused feature vector; wherein the dimension of the fused feature vector is equal to the sum of the dimensions corresponding to the feature vectors of the multiple different dimensions respectively, and the multidimensional health record information includes the patient's biological characteristics, clinical information, living habit data, living environment information, and psychosocial data.

5. The system according to claim 1, characterized in that: The processor also includes a trained multi-layer perceptron MLP model; The MLP model is used to reshape the fused feature vector into the two-dimensional feature map, and perform data manifold transformation on the two-dimensional feature map to obtain a one-dimensional flattened vector; wherein the data manifold transformation includes convolution calculation, activation function calculation, and pooling calculation; The recognition result inside the vagina is obtained based on the one-dimensional flattened vector.

6. The system according to claim 4, characterized in that: The feature extraction module includes a ResNet unit and a splicing unit; The ResNet unit is used to extract features from the multi-dimensional health record information to obtain feature vectors of the multiple different dimensions; The splicing unit is used to set corresponding weight values ​​for the feature vectors of the multiple different dimensions based on the correlation between the patient's chief complaint information and the multi-dimensional health record information, and to splice the feature vectors after the weight values ​​are set on the feature dimension to obtain the fused feature vector.

7. The system according to claim 4, characterized in that: The MLP model includes an input layer, a hidden layer, and an output layer: The input layer is used to receive the fused feature vector and pass the fused feature vector to the hidden layer; Based on the hidden layer, the fused feature vector is reshaped into a two-dimensional feature map, the size of which is equal to the total dimension of the fused feature vector; convolution calculation, activation function calculation, and pooling calculation are performed on the two-dimensional feature map in sequence, and the pooling calculation result is flattened to obtain the one-dimensional flattened vector; the one-dimensional flattened vector is fully connected and the calculation result is transmitted to the output layer; Based on the output layer, the one-dimensional flattened vector is converted into a probability distribution of each classification result, and the probability distribution is used as the recognition result.

8. The system according to claim 7, characterized in that: The convolution calculation includes: For each coordinate position of the two-dimensional feature map, perform the following operations respectively: Sliding the convolution kernel to a coordinate position on the two-dimensional feature map so that each element of the convolution kernel corresponds to a local area of ​​the two-dimensional feature map; Multiplying the multiple weights in the convolution kernel by the feature vector elements corresponding to the multiple weights at the local position, respectively, to obtain multiple products; The multiple products are summed to obtain a temporary accumulated value, and the accumulated value is added with the corresponding bias term to obtain the element of the output feature map.

9. The system according to claim 8, characterized in that The activation function calculation includes: Determining whether an element of the output feature map is a positive value; If yes, then taking the elements of the output feature map as the calculation result of the activation function calculation; If not, the calculation result of the activation function calculation is recorded as 0.

10. An intelligent identification method for vaginal endoscopy, based on any system in claims 1 to 9, characterized in that: The method comprises: collecting image data of the interior of the patient's vagina; Based on the image data and the multi-dimensional health file information of the patient, identifying the vaginal health condition of the patient to obtain an identification result; The recognition result is displayed visually to observe the internal conditions of the vagina in real time.

Images