Oral cancer risk assessment and prevention system based on big data

Abstract

The invention relates to the technical field of risk assessment, and discloses a big data-based oral cancer risk assessment and prevention system, which comprises a data collection module, a data preprocessing module, a category analysis module, a risk assessment module, a prevention scheme acquisition module, a scheme preferential module and a health management module. Through arrangement of a category analysis module, a risk assessment module and a scheme selection module, image categories corresponding to image data can be acquired based on a capsule neural network model, risk level division is performed based on risk assessment scores, and an optimal prevention scheme is selected through an improved optimization algorithm; a user can answer questionnaire survey questions by himself / herself so as to carry out risk self-test on the oral cancer, obtain corresponding prevention schemes, comprehensively consider personal living habits to carry out risk factor analysis and comprehensively evaluate the risk level of the individual oral cancer, the method is more convenient and effective, the early screening ability can be enhanced, and the morbidity of the oral cancer can be reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk assessment, more particularly to a mouth cancer risk assessment and prevention system based on big data. BACKGROUND

[0002] Mouth cancer is one of the malignant tumors with high incidence and high mortality worldwide, including tongue cancer, gum cancer, palate cancer, oropharyngeal cancer, and lip cancer. The early symptoms are not obvious, and most cases are found to be in the advanced stage. The clinical treatment effect is poor. If early detection and intervention measures can be taken, mouth cancer can be prevented to some extent. The current common mouth cancer screening methods often rely on a single examination method, such as tissue biopsy and imaging examination, which has certain limitations, and it is difficult to early detect high-risk groups. Studies have shown that a variety of risk factors can increase the risk of mouth cancer, such as smoking, alcoholism, and poor oral hygiene. Early detection of risk factors and intervention measures are of great significance in preventing mouth cancer.

[0003] Traditional mouth cancer risk assessment and prevention has certain limitations, and cannot analyze risk factors based on comprehensive consideration of individual living habits, comprehensively assess the risk level of individual mouth cancer, and formulate corresponding personalized effective prevention programs according to different risk levels and types of mouth cancer. In view of this, the present application provides an efficient, convenient and reliable mouth cancer risk assessment and prevention system to improve early screening ability and reduce the incidence of mouth cancer. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a mouth cancer risk assessment and prevention system based on big data to solve the problems in the background art.

[0005] The present application provides the following technical scheme: a mouth cancer risk assessment and prevention system based on big data, comprising a data collection module, a data preprocessing module, a category analysis module, a risk assessment module, a prevention program acquisition module, a scheme optimization module, and a health management module. The data collection module is used for collecting user data and transmitting it to the data preprocessing module. The data collection module comprises a text data acquisition unit and an image data acquisition unit. The data preprocessing module is used for preprocessing the data of the data collection module. The data preprocessing module comprises a text data preprocessing unit and an image data preprocessing unit. The category analysis module is used for receiving preprocessed image data, obtaining the image category corresponding to the image data based on a capsule neural network model, and transmitting the risk assessment module and the prevention program acquisition module. The risk assessment module is configured to receive data from the text data preprocessing unit and the category analysis module, calculate a risk assessment score, and divide the risk level based on the risk assessment score. The prevention scheme acquisition module is configured to acquire a plurality of different prevention schemes, and form a prevention scheme set corresponding to the risk level and the image category. The scheme optimization module is configured to select an optimal prevention scheme by using the improved optimization algorithm, and transmit the scheme to the health management module. The health management module is configured to output the content of the optimal prevention scheme, perform human-computer interaction display, regularly send health reminders, provide health consultation and medical guidance services, and can be connected with a medical institution.

[0006] Preferably, the text data acquisition unit is configured to collect text data of a user and transmit the text data to the data preprocessing module, and the image data acquisition unit is configured to collect image data of the user and transmit the image data to the data preprocessing module; the text data preprocessing unit is configured to preprocess the text data of the text data acquisition unit and transmit the text data to the risk assessment module, and the image data preprocessing unit is configured to preprocess the image data of the image data acquisition unit and transmit the image data to the category analysis module. The user data includes user text data and user image data, the text data includes life data and oral health data, and the image data is oral image; the user is an object requiring oral cancer screening or risk assessment; the life data includes user dietary preferences, oral hygiene habits, and life habits; the oral health data includes user oral symptom description, oral disease history, treatment history, and genetic history; and the oral symptom description is a recent oral condition description.

[0007] Preferably, the capsule neural network includes an input layer, a feature extraction layer, an initial capsule layer, a routing capsule layer, a loss function, a decoder, and an output layer; the input layer inputs an image, the feature extraction layer is a feature extraction network, the initial capsule layer is composed of a series of capsules, each capsule is a separate convolutional layer, and each capsule extracts information on each part of a feature map; the routing capsule layer passes all vectors through a dynamic routing structure, repeats until a specified number of iterations, and outputs a category vector from the output layer; the number of category vectors is the number of image categories, and the elements in the category vector represent features on a picture, and the length represents the probability that the input picture belongs to the category. The calculation formula of the routing capsule layer is represented as: ; ; ; ; Wherein, j = 1, 2, 3, …, m; m is the initial capsule layer output vector number, a is the category vector, s is the weighted vector sum, b j is the weight of the jth vector, v j is the intermediate vector after transformation of the jth vector, w j is the weight matrix or transformation matrix of the jth vector, u j is the jth vector of the initial capsule output, squash represents the squash function, softmax represents the softmax function, b j ´ is the new weight after the softmax function, and is used as the new b j is substituted and the next round of calculation is performed; repeat formulas ① to ④ until the specified number of iterations is reached, at which time the final category vector a is output, and the number of a is the number of image categories; The loss function uses a marginal loss function, and the formula is: Wherein, n is the number of categories of the image, that is, the number of categories of oral cancer, which includes the normal category, that is, the normal image; a i is the output vector of category i; T i represents whether the image belongs to category i, and takes the value of 0 or 1, if it belongs, it takes the value of 1, and if it does not belong, it takes the value of 0; m + and m - is a label smoothing term, which takes the value of 0.9 and 0.1, and the loss is divided into two parts, the length of the output vector of the true label is close to m + , and the length of the remaining output vector is close to m - ; λ is a balance parameter, i = 1, 2, 3, …, n; The image data preprocessed by the image data preprocessing unit is input into the capsule neural network model, and the category vector with the maximum vector length is taken as the category of the input image, that is, the category of oral cancer corresponding to the image.

[0008] Preferably, the calculation formula of the risk assessment score calculated by the risk assessment module is: Wherein, FX is the risk assessment score, γ h is the risk assessment coefficient of the hth risk assessment factor, BD h is the expression amount of the hth risk assessment factor, and H is the total amount of risk assessment factors; the risk assessment factor is the category of text data, and the number of risk assessment factors is the same as the total amount of questionnaire survey questions; The calculation formula of the risk assessment coefficient is: Wherein, R hN is the total number of samples; the life data and oral health data of N patients with oral cancer are collected in advance as sample data; if the yth sample data has the same data as the adverse answer corresponding to the hth risk assessment factor, the hth risk assessment factor exists in the yth sample data.

[0009] Preferably, the risk assessment module divides the risk level based on the risk assessment score, and the division is specifically: When the risk assessment score FX satisfies FX=0, it is classified as no risk; When the risk assessment score FX satisfies 0 When the risk assessment score FX satisfies YU2 When the risk assessment score FX satisfies FX>YU2, it is classified as high risk; The YU1 and YU2 are both determination thresholds, the YU1 is a first determination threshold, and the YU2 is a second determination threshold, YU1

[0010] Preferably, the set of prevention schemes of the prevention scheme acquisition module is represented as: , wherein J i is the set of prevention schemes of the ith oral cancer category, DE lowi is the set of low-risk prevention measures in the set of prevention schemes of the ith oral cancer category, DE midi is the set of medium-risk prevention measures in the set of prevention schemes of the ith oral cancer category, DE highi is the set of high-risk prevention measures in the set of prevention schemes of the ith oral cancer category; The results of the category analysis module and the results of the risk assessment module are combined to obtain a set of prevention measures corresponding to the risk level of the corresponding category, and all prevention schemes in the set are marked as candidate prevention schemes; If the output result of the category analysis module is a normal category or the output result of the risk assessment module is no risk, a health reminder instruction is directly sent to the health management module, and at this time the health management module displays the data through human-computer interaction and sends a health reminder to the user.

[0011] Preferably, the scheme optimization module selects the optimal prevention scheme in the following specific manner: Step S11: encoding the candidate prevention scheme, the encoding is O, O is a chromosome, obtaining the chromosome, and constructing an initial population ; Step S12: determining the fitness function; Step S13: natural selection of the chromosomes in the population; Step S14: performing crossover recombination on the chromosomes in the population; Step S15: performing mutation on the chromosomes in the population; Step S16: obtaining a new population, presetting a population generation number L and an adaptability threshold Q, L being an integer greater than 0 and Q being a real number greater than 0; repeating steps S13 to S15 until the population generation number of the new population is L or there is a chromosome in the new population corresponding to an adaptability greater than or equal to the adaptability threshold Q, and the cycle ends, the optimal preventive scheme corresponding to the chromosome corresponding to the maximum adaptability in the new population being obtained.

[0012] Preferably, the expression of the adaptability function is: , wherein f r is the adaptability corresponding to the rth chromosome, YZ r is the risk inhibition value of the preventive scheme corresponding to the rth chromosome, and r = 1, 2, 3, …, k; The natural selection is performed by combining the elite method and the rotation method; The expression of the rotation probability is: , wherein ζ r is the rotation probability corresponding to the rth chromosome.

[0013] Preferably, the risk inhibition value is obtained in the following manner: The oral cancer category, the risk assessment score, and the preventive scheme are taken as a group of input data, which are input into the risk inhibition prediction model to obtain the risk inhibition value corresponding to the preventive scheme; The risk inhibition prediction model is a neural network model, and the training method is: The oral cancer category, the risk assessment score, and the preventive scheme of each sample data in N sample data collected in advance are taken as a group of analysis data, the risk inhibition value corresponding to each group of analysis data is obtained, the risk inhibition value is obtained by subtracting the risk assessment score before the preventive scheme is adopted from the risk assessment score after the preventive scheme is adopted, and the analysis data and the corresponding risk inhibition value are converted into a corresponding group of feature vectors; Each group of feature vectors is taken as the input of the risk inhibition prediction model, the risk inhibition prediction model takes one predicted risk inhibition value corresponding to each group of analysis data as the output, takes the actual risk inhibition value corresponding to each group of analysis data as the prediction target, the actual risk inhibition value can be obtained through the data collected in advance, and takes the minimization of the sum of prediction errors of all analysis data as the training target; the formula of the prediction error is: , wherein ε p is the prediction error, p is the group number of the feature vector corresponding to the analysis data, and θ pa predicted risk inhibition value corresponding to the pth group of analysis data, μ p an actual risk inhibition value corresponding to the pth group of analysis data, μ

[0014] Technical effects and advantages of the present application: The present application is provided with a category analysis module, a risk assessment module and a scheme optimization module, which is conducive to obtaining the image category corresponding to the image data based on the capsule neural network model, that is, the oral cancer category corresponding to the image, then calculating the risk assessment score based on the life data and oral health data, and dividing the risk level based on the risk assessment score, and finally obtaining the alternative prevention scheme based on the obtained oral cancer category and risk level, and selecting the optimal prevention scheme through the improved optimization algorithm; the user can answer the questionnaire and upload the oral image, so as to perform the risk self-test of oral cancer, and obtain the corresponding prevention scheme, analyze the risk factors by comprehensively considering the individual life habits, and comprehensively evaluate the risk level of oral cancer, which is more efficient, more convenient and more effective, can enhance the early screening ability, and reduce the incidence of oral cancer. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The present application is provided with a category analysis module, a risk assessment module and a scheme optimization module, which is conducive to obtaining the image category corresponding to the image data based on the capsule neural network model, that is, the oral cancer category corresponding to the image, then calculating the risk assessment score based on the life data and oral health data, and dividing the risk level based on the risk assessment score, and finally obtaining the alternative prevention scheme based on the obtained oral cancer category and risk level, and selecting the optimal prevention scheme through the improved optimization algorithm; the user can answer the questionnaire and upload the oral image, so as to perform the risk self-test of oral cancer, and obtain the corresponding prevention scheme, analyze the risk factors by comprehensively considering the individual life habits, and comprehensively evaluate the risk level of oral cancer, which is more efficient, more convenient and more effective, can enhance the early screening ability, and reduce the incidence of oral cancer. DETAILED DESCRIPTION

[0016] The technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application, and the forms of each structure described in the following embodiments are only examples, and the present application is not limited to each structure described in the following embodiments, and all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] As shown in Figure 1 The present application provides a big data-based oral cancer risk assessment and prevention system, which comprises a data collection module, a data preprocessing module, a category analysis module, a risk assessment module, a prevention scheme acquisition module, a scheme optimization module and a health management module. The data collection module is used for collecting user data and transmitting to the data preprocessing module, wherein the user data includes user text data and user image data, the text data includes life data and oral health data, and the image data is an oral image; the user is an object requiring oral cancer screening or risk assessment; the data collection module includes a text data acquisition unit and an image data acquisition unit, the text data acquisition unit is used for collecting user text data and transmitting to the data preprocessing module, and the image data acquisition unit is used for collecting user image data and transmitting to the data preprocessing module; the life data includes but is not limited to user dietary preferences, oral hygiene habits and living habits, etc.; the oral health data includes but is not limited to user oral symptom description, oral disease history, treatment history and genetic history, etc.; the dietary preferences include but are not limited to milk tea, carbonated beverage intake, etc., the oral hygiene habits include but are not limited to tooth brushing frequency, toothpaste usage type and dental floss usage, the living habits include but are not limited to smoking, alcoholism, etc., and the oral symptom description is a recent oral condition description, such as whether dry mouth, pain, etc.; the text data can be obtained by setting a questionnaire interface to conduct questionnaire survey on the user, and the image data is uploaded by the user himself or herself, including oral image or professional image related to oral condition; The data preprocessing module is used for preprocessing the data of the data collection module, and includes a text data preprocessing unit and an image data preprocessing unit; the text data preprocessing unit is used for preprocessing the text data of the text data acquisition unit and transmitting to the risk assessment module, and the image data preprocessing unit is used for preprocessing the image data of the image data acquisition unit and transmitting to the category analysis module; the text data preprocessing unit is used for preprocessing the text data, including but not limited to data cleaning, data conversion and data standardization; the image data preprocessing unit is used for preprocessing the image data, including but not limited to noise reduction and binarization processing of the image; the data cleaning includes but is not limited to removing duplicate data and incomplete data, etc., so as to ensure data quality, the data conversion includes but is not limited to converting data in different formats into a unified format, so as to facilitate subsequent processing, and the data standardization includes but is not limited to normalizing and standardizing the data, so as to eliminate the dimensional difference between different data; The category analysis module is configured to receive the preprocessed image data, obtain the image category corresponding to the image data based on the capsule neural network model, and then transmit the risk assessment module and the prevention scheme acquisition module, wherein the image category corresponding to the image data is the oral cancer category corresponding to the image; different oral cancer categories have different symptoms and characteristics in different periods, so by identifying and classifying the oral image, the oral cancer category image closest to the current oral image can be obtained, and targeted prevention can be performed; the oral cancer category includes but is not limited to tongue cancer, gum cancer, palate cancer, oropharyngeal cancer, and lip cancer, etc.; after obtaining the oral cancer category through the capsule neural network, corresponding prevention measures can be taken for different categories, and the pertinence and effectiveness of oral cancer prevention can be improved; The risk assessment module is configured to receive the data from the text data preprocessing unit and the category analysis module, calculate a risk assessment score, and divide the risk level based on the risk assessment score; the purpose is to quantify the risk through risk assessment, so as to obtain the corresponding risk level; different risk levels have different prevention schemes, for example, for high-risk levels, timely medical treatment and professional in-depth examination are required, and for low-risk levels, oral hygiene knowledge education is required, such as normal tooth brushing and use of dental floss, so that the treatment and deterioration prevention of oral cancer can be more targeted and the effect can be enhanced; The prevention scheme acquisition module is configured to acquire a plurality of different prevention schemes, and form a prevention scheme set corresponding to the risk level and the image category; the purpose is to take different prevention schemes for different risk levels of different image categories, i.e., oral cancer categories, which lays a foundation for subsequent scheme optimization, improves the accuracy and effectiveness of prevention scheme selection, and improves the accuracy and effectiveness of prevention scheme selection; The scheme optimization module is configured to select the optimal prevention scheme through the improved optimization algorithm, and transmit the scheme to the health management module; the improved optimization algorithm adds a neural network to the fitness calculation of the optimization algorithm, which can fully utilize the nonlinear and multi-modal data modeling capabilities of the neural network, effectively capture complex data relationships, improve the computing efficiency according to the parallel computing mechanism, and evaluate the prevention scheme through the deep learning technology; the combination of the neural network and the optimization algorithm can comprehensively and accurately understand the effect of the prevention scheme, ensure that the prevention scheme is the optimal prevention scheme, and has effectiveness and pertinence; The health management module is configured to output the optimal prevention scheme content, perform human-computer interaction display, regularly send health reminders, provide health consultation and medical guidance services, and can be connected with medical institutions to facilitate users to seek medical treatment in time when needed.

[0018] In this embodiment, it needs to be specifically pointed out that the capsule neural network model used in the category analysis module is a kind of improved convolutional neural network; the capsule neural network proposes to encode the data features extracted from the image into the form of vector for output, the modulus of the vector represents the probability of the existence of the feature, and the direction of the vector represents the pose information of the feature, and the moving feature will change the vector without affecting the probability of the existence of the feature; The capsule neural network comprises an input layer, a feature extraction layer, an initial capsule layer, a routing capsule layer, a loss function, a decoder and an output layer; the input layer inputs an image, the feature extraction layer, that is, a feature extraction network, aims to expand the dimension of the image, extracts image features and encodes them into vectors composed of instance parameters, the initial capsule layer is composed of a series of capsules, each capsule is a separate convolutional layer, which respectively extracts information of each part on the feature map, and then outputs a tensor, taking the channel dimension as the vector dimension of the output, and integrating the outputs of each capsule together to output a vector, taking the capsule neural network for MNIST handwritten digit classification as an example, if the initial capsule layer is 32 9*9 convolutional layers, which respectively extract information of each part on the feature map, 32 6*6*8 tensors are output, and then taking the channel dimension as the vector dimension of the output, the outputs of each capsule are integrated together to output 32*6*6 8-dimensional vectors; the routing capsule layer passes all vectors through a dynamic routing structure, repeats until a specified number of iterations, and outputs a category vector from the output layer, the number of category vectors is the number of image categories, and the elements in the category vector represent the features on the picture, and the length represents the probability that the input picture belongs to the category, the category vector is updated after being output through the loss function, and the loss is balanced; the decoder is used to verify that the vector output by the capsule network is indeed the attribute features and pose information of the encoded picture, and the longest vector is extracted and reconstructed into the original image through a fully connected neural network; The calculation formula of the routing capsule layer is represented as: ; ; ; ; Wherein, j=1, 2, 3, …, m; m is the number of initial capsule layer output vectors, a is a category vector, s is a weighted vector sum, b j is the weight of the jth vector, v j is the intermediate vector of the jth vector after transformation, w j is the weight matrix or transformation matrix of the jth vector, u j is the jth vector output by the initial capsule, squash represents the squash function, softmax represents the softmax function, and b j´ is the new weight after the softmax function and is the new b j is substituted and the next round of calculation is performed; formulas ① to ④ are repeated until a specified number of iterations is reached, at which time the final category vector a is output, and the number of a is the number of image categories. For example, if an oral image is input, the number of the final output category vector a is 3, which indicates that the number of image categories is 3, that is, the input oral image corresponds to 3 oral cancer categories, and if the lengths of the 3 category vectors are A1, A2, and A3, respectively, then the probabilities of the oral image corresponding to the 3 oral cancer categories are A1, A2, and A3, respectively; The length of the vector a satisfies a∈(0, 1), formula ① is a nonlinear transformation from a vector to a vector, which compresses the length of the vector to below 1 without changing the direction of the vector; the number of iterations can be determined by a person skilled in the art by training the capsule neural network multiple times until an evaluation index is reached, at which time the training is stopped, and the average of the number of iterations in multiple training is taken as the specified number of iterations, and the evaluation index can be composed of one or more of accuracy, recall rate, precision, and F1 score; The loss function uses a margin loss function, which is expressed as: where n is the number of categories of the image, that is, the number of categories of oral cancer, which includes the normal category, that is, the normal image; a i is the output vector of category i; T i represents whether the image belongs to category i, and takes a value of 0 or 1, that is, 1 if it belongs to, and 0 if it does not belong to; m + is the number of categories of the image, that is, the number of categories of oral cancer, which includes the normal category, that is, the normal image; a - is a label smoothing term, which takes a value of 0.9 and 0.1, and the loss is divided into two parts, the length of the output vector of the true label is forced to approach m + , and the lengths of the remaining output vectors are forced to approach m - ; λ is a balance parameter for balancing the loss of the two parts to prevent a large difference in loss between the two parts when the number of categories is large; i = 1, 2, 3, …, n; The training of the capsule neural network model includes two parts of training set and validation set data, all images of each period and each performance symptom of each category of oral cancer and normal images are collected, and data division is performed according to the training set and the validation set, and the division standard can be determined by a person skilled in the art, and the present embodiment selects 80% training set and 20% validation set; the images of the training set are input into the capsule neural network model as training data for training, and the images of the validation set are input into the capsule neural network model as validation data for model evaluation and validation; The image data pre-processed by the image data pre-processing unit is input into the capsule neural network model, and a category vector is output. The category vector corresponding to the maximum vector length is taken as the category of the input image, i.e., the category of oral cancer corresponding to the image. For example, if the value of A1 is the largest among A1, A2 and A3, the category vector corresponding to the length A1 has the largest probability of the category of oral cancer, and thus the category of the input oral image is the category of oral cancer corresponding to A1, which is transmitted to the risk assessment module and the prevention scheme acquisition module.

[0019] In this embodiment, it is specifically pointed out that the calculation formula of the risk assessment score calculated by the risk assessment module is: wherein, FX is the risk assessment score, γ h is the risk assessment coefficient of the hth risk assessment factor, BD h is the expression amount of the hth risk assessment factor, and H is the total amount of risk assessment factors. The risk assessment factor is the category of text data. For example, smoking is a risk assessment factor. The number of risk assessment factors is the same as the total number of questionnaire survey questions. One question in the questionnaire survey corresponds to one risk assessment factor. The questionnaire survey is in the form of yes or no. For example, whether daily tooth brushing is more than twice, whether smoking, whether alcoholism, whether there is a history of oral cancer in the direct relatives, etc. The expression amount of the risk assessment factor is valued as 0 or 1. When the answer to the question in the questionnaire survey is an answer that is not conducive to oral health, the expression amount is valued as 1, otherwise as 0. For example, when the question is whether daily tooth brushing is more than once, if the user selects yes, the answer indicates that daily tooth brushing is more than once, and the behavior of daily tooth brushing more than once corresponding to the answer is conducive to oral health, so the expression amount of the risk assessment factor corresponding to the question is valued as 0. If the user selects no, the answer indicates that daily tooth brushing is not more than once, and the behavior of daily tooth brushing not more than once corresponding to the answer is not conducive to oral health, so the expression amount of the risk assessment factor corresponding to the question is valued as 1. The calculation formula of the risk assessment coefficient is: wherein, R hN is the total number of samples; the life data and oral health data of N patients with oral cancer are collected in advance as sample data; if the yth sample data has the same data as the adverse answer to the problem corresponding to the hth risk assessment factor, the hth risk assessment factor exists in the sample data; for example, when the problem corresponding to the hth risk assessment factor is whether daily tooth brushing is more than once, the adverse answer to the problem is no; if the answer to the problem by the yth sample data is no, the hth risk assessment factor exists in the yth sample, and thus the number of samples with the answer no in the N sample data is the number of samples with the hth risk assessment factor in the sample data; y = 1, 2, 3, …, N; The problems in the questionnaire survey can be reasonably set by those skilled in the art according to the content of the life data and oral health data, and the problems are not listed one by one in this embodiment.

[0020] In this embodiment, it needs to be specifically explained that the risk assessment module divides the risk level based on the risk assessment score, and the division is as follows: When the risk assessment score FX satisfies FX = 0, it is classified as no risk; When the risk assessment score FX satisfies 0 < FX ≤ YU1, it is classified as low risk level; When the risk assessment score FX satisfies YU2 < FX ≤ YU2, it is classified as medium risk level; When the risk assessment score FX satisfies FX > YU2, it is classified as high risk level; The YU1 and YU2 are both determination thresholds, the YU1 is a first determination threshold, the YU2 is a second determination threshold, YU1 < YU2, and the values of the YU1 and YU2 can be set by those skilled in the art, which are not specifically limited in this embodiment.

[0021] In this embodiment, it needs to be specifically explained that the set of prevention schemes of the prevention scheme acquisition module is represented as: , wherein, J i is the set of prevention schemes of the ith oral cancer category, DE lowi is the set of low-risk prevention measures in the set of prevention schemes of the ith oral cancer category, DE midi is the set of medium-risk prevention measures in the set of prevention schemes of the ith oral cancer category, DE highia high-risk prevention measure set in the prevention scheme set for the i-th oral cancer category; the low-risk prevention measure set, the medium-risk prevention measure set, and the high-risk prevention measure set each include a plurality of corresponding prevention measures, such as correct tooth brushing, use of dental floss, use of mouthwash, professional oral imaging examination, etc., and the prevention measures in each risk prevention measure set can be obtained by collecting effective prevention measures for different risk levels in the sample data and modifying them in combination with professional knowledge and oral professional content of technical personnel; combining the results of the category analysis module and the results of the risk assessment module to obtain a prevention measure set corresponding to the risk level of the corresponding category, all prevention schemes in the set being marked as optional prevention schemes; If the category analysis module outputs a normal category or the risk assessment module outputs no risk, a health reminder instruction is directly sent to the health management module, and at this time the health management module displays the data in human-computer interaction and sends a health reminder to the user, prompting the user to regularly analyze and detect.

[0022] In this embodiment, it needs to be specifically pointed out that the scheme selection module can select the optimal prevention scheme by using any one of natural heuristic algorithms such as a genetic algorithm, a cuckoo optimization algorithm, and a monkey swarm optimization algorithm. In this embodiment, the genetic algorithm is taken as an example for improvement, and the specific mode is as follows: Step S11: encoding the optional prevention scheme, the encoding being O, O being a chromosome, obtaining the chromosome, and constructing an initial population ; Step S12: determining a fitness function; Step S13: natural selection of the chromosomes in the population; Step S14: cross-recombination of the chromosomes in the population; Step S15: mutation of the chromosomes in the population; Step S16: obtaining a new population, the preset population number being L, the fitness threshold being Q, L being an integer greater than 0, and Q being a real number greater than 0; the steps S13 to S15 are cycled until the number of the new population is L or there is a chromosome in the new population whose fitness is greater than or equal to the fitness threshold Q, the cycle is ended, and the optional prevention scheme corresponding to the chromosome with the maximum fitness in the new population is obtained, which is the optimal prevention scheme; for example, if the preset population number is 1, the chromosomes in the initial population are naturally selected and cross-recombined and mutated to obtain a new population, at this time the number of the new population is 1, and therefore the cycle is ended; The fitness threshold Q is preset by a person skilled in the art according to the algorithm accuracy, the population generation number L is obtained by a person skilled in the art under a plurality of different prevention schemes, and the genetic algorithm is used multiple times to obtain a corresponding optimal operation category, in each time of using the genetic algorithm, when the fitness corresponding to a chromosome in a new population is greater than or equal to the fitness threshold Q, the cycle ends, and a generation number corresponding to the new population is obtained; and the maximum generation number in the plurality of generation numbers is taken as the population generation number L.

[0023] In this embodiment, it needs to be specifically explained that the expression of the fitness function is: Wherein, f r is the fitness corresponding to the rth chromosome, YZ r is the risk inhibition value of the candidate prevention scheme corresponding to the rth chromosome; r = 1, 2, 3, …, k; The natural selection is performed by combining the elite method and the rotation method; wherein the elite method generates F1 offspring chromosomes, for a population with a capacity of k, the fitnesses corresponding to the k chromosomes are arranged from large to small, and each of the F1 chromosomes arranged in the front generates an offspring chromosome; the rotation method generates F2 offspring chromosomes, that is, the k chromosomes generate F2 offspring chromosomes according to the corresponding rotation probability; F1+F2=k, so as to keep the offspring population capacity k unchanged and the population generation number increasing; The expression of the rotation probability is: Wherein, ζ r is the rotation probability corresponding to the rth chromosome; In the step S14, E chromosomes are randomly selected from the population for crossover recombination to obtain E new chromosomes; the crossover recombination adopts the PMX method, which is a prior art means, and will not be described in detail in this embodiment; after the crossover recombination of the chromosomes, the fitnesses of the E new chromosomes are calculated, the fitnesses of the E new chromosomes and the fitnesses of the E chromosomes are sorted from large to small to generate a sorting table, and the E new chromosomes in the sorting table are replaced with the E chromosomes in the population for crossover recombination in the positive order; preferably, E = 0.7k in this embodiment, if the calculated E is not an integer, E is rounded up to ensure that the calculated E is an integer; In the step S15, a mutation probability V is preset, and the k chromosomes in the population are mutated according to the mutation probability; the mutation method is to randomly select the positions of two genes in the chromosome and exchange the values of the two genes; preferably, V = 0.02 in this embodiment, and the mutation probability is preset by a person skilled in the art according to the algorithm efficiency and the algorithm accuracy.

[0024] In this embodiment, it needs to be specifically explained that the risk inhibition value is obtained in the following manner: The oral cancer category, risk assessment score and alternative prevention scheme are input into the risk inhibition prediction model as a group of input data, and a risk inhibition value corresponding to the alternative prevention scheme is obtained; The risk inhibition prediction model is a neural network model, and the training method is: Each sample data in the N sample data collected in advance is taken as a group of analysis data, and a risk inhibition value corresponding to each group of analysis data is obtained. The risk inhibition value is obtained by subtracting the risk assessment score before the prevention scheme is adopted from the risk assessment score after the prevention scheme is adopted. For example, if the risk assessment score of the yth sample data is 25, and the risk assessment score is 20 after a certain prevention scheme is adopted, the risk inhibition value is -5. The use time of all prevention schemes is consistent. A fixed scheme use time can be set in advance to determine the risk inhibition value of each scheme in the same time, and a more suitable prevention scheme can be selected according to the length of time. The analysis data and the corresponding risk inhibition value are converted into a corresponding group of feature vectors; Each group of feature vectors is taken as the input of the risk inhibition prediction model, the risk inhibition prediction model takes a predicted risk inhibition value corresponding to each group of analysis data as the output, and takes the actual risk inhibition value corresponding to each group of analysis data as the prediction target. The actual risk inhibition value can be obtained through the pre-collected data. The training target is to minimize the sum of prediction errors of all analysis data. The formula of the prediction error is: , wherein ε p is the prediction error, p is the group number of the feature vector corresponding to the analysis data, θ p is the predicted risk inhibition value corresponding to the pth group of analysis data, μ p is the actual risk inhibition value corresponding to the pth group of analysis data. The risk inhibition prediction model is trained until the sum of prediction errors reaches convergence.

[0025] In the embodiment, it needs to be specifically pointed out that the difference between the embodiment and the prior art is mainly that the embodiment has a category analysis module, a risk assessment module and a scheme optimization module, an image category corresponding to image data is obtained based on a capsule neural network model, that is, an oral cancer category corresponding to the image, a risk assessment score is calculated based on life data and oral health data, and a risk level is divided based on the risk assessment score, finally, an alternative prevention scheme is obtained based on the obtained oral cancer category and risk level, and the optimal prevention scheme is selected through the improved optimization algorithm; the user can answer the questionnaire problems and upload the oral image, so as to perform the risk self-test of oral cancer, and obtain the corresponding prevention scheme, the risk factors are analyzed by comprehensively considering the individual life habit, the risk level of the individual oral cancer is comprehensively evaluated, it is more efficient, more convenient and more effective, the early screening ability can be enhanced, and the incidence rate of oral cancer can be reduced.

[0026] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application, should be included in the protection scope of the present application.

[0027] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A big data-based oral cancer risk assessment and prevention system, characterized by: It includes data collection module, data preprocessing module, category analysis module, risk assessment module, prevention plan acquisition module, plan selection module and health management module; The data collection module is used to collect user data and transmit it to the data preprocessing module. The data collection module includes a text data acquisition unit and an image data acquisition unit; The data preprocessing module is used to perform preprocessing operations on the data of the data collection module, and the data preprocessing module includes a text data preprocessing unit and an image data preprocessing unit; The category analysis module is used to receive the pre-processed image data, obtain the image category corresponding to the image data based on the capsule neural network model, and then transmit it to the risk assessment module and the prevention plan acquisition module; The risk assessment module is used to receive data from the text data preprocessing unit and the category analysis module, calculate the risk assessment score, and divide the risk level based on the risk assessment score; The prevention solution acquisition module is used to acquire a plurality of different prevention solutions and form a prevention solution set corresponding to the risk level and image category; The optimal solution selection module is used to select the optimal prevention solution through the improved optimization algorithm and transmit the solution to the health management module; The health management module is used to output the content of the optimal prevention plan, perform human-computer interactive display and send health reminders regularly, provide health consultation and medical guidance services, and can also connect with medical institutions.

2. The oral cancer risk assessment and prevention system based on big data according to claim 1, characterized in that: The text data acquisition unit is used to collect the user's text data and transmit it to the data preprocessing module, and the image data acquisition unit is used to collect the user's image data and transmit it to the data preprocessing module; The text data preprocessing unit is used to preprocess the text data of the text data acquisition unit and transmit it to the risk assessment module; the image data preprocessing unit is used to preprocess the image data of the image data acquisition unit and transmit it to the category analysis module; The user data includes user text data and user image data, the text data includes life data and oral health data, and the image data is an oral image; the user is a subject who needs oral cancer screening or risk assessment; the life data includes the user's dietary preferences, oral hygiene habits and living habits; the oral health data includes the user's oral symptom description, oral disease history, treatment history and genetic history; the oral symptom description is a description of the recent oral condition.

3. The oral cancer risk assessment and prevention system based on big data according to claim 2, characterized in that: The capsule neural network includes an input layer, a feature extraction layer, an initial capsule layer, a routing capsule layer, a loss function, a decoder, and an output layer; the input layer inputs an image, the feature extraction layer is a feature extraction network, the initial capsule layer is composed of a series of capsules, each capsule is a separate convolution layer, and each capsule extracts different parts of information on the feature map. The routing capsule layer passes all vectors through a dynamic routing structure and repeats until a specified number of iterations is reached. The output layer outputs a category vector, the number of category vectors is the number of image categories, the elements in the category vector represent the features on the image, and the length represents the probability that the input image belongs to the category; The calculation formula of the routing capsule layer is expressed as: ； ； ； ； Where j = 1, 2, 3, ..., m; m is the number of output vectors of the initial capsule layer, a is the category vector, s is the weighted vector sum, b j is the weight of the jth vector, v j is the intermediate vector after the transformation of the jth vector, w j is the weight matrix or transformation matrix of the jth vector, u j is the jth vector output by the initial capsule, squash represents the squash function, softmax represents the softmax function, and b j ´ is the new weight after the softmax function and serves as the new b j Substitute and perform the next round of calculation; repeat formulas ① to ④ until the specified number of iterations is reached, at which point the final category vector a is output. The number of a is the number of image categories. The loss function adopts the marginal loss function, which is expressed as follows: , where n is the number of image categories, i.e. the number of oral cancer categories, including the normal category, i.e. normal images; a i is the output vector of category i; T i Indicates whether the image belongs to category i, with a value of 0 or 1. If it does, the value is 1, and if it does not, the value is 0; m + With m - is the label smoothing term, with values ​​of 0.9 and 0.

1. The loss is divided into two parts, and the output vector length of the true label should be close to m + , the length of the remaining output vectors should be close to m - ;λ is the balance parameter, i=1, 2, 3,…, n; The image data preprocessed by the image data preprocessing unit is input into the capsule neural network model, and a category vector is output. The category vector corresponding to the maximum vector length is used as the category of the input image, that is, the oral cancer category corresponding to the image.

4. The oral cancer risk assessment and prevention system based on big data according to claim 3, characterized in that: The risk assessment module calculates the risk assessment score using the following formula: , where FX is the risk assessment score, γ h is the risk assessment coefficient of the h-th risk assessment factor, BD h is the expression value of the hth risk assessment factor, H is the total number of risk assessment factors; the risk assessment factor is the text data category, and the number of risk assessment factors is the same as the total number of questionnaire questions; The calculation formula of the risk assessment coefficient is expressed as: , where R h is the number of h-th risk assessment factors in the sample data, and N is the total number of samples. The life data and oral health data of N patients with oral cancer are collected in advance as sample data. If the y-th sample data contains data that has the same unfavorable answer as the question corresponding to the h-th risk assessment factor, it means that the h-th risk assessment factor exists in the sample data.

5. The oral cancer risk assessment and prevention system based on big data according to claim 4, characterized in that: The risk assessment module divides the risk levels based on the risk assessment scores into the following categories: When the risk assessment score FX satisfies FX=0, it is classified as risk-free; When the risk assessment score FX satisfies 0<FX≤YU1, it is classified as a low risk level; When the risk assessment score FX satisfies YU2<FX≤YU2, it is classified as medium risk level; When the risk assessment score FX satisfies FX>YU2, it is classified as a high risk level; Both YU1 and YU2 are determination thresholds, YU1 is a first determination threshold, YU2 is a second determination threshold, and YU1<YU2.

6. The oral cancer risk assessment and prevention system based on big data according to claim 5, characterized in that: The prevention solution set of the prevention solution acquisition module is expressed as: , where J i is the set of prevention programs for the i-th oral cancer category, DE lowi is the set of low-risk prevention measures in the prevention program set for the i-th oral cancer category, DE midi is the set of medium-risk prevention measures in the prevention program set for the i-th oral cancer category, DE highi is the set of high-risk prevention measures in the prevention program set for the i-th oral cancer category; Combine the results of the category analysis module with the results of the risk assessment module to obtain a set of preventive measures corresponding to the risk level of the corresponding category. All preventive measures in this set are marked as alternative preventive measures. If the output result of the category analysis module is a normal category or the output result of the risk assessment module is no risk, a health reminder instruction is directly sent to the health management module. At this time, the health management module displays the data in a human-computer interactive manner and sends a health reminder to the user.

7. The oral cancer risk assessment and prevention system based on big data according to claim 6, characterized in that: The specific method for the scheme selection module to select the optimal prevention scheme is as follows: Step S11: Encode the alternative prevention plan, encode it as O, O is the chromosome, obtain the chromosome, and construct the initial population ; Step S12: determining the fitness function; Step S13: performing natural selection on chromosomes in the population; Step S14: performing crossover recombination on the chromosomes in the population; Step S15: mutating the chromosomes in the population; Step S16: Obtain a new population, preset the population generation to be L, the fitness threshold to be Q, where L is an integer greater than 0 and Q is a real number greater than 0; loop steps S13 to S15 until the generation corresponding to the new population is L or the fitness corresponding to the chromosome in the new population is greater than or equal to the fitness threshold Q, then the loop ends, and the alternative prevention plan corresponding to the chromosome with the maximum fitness in the new population is obtained, which is the optimal prevention plan.

8. The oral cancer risk assessment and prevention system based on big data according to claim 7, characterized in that: The expression of the fitness function is: , where f r is the fitness corresponding to the rth chromosome, YZ r is the risk suppression value of the alternative prevention plan corresponding to the rth chromosome; r = 1, 2, 3, ..., k; The natural selection is carried out by combining the elitist method and the rotation method; The expression of the rotation probability is: , where ζ r is the rotation probability corresponding to the rth chromosome.

9. The oral cancer risk assessment and prevention system based on big data according to claim 8, characterized in that: The risk suppression value is obtained as follows: The oral cancer category, risk assessment score, and alternative prevention options are used as a set of input data and input into the risk suppression prediction model to obtain the risk suppression value corresponding to the alternative prevention option; The risk suppression prediction model is a neural network model, and the training method is: The oral cancer category, risk assessment score, and adopted prevention plan of each sample data in the N pre-collected sample data are used as a set of analysis data. A risk suppression value corresponding to each set of analysis data is obtained by subtracting the risk assessment score before the prevention plan was adopted from the risk assessment score after the prevention plan was adopted. The analysis data and the corresponding risk suppression value are converted into a corresponding set of feature vectors. Each set of feature vectors is used as the input of the risk suppression prediction model. The risk suppression prediction model outputs a predicted risk suppression value corresponding to each set of analysis data, and uses the actual risk suppression value corresponding to each set of analysis data as the prediction target. The actual risk suppression value can be obtained through pre-collected data. The training goal is to minimize the sum of the prediction errors of all analysis data. The formula for the prediction error is expressed as: , where ε p is the prediction error, p is the group number of the eigenvector corresponding to the analysis data, θ p is the predicted risk suppression value corresponding to the p-th group analysis data, μ p The actual risk suppression value corresponding to the p-th group of analysis data is used to train the risk suppression prediction model until the sum of the prediction errors reaches convergence.

Images