Cervical fluorescence image processing and analyzing method and system and storage medium

Abstract

The invention discloses a cervical fluorescence image processing analysis method and system and a storage medium, and the method comprises the steps: extracting the three-primary color value of each pixel point after a cervical fluorescence image is obtained, and generating a plurality of color ratio matrixes of the cervical fluorescence image based on the three-primary color value of each pixel point; according to the proportion matrix of each color, preliminarily analyzing the cervical fluorescence image, and judging whether an abnormal region exists in the cervical fluorescence image or not; if yes, extracting an abnormal region image from the cervix uteri fluorescence image, and converting the abnormal region image into a cervix uteri grey-scale map; and recognizing the cervix uteri grey-scale map based on a preset recognition model to obtain a final recognition result. According to the invention, through a front-stage detection and recognition mechanism and a rear-stage detection and recognition mechanism, when normal cervical fluorescence images are quickly screened out, more detailed and comprehensive image recognition is carried out on the preliminarily abnormal cervical fluorescence images through the preset recognition model formed by convolutional neural network training, so that accurate and efficient analysis of the cervical fluorescence images is realized.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a cervical fluorescence image processing and analysis method, system and storage medium. Background Art

[0002] Cervical cancer is one of the most common malignant tumors of the reproductive tract that occurs in the female cervix. It is mainly squamous cell carcinoma, and the rest are cervical adenocarcinoma, small cell carcinoma, etc. It has been confirmed that long-term or repeated infection with human papillomavirus (HPV) is the main cause of cervical cancer. HPV infection develops into cervical cancer after a precancerous lesion stage. According to the WHO classification of female reproductive system tumors, the precancerous lesions of squamous cell carcinoma are named cervical squamous intraepithelial lesions (SIL), and are classified into two levels: low-grade SIL (LSIL) and high-grade SIL (HSIL). LSILL includes CIN1, p16-negative CIN2 lesions, and condyloma lesions caused by HPV infection; HSIL includes p16-positive CIN2 and CIN3 lesions.

[0003] Accurate and effective examination of precancerous lesions is an important means of cervical cancer screening, which can significantly reduce the incidence of cervical cancer. Commonly used inspection methods include naked eye observation, glacial acetic acid coating observation, conventional colposcopy combined with acetic acid and Lugol's iodine test, Pap smear, liquid-based cytology, etc. However, these methods have a high false negative rate and poor sensitivity, and cannot accurately determine the characteristics of the lesions; moreover, the patient's cervix needs to be contact coated, which may cause discomfort to the patient and introduce other pathogens during the contact process.

[0004] At present, with the development and innovation of medical imaging technology and image processing technology, taking cervical images and processing and analyzing them has gradually become one of the mainstream screening methods for cervical cancer. However, the current processing and analysis of cervical images is often relatively simple. For example, the image is analyzed to see if it is clear, and if it is not clear, the clarity is processed, and then the doctor diagnoses and conducts further examinations. This requires a high level of experience for the doctor and increases the diagnosis time for the patient. Therefore, how to accurately and efficiently analyze and process cervical images is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of the present invention is to provide a cervical fluorescence image processing and analysis method, system and storage medium, aiming to solve the technical problem of how to accurately and efficiently analyze and process cervical images in the prior art.

[0006] To achieve the above object, the present invention provides a method for processing and analyzing cervical fluorescence images, wherein the method comprises:

[0007] After acquiring a cervical fluorescence image based on a preset excitation light, extracting three primary color values ​​of each pixel in the cervical fluorescence image, and generating multiple color proportion matrices of the cervical fluorescence image based on the three primary color values ​​of each pixel;

[0008] Performing a preliminary analysis on the cervical fluorescence image according to each of the color proportion matrices to determine whether there is an abnormal area in the cervical fluorescence image;

[0009] If there is an abnormal area in the cervical fluorescence image, extracting the abnormal area image from the cervical fluorescence image, and converting the abnormal area image into a cervical grayscale image based on the three primary color values ​​of each pixel in the abnormal area image;

[0010] The cervical grayscale image is recognized based on a preset recognition model to obtain a final recognition result.

[0011] Preferably, the step of identifying the cervical grayscale image based on a preset recognition model to obtain a final recognition result includes:

[0012] Performing feature extraction on the cervical grayscale image based on a feature extraction module in a preset recognition model to obtain a feature vector of the cervical grayscale image;

[0013] Based on the classifier in the preset recognition model, the feature vector is classified and calculated to obtain a calculation result, and the calculation formula is:

[0014]

[0015] Where n represents the number of abnormal categories, ai and bi are the weight and bias values ​​of the i-th abnormality in the classifier, H is the feature vector, pi represents the calculation result of the i-th abnormality probability value, j represents the previous j calculations, and wi represents the sum of the classifier errors of the i-th abnormality calculation results in the previous j calculations;

[0016] According to the calculation result, the abnormal type corresponding to the abnormal area in the cervical fluorescence image is determined, and the area size of the abnormal area is identified, and the abnormal type and the area size are generated as the final identification result.

[0017] Preferably, before the step of identifying the cervical grayscale image based on a preset recognition model to obtain a final recognition result, the step includes:

[0018] Acquire cervical image samples, and divide the cervical image samples into training image samples and test image samples;

[0019] The training image samples are transferred to the preset initial model, the preset initial model is trained, and when the number of training times reaches the preset number of times, the loss function value of the preset initial model is calculated, and the calculation formula is:

[0020]

[0021] Where L represents the loss function value, M represents the number of samples in the training image samples that are identified as abnormal after training, N represents the number of samples in the training image samples that are identified as not abnormal after training, and P k1 represents the abnormal category identified by the training of the k1th training image sample, S k1 represents the abnormal category identified by the training of the k1th training image sample, S k1 represents the reference abnormal region corresponding to the k1th training image sample, S k1 ' represents the abnormal area identified by training for the k1th training image sample, P k2 represents the reference category corresponding to the k2th training image sample, P k2 ′ represents the category of the k2th training image sample after training and recognition, S k2 P represents the k2th training image sample k2 'With P k2 On the contrary, the reference abnormal area corresponding to the k2th training image sample;

[0022] Determine whether the loss function value is less than or equal to a preset threshold value, and if so, test the preset initial model based on the test sample image to obtain a test result;

[0023] A test error is generated according to the test result, and the preset recognition model is generated according to the size relationship between the test error and a preset threshold.

[0024] Preferably, the step of generating the preset recognition model according to the magnitude relationship between the test error and a preset threshold value comprises:

[0025] If the size relationship is that the test error is less than a preset threshold, generating the preset initial model as a preset recognition model;

[0026] If the test error is greater than or equal to the preset threshold, the calculation formula of the loss function value is updated, and the preset initial model is iteratively trained based on the updated calculation formula until the test error is less than the preset threshold, wherein the formula for updating the calculation formula is:

[0027]

[0028] Where L′ represents the updated calculation formula, L represents the calculation formula before the update, μ1 and μ2 represent the preset update parameters, and W t represents the weight matrix between the input layer and the convolutional layer in the preset initial model, m represents the number of elements in the weight matrix, It is used to calculate the cumulative value of each element in the weight matrix after taking the absolute value. Represents the transpose of the weight matrix.

[0029] Preferably, the step of performing a preliminary analysis on the cervical fluorescence image according to each of the color proportion matrices to determine whether there is an abnormal area in the cervical fluorescence image comprises:

[0030] Based on a preset comparison formula, the pixel color proportion values ​​in each color proportion matrix are compared with a preset proportion threshold value to generate multiple proportion conversion matrices, wherein the preset comparison formula is:

[0031]

[0032] k=1, 2, 3, x=1, 2, 3...X, y=1, 2, 3...Y;

[0033] Where k represents the three primary colors red, green and blue, X represents the number of rows in the color proportion matrix, Y represents the number of columns in the color proportion matrix, F represents the preset proportion threshold, and f' k (x, y) represents the proportion conversion matrix, f k (x, y) represents the color proportion matrix;

[0034] Performing a sum operation on each of the proportion conversion matrices to obtain a sum matrix, and performing a sum operation on each element value in the sum matrix to obtain a matrix value of the sum matrix;

[0035] It is determined whether the matrix value is greater than a preset matrix threshold value. If the matrix value is greater than the preset matrix threshold value, it is determined that an abnormal area exists in the cervical fluorescence image.

[0036] Preferably, if there is an abnormal area in the cervical fluorescence image, the step of extracting the abnormal area image from the cervical fluorescence image includes:

[0037] If there is an abnormal area in the cervical fluorescence image, searching for a target element whose values ​​of all elements in the sum matrix are zero;

[0038] Target pixel points corresponding to the target elements are found in the cervical fluorescence image, and the target pixel points are removed from the fluorescence image to form the abnormal area image.

[0039] Preferably, before the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image, the step includes:

[0040] Determining whether the cervical fluorescence image includes a complete cervical surface and whether the clarity of the cervical fluorescence image reaches a preset clarity threshold;

[0041] If the cervical fluorescence image includes the complete surface of the cervix and the clarity reaches a preset clarity threshold, the cervical fluorescence image is determined to be a valid image, and the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image is performed;

[0042] If the cervical fluorescence image does not include the complete cervical surface and / or the clarity does not reach a preset clarity threshold, the cervical fluorescence image is determined to be an invalid image, and a prompt message for re-collecting the cervical fluorescence image is output.

[0043] Preferably, the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image includes:

[0044] Generate a grayscale value of each pixel point according to the three primary color values ​​of each pixel point, determine the maximum value and the minimum value of each grayscale value, and perform a mean operation on the maximum value and the minimum value to obtain a calculation result as an initial threshold value;

[0045] Comparing the grayscale value of each pixel with the initial threshold, determining a first grayscale value set greater than the initial threshold among the grayscale values, and a second grayscale value set less than the initial threshold among the grayscale values;

[0046] Based on the first grayscale value set and the second grayscale value set, respectively generate a first average grayscale value and a second average grayscale value;

[0047] updating the initial threshold according to the first average grayscale value and the second average grayscale value, and determining whether the updated initial threshold is the same as the initial threshold before the update;

[0048] If they are the same, each pixel point corresponding to the second gray value set is removed from the cervical fluorescence image to form a new cervical fluorescence image;

[0049] If they are not the same, then based on the updated initial threshold, a step of comparing the grayscale value of each pixel with the initial threshold is performed until the updated initial threshold is the same as the initial threshold before the update.

[0050] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a cervical fluorescence image processing and analysis system, the cervical fluorescence image processing and analysis system comprising a storage device, a processor, a communication bus, and a control program stored in the storage device:

[0051] The communication bus is used to realize the connection and communication between the processor and the storage;

[0052] The processor is used to execute the control program to implement the steps of the cervical fluorescence image processing and analysis method as described above.

[0053] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a storage medium, on which a control program is stored, and when the control program is executed by a processor, the steps of the cervical fluorescence image processing and analysis method as described above are implemented.

[0054] The cervical fluorescence image processing and analysis method, system and storage medium of the present invention, after obtaining the cervical fluorescence image through the preset excitation light, extracts the three primary color values ​​of each pixel in the cervical fluorescence image, and generates multiple color proportion matrices of the cervical fluorescence image according to the three primary color values ​​of each pixel; then, based on each color proportion matrix, the cervical fluorescence image is preliminarily analyzed to determine whether there is an abnormal area in the cervical fluorescence image; if there is an abnormal area, the abnormal area image is segmented from the cervical fluorescence image, and the abnormal area image is converted into a cervical grayscale image based on the three primary color values ​​of each pixel in the abnormal area image; then, the converted cervical grayscale image is recognized by the preset recognition model to obtain the final recognition result. In this way, by setting the front and back two-stage detection and recognition mechanism, the front stage is to generate the acquired cervical fluorescence image into a color proportion matrix for preliminary detection based on the difference in the three primary color values ​​of the pixels between the normal cervical fluorescence image and the abnormal cervical fluorescence image, and the back stage is to re-detect the cervical fluorescence image detected as abnormal by the front stage according to the preset recognition model formed by pre-training a large number of abnormal image samples. In addition, the normal judgment standard in the pre-stage detection is relatively strict. Only when it fully meets the normal standard can the cervical fluorescence image be judged to be normal. In addition, as long as there is a suspected abnormal area, it is judged to be abnormal, and then the post-stage inspection is carried out. On the one hand, normal cervical fluorescence images can be quickly screened out, and on the other hand, the preset recognition model formed by the convolutional neural network training for the preliminary abnormal cervical fluorescence images can perform more detailed and comprehensive image recognition, with higher recognition accuracy, and realize accurate and efficient analysis and processing of cervical fluorescence images. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flowchart of the first embodiment of the cervical fluorescence image processing and analysis method of the present invention;

[0056] Figure 2 It is a flowchart of the second embodiment of the cervical fluorescence image processing and analysis method of the present invention;

[0057] Figure 3It is a flowchart of the third embodiment of the cervical fluorescence image processing and analysis method of the present invention;

[0058] Figure 4 It is a schematic diagram of the structure of the hardware operating environment involved in an embodiment of the cervical fluorescence image processing and analysis system of the present invention.

[0059] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0060] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0061] The present invention provides a cervical fluorescence image processing and analysis method, please refer to Figure 1 , Figure 1 It is a flowchart of the first embodiment of the method for processing and analyzing cervical fluorescence images of the present invention.

[0062] The embodiment of the present invention provides an embodiment of a method for processing and analyzing cervical fluorescence images. It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here. Specifically, the method for processing and analyzing cervical fluorescence images in this embodiment includes:

[0063] Step S10, after acquiring the cervical fluorescence image based on the preset excitation light, extracting the three primary color values ​​of each pixel in the cervical fluorescence image, and generating multiple color proportion matrices of the cervical fluorescence image based on the three primary color values ​​of each pixel.

[0064] The cervical fluorescence image processing and analysis method of this embodiment is applied to a server, and the server processes and analyzes the collected cervical fluorescence image to determine whether it reflects the characteristics of cervical cancer and precancerous lesions. This embodiment collects cervical fluorescence images by means of the autofluorescence characteristics of human tissue. Among them, human tissue can induce fluorescent groups or substances in the tissue, such as collagen, protein, porphyrin, tryptophan, etc., to produce their own intrinsic fluorescence under the irradiation of short-wavelength light such as ultraviolet light, purple light, and blue light, which is autofluorescence, and the short-wavelength light used for irradiation is excitation light. Specifically, a device for collecting cervical fluorescence images is set, and the device can set a short-wavelength light with a good induction effect as a preset laser according to the characteristics of the fluorescent substance of the cervix. At the same time, a shooting component is also set in the device, such as a high-sensitivity CCD (Charge Couple Device) camera. The preset excitation light irradiates the surface of the cervix with a specific wavelength to excite the cervical tissue to produce autofluorescence, and then the fluorescent image of the cervical surface is taken by the shooting component and transmitted to the server, and the server realizes the acquisition of the cervical fluorescence image based on the preset excitation light.

[0065] Furthermore, the server is provided with an extraction tool for extracting the three primary color values, through which the three primary color values ​​of each pixel in the cervical fluorescence image can be extracted. Among them, the three basic colors that cannot be decomposed in color are the three primary colors, including red (R), green (G), and blue (B). The three primary color values ​​of a pixel are the R value, G value, and B value of the pixel. After the three primary color values ​​of each pixel in the cervical fluorescence image are extracted, the three primary color values ​​of each pixel can be arranged according to color to form multiple color proportion matrices of the cervical fluorescence image. Since the number of colors in the three primary color values ​​is three, the color proportion matrices formed are preferably three, namely, the red color proportion matrix, the green color proportion matrix, and the blue color proportion matrix. The specific formation method is that, for each pixel point, the R value, G value and B value of the three primary color values ​​are added to obtain the calculation result, and then the R value, G value and B value are respectively ratioed with the calculation result to obtain the R value ratio, G value ratio and B value ratio of each pixel point, and then the R value ratio of each pixel point is arranged according to the position of each pixel point in the cervical fluorescence image to form a red color ratio matrix, the G value ratio of each pixel point is arranged according to the position of each pixel point in the fluorescent cervical image to form a green color ratio matrix, and the B value ratio of each pixel point is arranged according to the position of each pixel point in the fluorescent cervical image to form a blue color ratio matrix.

[0066] It should be noted that in the process of collecting cervical fluorescence images based on preset excitation light, the collected images may not be clear, or may not contain the complete cervical surface, etc. The three primary colors of the pixels extracted from such images cannot be used as the basis for subsequent analysis and processing. Therefore, before extracting the three primary colors of the pixels, it is necessary to determine the validity of the acquired cervical fluorescence image. Specifically, the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image includes:

[0067] Step a1, determining whether the cervical fluorescence image contains a complete cervical surface, and whether the clarity of the cervical fluorescence image reaches a preset clarity threshold;

[0068] Step a2, if the cervical fluorescence image includes the complete cervical surface and the clarity reaches a preset clarity threshold, the cervical fluorescence image is determined to be a valid image, and the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image is performed;

[0069] Step a3: if the cervical fluorescence image does not include the complete cervical surface and / or the clarity does not reach a preset clarity threshold, the cervical fluorescence image is determined to be an invalid image, and a prompt message for re-collecting the cervical fluorescence image is output.

[0070] Furthermore, the server is provided with a recognition module capable of recognizing the surface contour of the cervix, and the recognition module recognizes the cervical fluorescence image, and determines whether the cervical fluorescence image contains the complete cervical surface by determining whether the image has the surface contour corresponding to the cervix. At the same time, the server is also provided with a detection module for detecting the clarity of the image, and a preset clarity threshold for determining the clarity. The clarity of the cervical fluorescence image is detected by the detection module, and the detected clarity is compared with the preset clarity threshold. By determining whether the detected clarity is greater than the preset clarity threshold, it is determined whether the clarity of the cervical fluorescence image reaches the preset clarity threshold.

[0071] Furthermore, if it is determined through identification that the cervical fluorescence image has a surface contour corresponding to the cervix, it can be determined that the cervical fluorescence image contains a complete cervical surface; and if it is determined through comparison that the detection clarity is greater than a preset clarity threshold, it can be determined that the clarity of the cervical fluorescence image reaches the preset clarity threshold. The cervical fluorescence image is then determined to be a valid image, so that the three primary color values ​​of each pixel point can be extracted from it. On the contrary, if the cervical fluorescence image does not contain a complete cervical surface, or its clarity does not reach the preset clarity threshold, or the cervical fluorescence image neither contains the cervical surface nor reaches the preset clarity threshold, it means that the cervical fluorescence image is an invalid image. At this time, a prompt message for re-collecting the cervical fluorescence image can be output in the form of text or sound, so as to re-collect the valid cervical fluorescence image according to the prompt message for analysis and processing.

[0072] It is understandable that, in addition to the imaging of the cervical surface, the collected cervical fluorescence image also includes the imaging of the tissues around the cervix. This type of tissue has no effect on the analysis of cervical cancer and precancerous lesions, but will increase the amount of data for image processing, affecting the processing efficiency. Therefore, after extracting the three primary color values ​​of each pixel, the background of the cervical fluorescence image can be removed based on the differences between the pixels, so as to retain the area containing the cervical imaging in the cervical fluorescence image for analysis and processing. Specifically, the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image includes:

[0073] Step b1, generating a grayscale value of each pixel point according to the three primary color values ​​of each pixel point, determining a maximum value and a minimum value of each grayscale value, and performing a mean operation on the maximum value and the minimum value to obtain a calculation result as an initial threshold value;

[0074] Step b2, comparing the grayscale value of each pixel with the initial threshold, determining a first grayscale value set greater than the initial threshold among the grayscale values, and a second grayscale value set less than the initial threshold among the grayscale values;

[0075] Step b3, generating a first average grayscale value and a second average grayscale value based on the first grayscale value set and the second grayscale value set respectively;

[0076] Step b4, updating the initial threshold according to the first average gray value and the second average gray value, and determining whether the updated initial threshold is the same as the initial threshold before the update;

[0077] Step b5, if they are the same, then removing each pixel point corresponding to the second gray value set from the cervical fluorescence image to form a new cervical fluorescence image;

[0078] Step b6: If they are not the same, then based on the updated initial threshold, the step of comparing the grayscale value of each pixel with the initial threshold is performed until the updated initial threshold is the same as the initial threshold before the update.

[0079] Furthermore, a grayscale conversion formula is pre-set, and the three primary color values ​​of each pixel are added to the grayscale conversion formula for calculation, and the calculation result obtained is the grayscale value of each pixel. After the three primary color values ​​of all pixels are calculated, the grayscale values ​​of all pixels can be obtained. Then, the grayscale values ​​are compared, the maximum and minimum values ​​are determined, and the maximum and minimum values ​​are averaged, and the calculation result obtained forms the initial threshold.

[0080] Furthermore, two arrays are established, namely the first array and the second array. The grayscale value of each pixel is compared with the initial threshold value to determine the size relationship between each grayscale value and the initial threshold value. For the grayscale value determined to be greater than the initial threshold value after comparison, it is placed in the first array, and for the grayscale value determined to be less than the initial threshold value after comparison, it is placed in the second array. After all grayscale values ​​are compared, the first array forms a first grayscale value set, and the second array forms a second grayscale value set. Then, the mean operation is performed on each grayscale value in the first grayscale value set, and the operation result is obtained as the first average grayscale value, and the mean operation is performed on each grayscale value in the second grayscale value set, and the operation result is obtained as the second average grayscale value.

[0081] Furthermore, the first average grayscale value and the second average grayscale value are averaged, and the calculation result is obtained to replace and update the original initial threshold, and it is determined whether the updated initial threshold is the same as the initial threshold before the update. If they are the same, it means that the pixels corresponding to the first grayscale value set are all pixels of cervical surface imaging, and the pixels corresponding to the second grayscale value set are all pixels of cervical surrounding tissue imaging, so the pixels corresponding to the second grayscale value set can be removed from the cervical fluorescence image, and the new cervical fluorescence image formed by the remaining pixels after the removal is the image of the cervical surface imaging.

[0082] On the contrary, if it is determined that the updated initial threshold is different from the initial threshold before the update, it means that the pixels corresponding to the first gray value set and the pixels corresponding to the second gray value set both contain pixels from the cervical surface imaging and pixels from the cervical surrounding tissue imaging. At this time, it is necessary to compare the gray value of each pixel with the updated initial threshold, determine the size relationship between each gray value and the updated initial threshold, re-divide the first gray value set and the second gray value set, and calculate the first and second average gray values ​​to form a new initial threshold for judgment, until the updated initial threshold is the same as the initial threshold before the update, and generate a new cervical fluorescence image.

[0083] Step S20: Preliminary analysis is performed on the cervical fluorescence image according to each of the color proportion matrices to determine whether there is an abnormal area in the cervical fluorescence image.

[0084] Understandably, the material components and component contents of normal cervical tissue and abnormal cervical tissue are different, which leads to different cervical fluorescence images. A large number of cervical fluorescence images obtained from normal cervical tissue are analyzed in advance, and the proportion characteristics of each color in the three primary color values ​​are determined as a normal reference to analyze the generated color proportion matrix. Then, based on the analysis of each color proportion matrix, a preliminary analysis is performed on the reference cervical fluorescence image to determine whether there is an abnormal area in the cervical fluorescence image. Among them, the abnormal area includes at least a cancerous area and a precancerous lesion area.

[0085] It should be noted that in order to avoid missing cervical fluorescence images with abnormal areas, a normal reference is formed with strict judgment standards, and strict judgment standards are also used for the analysis of each color ratio matrix based on the normal reference. Only tissues that are strictly normal cervical tissue can be used to form a normal reference, and only when each color ratio matrix strictly matches the normal reference can it be determined that there is no abnormal area in the cervical fluorescence image. Otherwise, it is determined that there is an abnormal area, so as to ensure the accuracy of cervical abnormality screening.

[0086] Step S30: If there is an abnormal area in the cervical fluorescence image, an abnormal area image is extracted from the cervical fluorescence image, and the abnormal area image is converted into a cervical grayscale image based on the three primary color values ​​of each pixel in the abnormal area image.

[0087] Furthermore, if it is determined that there is an abnormal area in the cervical fluorescence image, the cervical fluorescence image can be segmented according to the boundary of the abnormal area, and the abnormal area can be extracted as an abnormal area image. Then, according to the grayscale value converted from the three primary color values ​​of each pixel in the abnormal area image, the abnormal area image is converted into a cervical grayscale image for subsequent analysis and processing.

[0088] Step S40, identifying the cervical grayscale image based on a preset recognition model to obtain a final recognition result.

[0089] Furthermore, a preset recognition model is pre-set in the server, and the preset recognition model is formed based on convolutional neural network training, including a convolution layer, a pooling layer, and a fully connected layer. Because normal cervical tissue has been excluded by the color ratio matrix, the cervical fluorescence image with abnormal areas after preliminary analysis needs to be processed. Therefore, in order to save the training time of the preset recognition model and the subsequent processing time, various abnormal images formed by various abnormal cervical tissues can be collected as training samples for training to form a preset recognition model, so that the preset recognition model has the ability to identify the abnormal type of cervical tissue. Of course, in order to avoid errors in cervical fluorescence images without abnormal areas excluded by the color ratio matrix, and to determine normal cervical fluorescence images as cervical fluorescence images with abnormal areas, the training samples can be set to include normal image samples formed by a large number of normal cervical tissues and various abnormal image samples formed by various abnormal cervical tissues. The preset recognition model is formed by training the training samples, so that the preset recognition model can distinguish between normal fluorescence images and abnormal fluorescence images while distinguishing various abnormal types. The converted cervical grayscale image is recognized and processed by the trained preset recognition model to obtain the final recognition result. The final recognition result at least includes information such as whether there is an abnormal area in the cervical fluorescence image, the size of the abnormal area if there is an abnormal area, and the corresponding abnormal type.

[0090] The cervical fluorescence image processing and analysis method implemented in this embodiment, after obtaining the cervical fluorescence image through the preset excitation light, extracts the three primary color values ​​of each pixel in the cervical fluorescence image, and generates multiple color proportion matrices of the cervical fluorescence image according to the three primary color values ​​of each pixel; then, based on each color proportion matrix, the cervical fluorescence image is preliminarily analyzed to determine whether there is an abnormal area in the cervical fluorescence image; if there is an abnormal area, the abnormal area image is segmented from the cervical fluorescence image, and the abnormal area image is converted into a cervical grayscale image based on the three primary color values ​​of each pixel in the abnormal area image; then, the converted cervical grayscale image is recognized by the preset recognition model to obtain the final recognition result. In this way, by setting up a front and back two-stage detection and recognition mechanism, the front stage is to generate the acquired cervical fluorescence image into a color proportion matrix for preliminary detection based on the difference in the three primary color values ​​of the pixels between the normal cervical fluorescence image and the abnormal cervical fluorescence image, and the back stage is to re-detect the cervical fluorescence image detected as abnormal by the front stage according to the preset recognition model formed by pre-training a large number of abnormal image samples. In addition, the normal judgment standard in the pre-stage detection is relatively strict. Only when it fully meets the normal standard can the cervical fluorescence image be judged to be normal. In addition, as long as there is a suspected abnormal area, it is judged to be abnormal, and then the post-stage inspection is carried out. On the one hand, normal cervical fluorescence images can be quickly screened out, and on the other hand, the preset recognition model formed by the convolutional neural network training for the preliminary abnormal cervical fluorescence images can perform more detailed and comprehensive image recognition, with higher recognition accuracy, and realize accurate and efficient analysis and processing of cervical fluorescence images.

[0091] For further information, please refer to Figure 2 Based on the first embodiment of the cervical fluorescence image processing and analysis method of the present invention, a second embodiment of the cervical fluorescence image processing and analysis method of the present invention is proposed.

[0092] The difference between the second embodiment of the cervical fluorescence image processing and analysis method and the first embodiment of the cervical fluorescence image processing and analysis method is that the step of identifying the cervical grayscale image based on a preset recognition model to obtain a final recognition result includes:

[0093] Step S41, extracting features from the cervical grayscale image based on a feature extraction module in a preset recognition model to obtain a feature vector of the cervical grayscale image;

[0094] Step S42, performing classification calculation on the feature vector based on the classifier in the preset recognition model to obtain a calculation result;

[0095] Step S43, determining the abnormal type corresponding to the abnormal area in the cervical fluorescence image according to the calculation result, identifying the area size of the abnormal area, and generating the abnormal type and the area size as the final recognition result.

[0096] Furthermore, the cervical grayscale image can be directly processed by a preset recognition model, or divided into multiple sub-grayscale images and processed by preset recognition models respectively, or divided into multiple sub-grayscale images for processing when the size of the cervical sub-grayscale image is larger than a certain size, and directly processed when it is not larger than the size. Regardless of whether it is divided or not, the processing flow is the same, but there is a difference in processing efficiency. This embodiment is described with an undivided cervical grayscale image.

[0097] Specifically, the preset recognition model is divided into a feature extraction module, which includes multiple convolution layers and multiple pooling layers, for example, at least three convolution layers and three pooling layers. The convolution layer is used to extract features in the image, and the pooling layer is used to select the features extracted by the convolution layer, so it can be set to connect a pooling layer after a convolution layer. In addition, it can also be set that the number of convolution layers is more than the number of pooling layers, for example, twice, so that two convolution layers correspond to one pooling layer. The cervical grayscale image is feature extracted by this type of feature extraction module in the preset recognition model to obtain multiple feature matrices, and then the feature matrices are connected by the fully connected layer in the preset recognition module to form a feature vector of the cervical grayscale image.

[0098] Furthermore, the preset recognition model also includes a classifier, through which each feature vector is classified and calculated to obtain a calculation result representing the probability value of the feature vector in each category of abnormality. Among them, the category of abnormality at least includes no abnormality, CIN1 abnormality, negative CIN2 abnormality, condyloma abnormality, positive CIN2 abnormality, positive CIN3 abnormality, cancer abnormality, etc. The calculation formula can be specifically referred to in the following formula (1).

[0099]

[0100] Where n represents the number of abnormal categories, ai and bi are the weight value and bias value of the i-th abnormality in the classifier, H is the feature vector, pi represents the calculation result of the i-th abnormality probability value, j represents the previous j calculations, and wi represents the sum of the classifier errors of the i-th abnormality calculation result in the previous j calculations.

[0101] In order to ensure the accuracy of probability value calculation in each category of abnormality, this embodiment is provided with a mechanism for correcting the classifier errors that occurred in the previous calculation processes for each category of abnormality. The specific number of times can be determined by testing. For example, if the correction effect of the first 5 times is good after testing, it is set to the first 5 times. Among them, the classifier error is the calculation error introduced due to its weight value and bias value. After each preset recognition model is recognized, the calculation error will be evaluated to obtain the corresponding classifier error. By averaging the classifier errors of the previous times and using them to correct the current classification calculation, the current calculation result is more accurate.

[0102] Furthermore, after obtaining the probability values ​​of each category of abnormalities represented by the calculation results, the probability values ​​represented by the calculation results are compared to determine the probability value with the largest value. Then, the abnormal type corresponding to the probability value with the largest value is found, and the abnormal type is the abnormal type corresponding to the abnormal area image. In addition, the abnormal area boundary in the abnormal area image is identified, and the abnormal range of the abnormal area is determined by the area boundary, and the area size corresponding to the abnormal area is obtained. In addition, for the abnormal area image formed by the boundary segmentation of the abnormal area, the area size corresponding to the abnormal area can be directly determined according to the boundary of the abnormal area image. Then, the abnormal type and the abnormal area size are jointly formed as the final recognition result.

[0103] It is understandable that the preset recognition model needs to be trained in advance. Specifically, the step of recognizing the cervical grayscale image based on the preset recognition model to obtain the final recognition result includes:

[0104] Step c1, obtaining cervical image samples, and dividing the cervical image samples into training image samples and test image samples;

[0105] Step c2, transmitting the training image sample to a preset initial model, training the preset initial model, and calculating the loss function value of the preset initial model when the number of training times reaches a preset number;

[0106] Step c3, determining whether the loss function value is less than or equal to a preset threshold value, and if so, testing the preset initial model based on the test sample image to obtain a test result;

[0107] Step c4, generating a test error according to the test result, and generating the preset recognition model according to the size relationship between the test error and a preset threshold.

[0108] Furthermore, a large number of cervical image samples are collected for screening, and those that are unclear are eliminated, and the remaining ones are marked with abnormal types, abnormal areas, etc. before being transmitted to the server. After the server obtains the cervical image samples, it is divided into training image samples and test image samples according to a preset ratio, such as 8 to 2, or 7 to 3. The training image samples are then transmitted to a preset initial model built based on a convolutional neural network, and deep learning training is performed on the preset initial model.

[0109] Furthermore, a preset number of times and a loss function value calculation formula are preset for the preset initial model. The loss function value is used to evaluate the difference between the model output of the preset initial model after training and the identified target output. The smaller the difference, the closer the model output of the preset initial model is to the target output, and the better the model training effect. When the number of training times reaches the preset number, the loss function value is calculated. The specific calculation formula can refer to the following formula (2).

[0110]

[0111] Where L represents the loss function value, M represents the number of samples in the training image samples that are identified as abnormal after training, N represents the number of samples in the training image samples that are identified as not abnormal after training, and P k1 represents the abnormal category identified by the training of the k1th training image sample, S k1 represents the abnormal category identified by the training of the k1th training image sample, S k1 represents the reference abnormal region corresponding to the k1th training image sample, S k1 ' represents the abnormal area identified by training for the k1th training image sample, P k2 represents the reference category corresponding to the k2th training image sample, P k2 ' represents the category of the k2th training image sample after training and identification, S k2 P represents the k2th training image sample k2 'With P k2 On the contrary, the reference abnormal area corresponding to the k2th training image sample.

[0112] It should be noted that the cervical image samples in this embodiment include both normal image samples formed by normal cervical tissue and various abnormal image samples formed by various abnormal cervical tissues. When calculating the loss function value, the error loss of normal image samples and abnormal image samples shall be considered. In the above formula (2), the reference abnormal category and the reference abnormal area are the marked abnormal category and abnormal area, respectively, and the reference category is the category marked as non-abnormal. For a training image sample, if it is an abnormal category, but is identified as a non-abnormal category, that is, the P of the above training image sample k2 'With P k2 On the contrary, the corresponding abnormal area is not identified at this time, and the abnormal area identified by the training image sample of the abnormal type constitutes its error. By considering the errors caused by various factors, the calculated loss function value is more accurate.

[0113] Furthermore, a preset threshold representing the error size is pre-set, and the calculated loss function value is compared with the preset threshold to determine whether the loss function value is less than or equal to the preset threshold. If it is less than or equal to, it means that the loss function value is smaller, and the closer the model output of the trained preset initial model is to the target output. At this time, in order to ensure the accuracy of the preset initial model, the preset initial model is tested by a test sample image, and the test sample image is processed by the preset initial model to obtain a test result, and a test error is generated based on the test result. The test error can be generated in the same way as the loss function value, or it can be different, and there is no limitation here.

[0114] Furthermore, the test error is compared with a preset threshold value to generate a size relationship between the two, and then a preset recognition model is generated based on the size relationship between the two. Specifically, the step of generating the preset recognition model based on the size relationship between the test error and the preset threshold value includes:

[0115] Step c41, if the size relationship is that the test error is less than a preset threshold, generating the preset initial model as a preset recognition model;

[0116] Step c42: If the test error is greater than or equal to a preset threshold, the calculation formula for the loss function value is updated, and the preset initial model is iteratively trained based on the updated calculation formula until the test error is less than the preset threshold.

[0117] Furthermore, if the relationship between the test error and the preset threshold is that the prediction error is less than the preset threshold, it means that the trained preset initial model can accurately process the test image samples, and the preset initial model has a better image processing effect after training, thereby generating the preset initial model as the preset recognition model.

[0118] If the relationship between the test error and the preset threshold is that the prediction error is greater than or equal to the preset threshold, it means that the trained preset initial model has a poor processing effect on the test image sample, but a good processing effect on the training image sample, and the preset initial model fits the training image sample too well. At this time, the calculation formula of the loss function value is adjusted, and a penalty term for limiting overfitting is added to the calculation formula to update the calculation formula. Then, the preset initial model is iteratively trained according to the updated calculation formula, and the loss function value is recalculated and tested again with the test sample when it is less than or equal to the preset threshold, until the test error generated by the test is also less than the preset threshold, and the preset initial model is generated as the preset recognition model.

[0119] The formula for updating the calculation formula of the loss function value is specifically shown in the following formula (3).

[0120]

[0121] Where L′ represents the updated calculation formula, L represents the calculation formula before the update, μ1 and μ2 represent the preset update parameters, and W t represents the weight matrix between the input layer and the convolutional layer in the preset initial model, m represents the number of elements in the weight matrix, It is used to calculate the cumulative value of each element in the weight matrix after taking the absolute value. Represents the transpose of the weight matrix.

[0122] This embodiment uses a preset recognition model for pre-training to recognize cervical grayscale images, fully considers various factors that cause model recognition errors during the training process, and prevents the preset recognition model from overfitting on training image samples by updating the loss function value calculation formula, thereby ensuring the accuracy of the preset recognition model, thereby making the preset recognition model more accurate in recognizing cervical grayscale images. At the same time, for the classifier in the preset recognition model, an evaluation and correction mechanism for the classifier error is set, which further improves the recognition accuracy of abnormal categories of cervical grayscale images.

[0123] For further information, please refer to Figure 3 Based on the first and second embodiments of the cervical fluorescence image processing and analysis method of the present invention, a third embodiment of the cervical fluorescence image processing and analysis method of the present invention is proposed.

[0124] The difference between the third embodiment of the cervical fluorescence image processing and analysis method and the first and second embodiments of the cervical fluorescence image processing and analysis method is that the step of performing a preliminary analysis on the cervical fluorescence image according to each color proportion matrix to determine whether there is an abnormal area in the cervical fluorescence image includes:

[0125] Step S21, based on a preset comparison formula, comparing the pixel color proportion value in each color proportion matrix with a preset proportion threshold value to generate a plurality of proportion conversion matrices;

[0126] Step S22, performing a sum operation on each of the proportion conversion matrices to obtain a sum matrix, and performing a sum operation on each element value in the sum matrix to obtain a matrix value of the sum matrix;

[0127] Step S23, determining whether the matrix value is greater than a preset matrix threshold value, and if so, determining that an abnormal area exists in the cervical fluorescence image.

[0128] Furthermore, the server is pre-set with a preset comparison formula and a preset ratio threshold for comparing the color ratio matrix. The preset ratio threshold is generated based on the pixel color ratio in the normal cervical fluorescence image formed by the normal cervical tissue, and is used to determine whether the pixel points in the cervical fluorescence image match the pixel points in the normal cervical fluorescence image. In addition, for three different color ratio matrices, the preset ratio threshold can be set to different values ​​according to the characteristics of the pixel points in the normal cervical fluorescence image, or it can be set to the same value. Specifically, the elements in the color ratio matrix are used as pixel color ratio values, and each color ratio value is compared with the preset ratio threshold through the preset comparison formula to determine whether the color ratio value is greater than or equal to the preset ratio threshold. If it is greater than or equal to the preset ratio threshold, the element corresponding to the color ratio value in the color ratio matrix is ​​generated as a value of 1, indicating that the pixel point corresponding to the color ratio value may be abnormal. On the contrary, if it is determined to be less than the preset ratio threshold, the element corresponding to the color ratio value in the color ratio matrix is ​​generated as a value of 0, indicating that the pixel point corresponding to the color ratio value is normal. The specific preset comparison formula can be seen in the following formula (4).

[0129]

[0130] k=1, 2, 3, x=1, 2, 3...X, y=1, 2, 3...Y; (4);

[0131] Where k represents the three primary colors red, green and blue, X represents the number of rows in the color proportion matrix, Y represents the number of columns in the color proportion matrix, F represents the preset proportion threshold, and f' k (x, y) represents the proportion conversion matrix, f k (x, y) represents the color proportion matrix.

[0132] Furthermore, after comparing all pixel color proportion values ​​in each color proportion matrix, the matrix formed after comparison is used as a proportion conversion matrix, and each proportion conversion matrix is ​​summed up. That is, each element in the red color proportion matrix, the green color proportion matrix, and the blue color proportion matrix is ​​summed up according to their respective positions to form a sum matrix, for example, the elements in the first row and first column of the three color proportion matrices are summed up, the elements in the first row and second column are summed up, the elements in the first row and third row are summed up..., and the elements in the first row, the first column, the second column, the third column... in the sum matrix are obtained respectively. Then, the values ​​of each element in the sum matrix are summed up again, for example, the values ​​of the elements in the first column, the second column, the third column... in the first row are summed up to obtain the matrix value of the sum matrix.

[0133] Furthermore, a preset matrix threshold formed according to a normal cervical fluorescence image is pre-set, and the normal cervical fluorescence image is formed by a normal cervical component. The preset matrix threshold is used to determine whether the currently collected cervical fluorescence image is abnormal. Specifically, the matrix value obtained by the operation is compared with the preset matrix threshold to determine whether the matrix value is greater than the preset matrix threshold. If it is greater than the preset matrix threshold, it means that there are more pixel color proportion values ​​greater than the preset proportion threshold in the pixel color proportion values ​​of the color proportion matrix, and the possibility that the cervix corresponding to the cervical fluorescence image from which the color proportion matrix is ​​derived is greater than that of abnormality, thereby determining that there is an abnormal area in the cervical fluorescence image. Otherwise, it is determined that there is no abnormal area in the cervical fluorescence image.

[0134] Furthermore, for a cervical fluorescence image with an abnormal area, the cervical fluorescence image can be segmented by identifying the boundary of the abnormal area, and the abnormal area can be extracted therefrom for abnormal type identification, so as to avoid identifying the entire cervical fluorescence image and improve identification efficiency. Specifically, if there is an abnormal area in the cervical fluorescence image, the step of extracting the abnormal area image from the cervical fluorescence image includes:

[0135] Step S31, if there is an abnormal area in the cervical fluorescence image, searching for a target element whose values ​​of all elements in the sum matrix are zero;

[0136] Step S32, searching for target pixel points corresponding to each of the target elements in the cervical fluorescence image, and removing the target pixel points from the fluorescence image to form the abnormal area image.

[0137] Furthermore, for cervical fluorescence images determined to have abnormal areas, the pixel points corresponding to the elements in the color proportion matrix whose color proportion values ​​are greater than the preset proportion threshold may be abnormal, and the value of the element after summing up in the sum matrix is ​​1, 2 or 3. Therefore, the abnormal area in the cervical fluorescence image can be determined by finding all elements in the sum matrix whose element values ​​are zero.

[0138] Specifically, the element values ​​in the sum matrix are identified, and all elements whose element values ​​are zero are determined as target elements. Then, the target position of each target element in the sum matrix is ​​determined, and the pixel points corresponding to each target position in the cervical fluorescence image are identified as the target pixel points corresponding to each target element in the cervical fluorescence image. The target pixel points are points in the non-abnormal area of ​​the cervical fluorescence image. All identified target pixel points are removed from the cervical fluorescence image, and the area formed by the remaining pixel points is the abnormal area in the cervical fluorescence image.

[0139] In this embodiment, each color proportion matrix is ​​converted into a proportion conversion matrix. For the same pixel point, as long as the pixel color proportion value of the pixel point in any color proportion matrix is ​​abnormal, the pixel point is generated as an abnormal element in the proportion conversion matrix, and the cervical fluorescence image is segmented according to the abnormal element and other normal matrices in the proportion conversion matrix, and the abnormal area image is extracted therefrom, so that the determination of whether there is an abnormal area in the cervical fluorescence image and the extraction of the abnormal area image are more accurate.

[0140] In addition, the embodiment of the present invention also provides a cervical fluorescence image processing and analysis system. Figure 4 , Figure 4 It is a structural schematic diagram of the equipment hardware operating environment involved in the embodiment of the cervical fluorescence image processing and analysis system of the present invention.

[0141] like Figure 4 As shown, the cervical fluorescence image processing and analysis system may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a storage 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The storage 1005 may be a high-speed RAM storage, or a stable storage (non-volatile memory), such as a disk storage. The storage 1005 may also be a storage device independent of the aforementioned processor 1001.

[0142] Those skilled in the art will understand that Figure 4 The hardware structure of the cervical fluorescence image processing and analysis system shown in the figure does not constitute a limitation on the cervical fluorescence image processing and analysis system, and may include more or less components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0143] like Figure 4 As shown, the storage 1005 as a storage medium may include an operating system, a network communication module, a user interface module and a control program. The operating system is a program for managing and controlling the cervical fluorescence image processing and analysis system and software resources, and supports the operation of the network communication module, the user interface module, the control program and other programs or software; the network communication module is used to manage and control the network interface 1004; the user interface module is used to manage and control the user interface 1003.

[0144] exist Figure 4 In the hardware structure of the cervical fluorescence image processing and analysis system shown in the figure, the network interface 1004 is mainly used to connect to other system servers and perform data communication with other system servers; the user interface 1003 is mainly used to connect to the client (user end) and perform data communication with the client; the processor 1001 can call the control program stored in the storage 1005 and perform the following operations:

[0145] After acquiring a cervical fluorescence image based on a preset excitation light, extracting three primary color values ​​of each pixel in the cervical fluorescence image, and generating multiple color proportion matrices of the cervical fluorescence image based on the three primary color values ​​of each pixel;

[0146] Performing a preliminary analysis on the cervical fluorescence image according to each of the color proportion matrices to determine whether there is an abnormal area in the cervical fluorescence image;

[0147] If there is an abnormal area in the cervical fluorescence image, extracting the abnormal area image from the cervical fluorescence image, and converting the abnormal area image into a cervical grayscale image based on the three primary color values ​​of each pixel in the abnormal area image;

[0148] The cervical grayscale image is recognized based on a preset recognition model to obtain a final recognition result.

[0149] Furthermore, the step of identifying the cervical grayscale image based on a preset recognition model to obtain a final recognition result includes:

[0150] Performing feature extraction on the cervical grayscale image based on a feature extraction module in a preset recognition model to obtain a feature vector of the cervical grayscale image;

[0151] Based on the classifier in the preset recognition model, the feature vector is classified and calculated to obtain a calculation result, and the calculation formula is:

[0152]

[0153] Where n represents the number of abnormal categories, ai and bi are the weight and bias values ​​of the i-th abnormality in the classifier, H is the feature vector, pi represents the calculation result of the i-th abnormality probability value, j represents the previous j calculations, and wi represents the sum of the classifier errors of the i-th abnormality calculation results in the previous j calculations;

[0154] According to the calculation result, the abnormal type corresponding to the abnormal area in the cervical fluorescence image is determined, and the area size of the abnormal area is identified, and the abnormal type and the area size are generated as the final identification result.

[0155] Furthermore, before the step of identifying the cervical grayscale image based on the preset recognition model to obtain the final recognition result, the processor 1001 may call the control program stored in the storage 1005 and perform the following operations:

[0156] Acquire cervical image samples, and divide the cervical image samples into training image samples and test image samples;

[0157] The training image samples are transferred to the preset initial model, the preset initial model is trained, and when the number of training times reaches the preset number of times, the loss function value of the preset initial model is calculated, and the calculation formula is:

[0158]

[0159] Where L represents the loss function value, M represents the number of samples in the training image samples that are identified as abnormal after training, N represents the number of samples in the training image samples that are identified as not abnormal after training, and P k1 represents the abnormal category identified by the training of the k1th training image sample, S k1 represents the abnormal category identified by the training of the k1th training image sample, S k1 represents the reference abnormal region corresponding to the k1th training image sample, S k1 ' represents the abnormal area identified by training for the k1th training image sample, P k2 represents the reference category corresponding to the k2th training image sample, P k2 ' represents the category of the k2th training image sample after training and identification, S k2 P represents the k2th training image sample k2 'With P k2 On the contrary, the reference abnormal area corresponding to the k2th training image sample;

[0160] Determine whether the loss function value is less than or equal to a preset threshold value, and if so, test the preset initial model based on the test sample image to obtain a test result;

[0161] A test error is generated according to the test result, and the preset recognition model is generated according to the size relationship between the test error and a preset threshold.

[0162] Furthermore, the step of generating the preset recognition model according to the magnitude relationship between the test error and the preset threshold value includes:

[0163] If the size relationship is that the test error is less than a preset threshold, generating the preset initial model as a preset recognition model;

[0164] If the test error is greater than or equal to the preset threshold, the calculation formula of the loss function value is updated, and the preset initial model is iteratively trained based on the updated calculation formula until the test error is less than the preset threshold, wherein the formula for updating the calculation formula is:

[0165]

[0166] Where L′ represents the updated calculation formula, L represents the calculation formula before the update, μ1 and μ2 represent the preset update parameters, and W t represents the weight matrix between the input layer and the convolutional layer in the preset initial model, m represents the number of elements in the weight matrix, It is used to calculate the cumulative value of each element in the weight matrix after taking the absolute value. Represents the transpose of the weight matrix.

[0167] Furthermore, the step of performing a preliminary analysis on the cervical fluorescence image according to each of the color proportion matrices to determine whether there is an abnormal area in the cervical fluorescence image includes:

[0168] Based on a preset comparison formula, the pixel color proportion values ​​in each color proportion matrix are compared with a preset proportion threshold value to generate multiple proportion conversion matrices, wherein the preset comparison formula is:

[0169]

[0170] k=1, 2, 3, x=1, 2, 3...X, y=1, 2, 3...Y;

[0171] Where k represents the three primary colors red, green and blue, X represents the number of rows in the color proportion matrix, Y represents the number of columns in the color proportion matrix, F represents the preset proportion threshold, and f'k (x, y) represents the proportion conversion matrix, f k (x, y) represents the color proportion matrix;

[0172] Performing a sum operation on each of the proportion conversion matrices to obtain a sum matrix, and performing a sum operation on each element value in the sum matrix to obtain a matrix value of the sum matrix;

[0173] It is determined whether the matrix value is greater than a preset matrix threshold value. If the matrix value is greater than the preset matrix threshold value, it is determined that an abnormal area exists in the cervical fluorescence image.

[0174] Furthermore, if there is an abnormal area in the cervical fluorescence image, the step of extracting the abnormal area image from the cervical fluorescence image includes:

[0175] If there is an abnormal area in the cervical fluorescence image, searching for a target element whose values ​​of all elements in the sum matrix are zero;

[0176] Target pixel points corresponding to the target elements are found in the cervical fluorescence image, and the target pixel points are removed from the fluorescence image to form the abnormal area image.

[0177] Furthermore, before the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image, the processor 1001 may call the control program stored in the storage 1005 and perform the following operations:

[0178] Determining whether the cervical fluorescence image includes a complete cervical surface and whether the clarity of the cervical fluorescence image reaches a preset clarity threshold;

[0179] If the cervical fluorescence image includes the complete surface of the cervix and the clarity reaches a preset clarity threshold, the cervical fluorescence image is determined to be a valid image, and the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image is performed;

[0180] If the cervical fluorescence image does not include the complete cervical surface and / or the clarity does not reach a preset clarity threshold, the cervical fluorescence image is determined to be an invalid image, and a prompt message for re-collecting the cervical fluorescence image is output.

[0181] Furthermore, after the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image, the processor 1001 may call the control program stored in the storage 1005 and perform the following operations:

[0182] Generate a grayscale value of each pixel point according to the three primary color values ​​of each pixel point, determine the maximum value and the minimum value of each grayscale value, and perform a mean operation on the maximum value and the minimum value to obtain a calculation result as an initial threshold value;

[0183] Comparing the grayscale value of each pixel with the initial threshold, determining a first grayscale value set greater than the initial threshold among the grayscale values, and a second grayscale value set less than the initial threshold among the grayscale values;

[0184] Based on the first grayscale value set and the second grayscale value set, respectively generate a first average grayscale value and a second average grayscale value;

[0185] updating the initial threshold according to the first average grayscale value and the second average grayscale value, and determining whether the updated initial threshold is the same as the initial threshold before the update;

[0186] If they are the same, each pixel point corresponding to the second gray value set is removed from the cervical fluorescence image to form a new cervical fluorescence image;

[0187] If they are not the same, then based on the updated initial threshold, a step of comparing the grayscale value of each pixel with the initial threshold is performed until the updated initial threshold is the same as the initial threshold before the update.

[0188] The specific implementation of the cervical fluorescence image processing and analysis system of the present invention is basically the same as the above-mentioned embodiments of the cervical fluorescence image processing and analysis method, and will not be repeated here.

[0189] The embodiment of the present invention further provides a storage medium on which a control program is stored, and when the control program is executed by a processor, the steps of the cervical fluorescence image processing and analysis method described above are implemented.

[0190] The storage medium of the present invention may be a computer-readable storage medium, and its implementation is substantially the same as the above-mentioned embodiments of the cervical fluorescence image processing and analysis method, and will not be described in detail here.

[0191] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims. All equivalent structures or equivalent process changes made using the contents of the specification and drawings of the present invention, or directly or indirectly used in other related technical fields, are protected by the present invention.

Claims

1. A cervical fluorescence image processing and analysis method, characterized in that: The cervical fluorescence image processing and analysis method comprises: After acquiring a cervical fluorescence image based on a preset excitation light, extracting three primary color values ​​of each pixel in the cervical fluorescence image, and generating multiple color proportion matrices of the cervical fluorescence image based on the three primary color values ​​of each pixel; Performing a preliminary analysis on the cervical fluorescence image according to each of the color proportion matrices to determine whether there is an abnormal area in the cervical fluorescence image; If there is an abnormal area in the cervical fluorescence image, extracting the abnormal area image from the cervical fluorescence image, and converting the abnormal area image into a cervical grayscale image based on the three primary color values ​​of each pixel in the abnormal area image; The cervical grayscale image is recognized based on a preset recognition model to obtain a final recognition result.

2. The cervical fluorescence image processing and analysis method according to claim 1, characterized in that: The step of identifying the cervical grayscale image based on a preset recognition model to obtain a final recognition result comprises: Performing feature extraction on the cervical grayscale image based on a feature extraction module in a preset recognition model to obtain a feature vector of the cervical grayscale image; Based on the classifier in the preset recognition model, the feature vector is classified and calculated to obtain a calculation result, and the calculation formula is: Where n represents the number of abnormal categories, ai and bi are the weight and bias values ​​of the i-th abnormality in the classifier, H is the feature vector, pi represents the calculation result of the i-th abnormality probability value, j represents the previous j calculations, and wi represents the sum of the classifier errors of the i-th abnormality calculation results in the previous j calculations; According to the calculation result, the abnormal type corresponding to the abnormal area in the cervical fluorescence image is determined, and the area size of the abnormal area is identified, and the abnormal type and the area size are generated as the final identification result.

3. The cervical fluorescence image processing and analysis method according to claim 1, characterized in that: The step of identifying the cervical grayscale image based on a preset recognition model to obtain a final recognition result includes: Acquire cervical image samples, and divide the cervical image samples into training image samples and test image samples; The training image samples are transferred to the preset initial model, the preset initial model is trained, and when the number of training times reaches the preset number of times, the loss function value of the preset initial model is calculated, and the calculation formula is: Where L represents the loss function value, M represents the number of samples in the training image samples that are identified as abnormal after training, N represents the number of samples in the training image samples that are identified as not abnormal after training, and P k1 represents the reference anomaly category corresponding to the k1th training image sample, P k1 ' represents the abnormal category of the k1th training image sample after training and identification, S k1 represents the reference abnormal region corresponding to the k1th training image sample, S k1 ' represents the abnormal area identified by training for the k1th training image sample, P k2 represents the reference category corresponding to the k2th training image sample, P k2 ' represents the category of the k2th training image sample after training and identification, S k2 P represents the k2th training image sample k2 'With P k2 On the contrary, the reference abnormal area corresponding to the k2th training image sample; Determine whether the loss function value is less than or equal to a preset threshold value, and if so, test the preset initial model based on the test sample image to obtain a test result; A test error is generated according to the test result, and the preset recognition model is generated according to the size relationship between the test error and a preset threshold.

4. The cervical fluorescence image processing and analysis method according to claim 3, characterized in that: The step of generating the preset recognition model according to the magnitude relationship between the test error and the preset threshold comprises: If the size relationship is that the test error is less than a preset threshold, generating the preset initial model as a preset recognition model; If the test error is greater than or equal to the preset threshold, the calculation formula of the loss function value is updated, and the preset initial model is iteratively trained based on the updated calculation formula until the test error is less than the preset threshold, wherein the formula for updating the calculation formula is: Where L′ represents the updated calculation formula, L represents the calculation formula before the update, μ1 and μ2 represent the preset update parameters, and W t represents the weight matrix between the input layer and the convolutional layer in the preset initial model, m represents the number of elements in the weight matrix, It is used to calculate the cumulative value of each element in the weight matrix after taking the absolute value. Represents the transpose of the weight matrix.

5. The cervical fluorescence image processing and analysis method according to claim 1, characterized in that: The step of performing a preliminary analysis on the cervical fluorescence image according to each of the color proportion matrices to determine whether there is an abnormal area in the cervical fluorescence image comprises: Based on a preset comparison formula, the pixel color proportion values ​​in each color proportion matrix are compared with a preset proportion threshold value to generate multiple proportion conversion matrices, wherein the preset comparison formula is: k=1, 2, 3, x=1, 2, 3...X, y=1, 2, 3...Y; Where k represents the three primary colors red, green and blue, X represents the number of rows in the color proportion matrix, Y represents the number of columns in the color proportion matrix, F represents the preset proportion threshold, and f′ k (x, y) represents the proportion conversion matrix, f k (x, y) represents the color proportion matrix; Performing a sum operation on each of the proportion conversion matrices to obtain a sum matrix, and performing a sum operation on each element value in the sum matrix to obtain a matrix value of the sum matrix; It is determined whether the matrix value is greater than a preset matrix threshold value. If the matrix value is greater than the preset matrix threshold value, it is determined that an abnormal area exists in the cervical fluorescence image.

6. The cervical fluorescence image processing and analysis method according to claim 5, characterized in that: If there is an abnormal area in the cervical fluorescence image, the step of extracting the abnormal area image from the cervical fluorescence image includes: If there is an abnormal area in the cervical fluorescence image, searching for a target element whose values ​​of all elements in the sum matrix are zero; Target pixel points corresponding to the target elements are found in the cervical fluorescence image, and the target pixel points are removed from the fluorescence image to form the abnormal area image.

7. The cervical fluorescence image processing and analysis method according to any one of claims 1 to 6, characterized in that: The step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image includes: Determining whether the cervical fluorescence image includes a complete cervical surface and whether the clarity of the cervical fluorescence image reaches a preset clarity threshold; If the cervical fluorescence image includes the complete surface of the cervix and the clarity reaches a preset clarity threshold, the cervical fluorescence image is determined to be a valid image, and the step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image is performed; If the cervical fluorescence image does not include the complete cervical surface and / or the clarity does not reach a preset clarity threshold, the cervical fluorescence image is determined to be an invalid image, and a prompt message for re-collecting the cervical fluorescence image is output.

8. The cervical fluorescence image processing and analysis method according to any one of claims 1 to 6, characterized in that: The step of extracting the three primary color values ​​of each pixel in the cervical fluorescence image includes: Generate a grayscale value of each pixel point according to the three primary color values ​​of each pixel point, determine the maximum value and the minimum value of each grayscale value, and perform a mean operation on the maximum value and the minimum value to obtain a calculation result as an initial threshold value; Comparing the grayscale value of each pixel with the initial threshold, determining a first grayscale value set greater than the initial threshold among the grayscale values, and a second grayscale value set less than the initial threshold among the grayscale values; Based on the first grayscale value set and the second grayscale value set, respectively generate a first average grayscale value and a second average grayscale value; updating the initial threshold according to the first average grayscale value and the second average grayscale value, and determining whether the updated initial threshold is the same as the initial threshold before the update; If they are the same, each pixel point corresponding to the second gray value set is removed from the cervical fluorescence image to form a new cervical fluorescence image; If they are not the same, then based on the updated initial threshold, a step of comparing the grayscale value of each pixel with the initial threshold is performed until the updated initial threshold is the same as the initial threshold before the update.

9. A cervical fluorescence image processing and analysis system, characterized in that: The cervical fluorescence image processing and analysis system includes a storage device, a processor, a communication bus, and a control program stored in the storage device: The communication bus is used to realize the connection and communication between the processor and the storage; The processor is used to execute the control program to implement the steps of the cervical fluorescence image processing and analysis method according to any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium stores a control program, and when the control program is executed by the processor, the steps of the cervical fluorescence image processing and analysis method according to any one of claims 1 to 8 are implemented.

Images