Erash disease prediction method and system based on convolutional neural network

Abstract

The invention relates to the field of eruption disease prediction, and discloses an eruption disease prediction method and system based on a convolutional neural network, and the method comprises the steps: firstly collecting a preliminary image of a patient; marking to obtain a preliminary image data set of the patient; training a patient image convolutional neural network; performing a first classification process to obtain a first patient image group, and extracting feature information of the patient preliminary image; performing eruption disease recognition analysis to obtain a patient image analysis result, and performing a second classification process on the patient preliminary image to obtain a second patient image group; establishing an eruption estimation tree diagram according to the second patient image group, and establishing an eruption feature vector according to the feature information of the patient preliminary image; estimating the eruption condition of the patient, and establishing a multi-dimensional eruption disease estimation map; the problems that in the prior art, eruption disease prediction is not accurate, the existing eruption disease prediction time period is long, the treatment time is later, the treatment difficulty is increased, and the treatment time is prolonged are solved.

Description

Technical Field

[0001] This invention relates to the field of rash disease prediction, and more particularly to a method and system for predicting rash diseases based on convolutional neural networks. Background Technology

[0002] Fever-rash-like illnesses (FRIs) are diseases characterized by fever and rash as the main clinical manifestations, including measles, rubella, and other RFIs, often occurring in outbreaks. The incidence of FRIs is high, especially among infants and young children, posing a significant threat to human health and social development. While the clinical symptoms of these diseases are generally mild, the serious complications and histological similarities make accurate diagnosis of the pathogen and appropriate treatment crucial.

[0003] Currently, most diagnostic methods for febrile rash-related diseases, such as pathogen culture, serological diagnosis, and molecular diagnosis, are not accurate or efficient enough, delaying treatment and even causing serious consequences. With the application of convolutional neural networks in various fields, convolutional neural networks can predict and analyze diseases including rash-related diseases through image recognition and analysis. However, there is currently no relevant technology to apply convolutional neural networks to the prediction, diagnosis, and analysis of rash-related diseases. Therefore, there is an urgent need for a method that can quickly and accurately predict and diagnose rash-related diseases using convolutional neural networks. Summary of the Invention

[0004] The present invention aims to provide a method and system for predicting exanthematous diseases based on convolutional neural networks, in order to solve the problems of inaccurate prediction of exanthematous diseases in the existing technology, the long prediction time of existing exanthematous diseases, resulting in delayed treatment time, increased treatment difficulty and prolonged treatment time.

[0005] To achieve the above objectives, the present invention provides the following method:

[0006] This invention provides a method for predicting rash-causing diseases based on convolutional neural networks:

[0007] S1: Acquire photographs of various parts of the patient, including images of different parts under different lighting conditions, to obtain a preliminary image of the patient;

[0008] S2: Annotate the preliminary patient images to obtain a preliminary patient image dataset;

[0009] S3: Train a convolutional neural network for patient images based on the preliminary patient image dataset;

[0010] S4: Perform a first classification process on the preliminary patient images using the patient image convolutional neural network to obtain a first patient image group, and extract the feature information of the preliminary patient images;

[0011] S5: Perform rash disease identification analysis on the feature information of the preliminary patient image to obtain patient image analysis results, bind the patient image analysis results with the preliminary patient image, and perform a second classification process on the preliminary patient image to obtain a second patient image group;

[0012] S6: Establish a rash prediction tree diagram based on the second patient image group, and establish a rash feature vector based on the feature information of the preliminary patient image corresponding to the image in the second patient image group;

[0013] S7: Based on the changes in the rash prediction tree diagram and rash feature vector, predict the patient's rash status and establish a multidimensional rash disease prediction map.

[0014] Preferably, the step of acquiring photographs of various parts of the patient, including images of different parts under different lighting conditions, to obtain a preliminary image of the patient includes: illuminating the patient's limbs, abdomen, back, and face with different types of light from different angles, and acquiring photographs of the patient's hands, legs, abdomen, back, and face to obtain a preliminary image of the patient; the different types of light include different lighting conditions such as dim nighttime light, natural light, and indoor lighting, and the different angles include frontal, side, and oblique shooting angles.

[0015] Preferably, the step of annotating the preliminary patient images to obtain a preliminary patient image dataset includes: comparing the preliminary patient images under different lighting conditions and at the same angle, and annotating abnormal areas of the patient's skin under different lighting conditions in the preliminary patient images to obtain a first annotated region; comparing the preliminary patient images under the same lighting conditions and at different angles, and annotating abnormal areas of the patient's skin under different angles in the preliminary patient images to obtain a second annotated region; and integrating the first annotated region and the second annotated region of the same patient in the preliminary patient images of the same location to obtain a preliminary patient image dataset.

[0016] Preferably, the step of training a patient image convolutional neural network based on the preliminary patient image dataset includes: training the patient image convolutional neural network to be trained using the preliminary patient image dataset, wherein the convolutional neural network to be trained includes one or more convolutional kernels, each of the convolutional kernels including multiple channels, for convolutional processing of image features including multiple channels; validating the trained convolutional neural network by clustering the channel vectors of the image features based on the preliminary patient image dataset to obtain a set of reference channel vectors; for each convolutional kernel used for convolutional processing of the image features, calculating the dot product of each channel vector of the convolutional kernel with each reference channel vector of the image features to obtain a channel vector dot product table; storing the set of reference channel vectors and one or more channel vector dot product tables as parameters of the trained convolutional neural network model, wherein each channel vector dot product table corresponds to a convolutional kernel; wherein the channel vector is a vector composed of values ​​at the same coordinates in different channels of the image features including multiple channels, or a vector composed of values ​​at the same coordinates in different channels of the convolutional kernel including multiple channels.

[0017] Preferably, the step of performing a first classification process on the preliminary patient images using the patient image convolutional neural network to obtain a first patient image group, and extracting feature information from the preliminary patient images, includes: classifying the preliminary patient images according to the same body parts of the same patient using the patient image convolutional neural network to obtain a first patient image group; expanding the preliminary patient image dataset in the first patient image group by uniformly expanding outwards with each pixel as an expansion unit, based on the first and second labeled regions as the central regions, until the central region is connected to an adjacent central region, and recording the straight-line distance between the center points of the two connected central regions to obtain the abnormal region distance; establishing a feature information analysis map of the preliminary patient images based on the first labeled region, the second labeled region, and multiple abnormal region distances; and extracting the feature information of the preliminary patient images based on the feature information analysis map of the preliminary patient images.

[0018] Preferably, the step of establishing a feature information analysis map of the patient's preliminary image based on the first labeled area, the second labeled area, and the distances of multiple abnormal areas, and extracting feature information of the patient's preliminary image based on the feature information analysis map of the patient's preliminary image, includes: monitoring the changes in the size of the first labeled area and the second labeled area, and establishing a feature information analysis map of the patient's preliminary image based on the changes in the size of the areas; binding the distances of multiple abnormal areas to the first labeled area and the second labeled area corresponding to the feature information analysis map of the patient's preliminary image; monitoring the changes in the size of the first labeled area and the second labeled area in the feature information analysis map of the patient's preliminary image, as well as the changes in the distances of the multiple abnormal areas corresponding to them, recording the distances of the first labeled area, the second labeled area, and the abnormal area that shrinks the most at each stage of change, and recording the change cycle and pattern of the distances of the first labeled area, the second labeled area, and the abnormal area, to obtain the feature information of the patient's preliminary image.

[0019] Preferably, the steps of performing rash disease identification analysis on the feature information of the preliminary patient image to obtain patient image analysis results, binding the patient image analysis results with the preliminary patient image, and performing a second classification process on the preliminary patient image to obtain a second patient image group include: performing rash disease identification analysis based on the feature information of the preliminary patient image, with a weekly cycle; if the distance between the first labeled area, the second labeled area, and the abnormal area changes periodically, then the feature information of the preliminary patient image is marked as a first rash disease category; if the range of the first labeled area and the second labeled area gradually increases, and the distance between the abnormal areas gradually decreases, and the... If the changes in the range of the first labeled region, the second labeled region, and the distance to the abnormal region are not periodic, then the feature information of the patient's preliminary image is labeled as a second rash disease category; if the changes in the range of the first labeled region, the second labeled region, and the distance to the abnormal region are not significant, then the feature information of the patient's preliminary image is labeled as a third rash disease category; the patient image analysis results are bound to the patient's preliminary image, and a second classification process is performed on the patient's preliminary image, classifying the patient's preliminary image according to the first rash disease category, the second rash disease category, and the third rash disease category based on different labels, to obtain a second patient image group.

[0020] Preferably, the rash prediction tree diagram is constructed by using each patient's personal information as the main point, and then branching downwards from the main point. The first branch branch represents the preliminary images of the patient at different locations. A second branch branch then expands downwards from the first branch branch, representing the first and second patient image groups at corresponding locations. A third branch branch then expands from the second branch branch, representing the rash disease identification and analysis status. This status includes the current rash disease identification and analysis result and the rash disease identification and analysis prediction result. The current rash disease identification and analysis result comprises the feature information of the preliminary patient images corresponding to the first, second, and third rash disease categories. The rash disease identification and analysis prediction result is obtained by inputting the monitoring results of the feature information of the preliminary patient images into a big data prediction model.

[0021] Preferably, the step of establishing a rash feature vector based on the feature information of the preliminary patient image corresponding to the image in the second patient image group includes: establishing a rash feature vector based on the feature information of the preliminary patient image corresponding to the image in the second patient image group; obtaining a rash disease probability gradient based on the first rash disease category, the second rash disease category, and the third rash disease category to which the preliminary patient image belongs, wherein the rash disease probability of the first rash disease category is 40%, the rash disease probability of the second rash disease category is 70%, and the rash disease probability of the third rash disease category is 20%; the rash feature vector includes the corresponding rash disease probability gradient to which the preliminary patient image belongs, the number of the first labeled regions, and the number of the second labeled regions.

[0022] This invention provides a rash disease prediction system based on convolutional neural networks, characterized in that the system comprises:

[0023] Image acquisition module: Acquires photographs of various parts of the patient, including images of different parts under different lighting conditions, to obtain a preliminary image of the patient;

[0024] Convolutional Neural Network Training Module: Annotates the preliminary patient images to obtain a preliminary patient image dataset; trains a patient image convolutional neural network based on the preliminary patient image dataset;

[0025] Feature extraction module: Performs a first classification process on the preliminary patient images using the patient image convolutional neural network to obtain a first patient image group, and extracts the feature information of the preliminary patient images;

[0026] Rash Disease Identification and Analysis Module: Performs rash disease identification and analysis on the feature information of the preliminary patient image to obtain patient image analysis results, binds the patient image analysis results with the preliminary patient image, and performs a second classification process on the preliminary patient image to obtain a second patient image group;

[0027] Rash prediction module: A rash prediction tree diagram is established based on the second patient image group, and a rash feature vector is established based on the feature information of the patient's preliminary image corresponding to the image in the second patient image group; the patient's rash status is predicted based on the changes in the rash prediction tree diagram and the rash feature vector, and a multidimensional rash prediction map is established.

[0028] The beneficial effects of this invention are as follows: By analyzing the collected patient images, this invention covers all key areas where patients are most prone to rashes, and uses different lighting and angles for acquisition, making the pre-rash symptoms more obvious. By training a convolutional neural network to analyze the collected images, a first classification process for patients of the same type and a second classification process for different patients of different types are performed to extract and identify the feature information of the images, obtaining preliminary rash disease prediction results. Then, a rash prediction dendrogram that can dynamically monitor changes and a rash feature vector that includes all patient feature information are established for secondary prediction, resulting in a multidimensional rash disease prediction map that can monitor and analyze the patient's rash disease in real time. This can predict the probability of the patient's rash disease in the shortest time, thereby enabling treatment and greatly reducing the difficulty and cycle of treatment. Attached Figure Description

[0029] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0030] Figure 1 A flowchart illustrating a method for predicting rash-causing diseases based on a convolutional neural network, provided in an embodiment of the present invention;

[0031] Figure 2 This is a flowchart illustrating a rash prediction system based on a convolutional neural network, provided as an embodiment of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0034] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0035] Currently, most diagnostic methods for febrile rash-related diseases, such as pathogen culture, serological diagnosis, and molecular diagnosis, are not accurate or efficient enough, delaying treatment and even causing serious consequences. With the application of convolutional neural networks in various fields, convolutional neural networks can predict and analyze diseases including rash-related diseases through image recognition and analysis. However, there is currently no relevant technology to apply convolutional neural networks to the prediction, diagnosis, and analysis of rash-related diseases. Therefore, there is an urgent need for a method that can quickly and accurately predict and diagnose rash-related diseases using convolutional neural networks.

[0036] The present invention aims to provide a method and system for predicting exanthematous diseases based on convolutional neural networks, in order to solve the problems of inaccurate prediction of exanthematous diseases in the existing technology, the long prediction time of existing exanthematous diseases, resulting in delayed treatment time, increased treatment difficulty and prolonged treatment time.

[0037] like Figure 1 As shown in the figure, a specific embodiment of the present invention provides a method for predicting rash-causing diseases based on convolutional neural networks, comprising the following steps:

[0038] S1: Collect photos of various parts of the patient's body, including images of different parts under different lighting conditions, to obtain preliminary images of the patient.

[0039] In this embodiment of the invention, different types of light are used to illuminate the patient's limbs, abdomen, back, and face from different angles to collect photos of the patient's hands, legs, abdomen, back, and face, thus obtaining a preliminary image of the patient. The different types of light include different lighting conditions such as dim night light, natural light, and indoor lighting. The different angles include frontal, side, and oblique shooting angles. It also includes photos of various body parts of the patient in different states, such as the state of the arm when the muscles are tense and when it is relaxed, and the state of the face when it is calm and when it is accompanied by different emotions.

[0040] S2: Annotate the preliminary patient images to obtain the preliminary patient image dataset.

[0041] In this embodiment of the invention, preliminary patient images under different lighting conditions and at the same angle are compared, and abnormal areas of the patient's skin under different lighting conditions in the preliminary patient images are marked to obtain a first marked area; preliminary patient images under the same lighting conditions and at different angles are compared, and abnormal areas of the patient's skin under different angles in the preliminary patient images are marked to obtain a second marked area; the first marked area and the second marked area of ​​the same patient are integrated into preliminary patient images of the same location to obtain a preliminary patient image dataset.

[0042] S3: Train a convolutional neural network for patient images based on the initial patient image dataset.

[0043] In this embodiment of the invention, a patient image convolutional neural network is trained using a preliminary patient image dataset. The convolutional neural network includes one or more convolutional kernels, each kernel including multiple channels, for convolutional processing of image features including multiple channels. The trained convolutional neural network is validated by clustering the channel vectors of the image features based on the preliminary patient image dataset to obtain a set of reference channel vectors. For each convolutional kernel used for convolutional processing of image features, the dot product of each channel vector of the convolutional kernel and each reference channel vector of the image features is calculated to obtain a channel vector dot product table. A set of reference channel vectors and one or more channel vector dot product tables are stored as parameters of the trained convolutional neural network model, wherein each channel vector dot product table corresponds to a convolutional kernel. The channel vector is a vector composed of values ​​at the same coordinates in different channels of the image features including multiple channels, or a vector composed of values ​​at the same coordinates in different channels of the convolutional kernel including multiple channels.

[0044] S4: Perform the first classification process on the preliminary patient images using a convolutional neural network to obtain the first patient image group, and extract the feature information of the preliminary patient images.

[0045] In this embodiment of the invention, preliminary patient images are classified according to the same body parts of the same patient using a patient image convolutional neural network to obtain a first patient image group; the preliminary patient image dataset in the first patient image group is expanded, using the first and second labeled regions as the central regions, and expanding uniformly outwards with each pixel as the expansion unit until the central region connects with an adjacent central region, and the straight-line distance between the center points of the two connected central regions is recorded to obtain the abnormal region distance; a feature information analysis map of the preliminary patient images is established based on the first labeled region, the second labeled region, and multiple abnormal region distances, and the feature information of the preliminary patient images is extracted based on the feature information analysis map of the preliminary patient images; the monitoring of the first... The changes in the size of the first and second labeled regions are analyzed, and a feature information analysis map of the patient's preliminary image is established based on these changes. The distances of multiple abnormal regions are bound to the first and second labeled regions corresponding to the feature information analysis map of the patient's preliminary image. The changes in the size of the first and second labeled regions in the feature information analysis map of the patient's preliminary image, as well as the changes in the distances of the corresponding multiple abnormal regions, are monitored. The distances of the first and second labeled regions with the largest changes and the abnormal regions with the largest reductions in distance are recorded at each stage. The periodicity and pattern of the changes in the distances of the first and second labeled regions and the abnormal regions are also recorded to obtain the feature information of the patient's preliminary image.

[0046] S5: Perform rash disease identification analysis on the feature information of the patient's preliminary image to obtain the patient image analysis results. Bind the patient image analysis results to the patient's preliminary image and perform a second classification process on the patient's preliminary image to obtain the second patient image group.

[0047] In this embodiment of the invention, a rash disease identification analysis is performed based on the feature information of the patient's preliminary image. Using a weekly cycle, if the distances between the first labeled region, the second labeled region, and the abnormal region change periodically, the feature information of the patient's preliminary image is labeled as a first rash disease category. If the ranges of the first labeled region and the second labeled region gradually increase, the distances of the abnormal regions gradually decrease, and the changes in the ranges of the first labeled region, the second labeled region, and the abnormal region do not show periodicity, the feature information of the patient's preliminary image is labeled as a second rash disease category. If the ranges of the first labeled region, the second labeled region, and the abnormal region do not change significantly, the feature information of the patient's preliminary image is labeled as a third rash disease category. The patient image analysis results are bound to the patient's preliminary image, and a second classification process is performed on the patient's preliminary image. The patient's preliminary image is classified according to different labels into the first rash disease category, the second rash disease category, and the third rash disease category, resulting in a second patient image group. The second patient image group also includes a fourth patient image group where the ranges of the first labeled region, the second labeled region, and the abnormal region do not change.

[0048] S6: Establish a rash prediction dendrogram based on the second patient image group, and establish a rash feature vector based on the feature information of the patient's preliminary image corresponding to the image in the second patient image group.

[0049] In this embodiment of the invention, the rash prediction tree diagram is constructed with each patient's personal information as the main point. A first branch expands downwards from the main point, representing preliminary images of different body parts of the patient. A second branch expands downwards from the first branch, representing a first and second patient image group corresponding to the same body part. A third branch expands from the second branch, representing the rash disease identification and analysis status, including the current rash disease identification and analysis result and the rash disease identification and analysis prediction result. The current rash disease identification and analysis result includes the feature information of the preliminary patient images corresponding to the first, second, and third rash disease categories. The rash disease identification and analysis prediction result is obtained by inputting the monitoring results of the feature information of the preliminary patient images into a big data dataset. The predictive model obtains the estimated results of rash disease identification and analysis. The steps for establishing a rash feature vector based on the feature information of the preliminary patient images corresponding to the images in the second patient image group include: establishing a rash feature vector from the feature information of the preliminary patient images corresponding to the images in the second patient image group; obtaining the rash disease probability gradient based on the first, second, and third rash disease categories to which the preliminary patient images belong, with the probability of rash disease in the first category being 40%, the probability in the second category being 70%, and the probability in the third category being 20%; the rash feature vector includes the corresponding rash disease probability gradient to which the preliminary patient images belong, the number of first and second labeled regions, and the different rash disease probabilities represent different directions of the rash feature vector, with the rash feature vector angle corresponding to 20% being 30 degrees. o The rash feature vector corresponding to 40% has an angle of 60°. o The rash feature vector corresponding to 70% has an angle of 75 degrees. o The rash feature vector angle for the fourth category of rash-causing diseases is 0.

[0050] S7: Based on the rash prediction tree diagram and the changing status of the rash feature vector, predict the patient's rash status and establish a multidimensional rash disease prediction map.

[0051] In this embodiment of the invention, a first rash disease prediction probability change value is obtained based on the real-time rash disease identification and prediction results of the rash prediction tree diagram; a second rash disease prediction probability change value is obtained based on the real-time change status of the rash feature vector; a multidimensional rash disease prediction map is established based on the sum of the first and second rash disease prediction probability change values; after establishing the multidimensional rash disease prediction map, epidemiological data modeling is also included: epidemiological data is input into a big data analysis model to obtain the influence coefficient of epidemiological data on rash diseases; the multidimensional rash disease prediction map is adjusted accordingly based on the influence coefficient of epidemiological data on rash diseases to obtain a corrected multidimensional rash disease prediction map; and the final result of rash disease prediction is obtained by analyzing the corrected multidimensional rash disease prediction map.

[0052] like Figure 2 As shown, the present invention provides a rash disease prediction system based on convolutional neural networks, characterized in that the system includes:

[0053] Image acquisition module: Acquires photos of various parts of the patient, including images of different parts under different lighting conditions, to obtain preliminary images of the patient;

[0054] Convolutional Neural Network Training Module: Annotates preliminary patient images to obtain a preliminary patient image dataset; trains a convolutional neural network based on the preliminary patient image dataset;

[0055] Feature extraction module: Performs a first classification process on the preliminary patient images using a convolutional neural network to obtain the first patient image group, and extracts the feature information of the preliminary patient images;

[0056] Rash Disease Identification and Analysis Module: Performs rash disease identification and analysis on the feature information of the patient's preliminary image to obtain the patient image analysis results, binds the patient image analysis results with the patient's preliminary image, and performs a second classification process on the patient's preliminary image to obtain a second patient image group;

[0057] Rash prediction module: A rash prediction tree diagram is built based on the second patient image group, and a rash feature vector is built based on the feature information of the patient's preliminary image corresponding to the image in the second patient image group; the patient's rash status is predicted based on the changes in the rash prediction tree diagram and the rash feature vector, and a multidimensional rash prediction map is built.

[0058] The beneficial effects of this invention are as follows: By analyzing the collected patient images, this invention covers all key areas where patients are most prone to rashes, and uses different lighting and angles for acquisition, making the pre-rash symptoms more obvious. By training a convolutional neural network to analyze the collected images, a first classification process for patients of the same type and a second classification process for different patients of different types are performed to extract and identify the feature information of the images, obtaining preliminary rash disease prediction results. Then, a rash prediction dendrogram that can dynamically monitor changes and a rash feature vector that includes all patient feature information are established for secondary prediction, resulting in a multidimensional rash disease prediction map that can monitor and analyze the patient's rash disease in real time. This can predict the probability of the patient's rash disease in the shortest time, thereby enabling treatment and greatly reducing the difficulty and cycle of treatment.

[0059] The above descriptions are merely embodiments of the present invention. Commonly known technical solutions or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A rash disease prediction method based on a convolutional neural network, characterized by, The method comprises: S1: collecting photographs of each part of a patient, the photographs of each part of the patient comprising images under different light conditions of different parts, to obtain preliminary images of the patient; S2: labeling the preliminary images of the patient to obtain a preliminary image dataset of the patient; S3: training a patient image convolutional neural network according to the preliminary image dataset of the patient; S4: performing a first classification process on the preliminary images of the patient by using the patient image convolutional neural network, obtaining a first patient image group, and extracting feature information of the preliminary images of the patient; S5: performing rash disease recognition analysis on the feature information of the preliminary images of the patient to obtain patient image analysis results, binding the patient image analysis results with the preliminary images of the patient, and performing a second classification process on the preliminary images of the patient to obtain a second patient image group; S6: establishing a rash estimation tree diagram according to the second patient image group, and establishing a rash feature vector according to the feature information of the preliminary images of the patient corresponding to the images in the second patient image group; S7: performing patient rash condition estimation according to the change state of the rash estimation tree diagram and the rash feature vector, and establishing a multi-dimensional rash disease estimation diagram. 2.The rash disease prediction method based on convolutional neural network according to claim 1, characterized in that, The step of collecting photographs of each part of a patient, the photographs of each part of the patient comprising images under different light conditions of different parts, to obtain preliminary images of the patient, comprises: Irradiating the limbs, abdomen, back and face of the patient at different angles with different kinds of light, collecting photographs of the hands, legs, abdomen, back and face of the patient, and obtaining preliminary images of the patient; The different kinds of light comprise night dim light, natural light, and different light conditions of indoor light, and the different angles comprise front, side and oblique shooting angles. 3.The rash disease prediction method based on convolutional neural network according to claim 2, characterized in that, The step of labeling the preliminary images of the patient to obtain a preliminary image dataset of the patient comprises: Comparing the preliminary images of the patient under different light conditions and the same angle, labeling abnormal areas of the patient's skin under different light conditions in the preliminary images of the patient, and obtaining a first labeled area; Comparing the preliminary images of the patient under the same light condition and different angles, labeling abnormal areas of the patient's skin under different angles in the preliminary images of the patient, and obtaining a second labeled area; Integrating the first labeled area and the second labeled area of the same patient in the preliminary images of the patient at the same part to obtain a preliminary image dataset of the patient. 4.The rash disease prediction method based on convolutional neural network according to claim 1, characterized in that, The step of training a patient image convolutional neural network according to the preliminary image dataset of the patient comprises: Training a patient image convolutional neural network to be trained using a preliminary image dataset of the patient, wherein the convolutional neural network to be trained comprises one or more convolutional kernels, each of which comprises a plurality of channels for convolution processing of image features comprising a plurality of channels; Verifying the trained convolutional neural network, clustering channel vectors of image features based on the preliminary image dataset of the patient to obtain a group of reference channel vectors; For each convolution kernel used for convolution processing of the image feature, a dot product result of each channel vector of the convolution kernel and each reference channel vector of the image feature is calculated to obtain a channel vector dot product table; The set of reference channel vectors and one or more channel vector dot product tables are stored as parameters of the trained convolutional neural network model, wherein each channel vector dot product table corresponds to a convolution kernel respectively; The channel vector is a vector composed of values at the same coordinates in different channels of the image feature including multiple channels, or a vector composed of values at the same coordinates in different channels of the convolution kernel. 5.The rash disease prediction method based on convolutional neural network according to claim 3, characterized in that, The step of performing a first classification process on the patient preliminary image through the patient image convolutional neural network to obtain a first patient image group and extracting feature information of the patient preliminary image, comprising: Classifying the patient preliminary image according to the same part of the same patient through the patient image convolutional neural network to obtain a first patient image group; The patient preliminary image data set in the first patient image group is expanded, and each pixel point is taken as an expansion unit to uniformly expand to the periphery according to the first annotation region and the second annotation region as a central region, until the central region is connected with an adjacent central region to stop expansion, and the straight line distance between the centers of the two connected central regions is recorded to obtain an abnormal region distance; According to the first annotation region, the second annotation region and a plurality of abnormal region distances, a feature information analysis graph of the patient preliminary image is established, and feature information of the patient preliminary image is extracted according to the feature information analysis graph of the patient preliminary image. 6.The rash disease prediction method based on convolutional neural network according to claim 5, characterized in that, The step of establishing a feature information analysis graph of the patient preliminary image according to the first annotation region, the second annotation region and a plurality of abnormal region distances, and extracting feature information of the patient preliminary image according to the feature information analysis graph of the patient preliminary image, comprising: Monitoring the size change of the first annotation region and the second annotation region, and establishing a feature information analysis graph of the patient preliminary image according to the size change; Binding a plurality of abnormal region distances with the first annotation region and the second annotation region corresponding to the feature information analysis graph of the patient preliminary image; Monitoring the size change of the first annotation region and the second annotation region in the feature information analysis graph of the patient preliminary image and the change of a plurality of abnormal region distances corresponding thereto, recording the first annotation region, the second annotation region and the abnormal region distance with the largest change amount in each stage, and recording the change period and the law of the change of the first annotation region, the second annotation region and the abnormal region distance, to obtain the feature information of the patient preliminary image.

7. The rash disease prediction method based on a convolutional neural network according to claim 6, characterized in that, The step of performing rash disease identification analysis on the feature information of the patient preliminary image, obtaining a patient image analysis result, binding the patient image analysis result with the patient preliminary image, and performing a second classification process on the patient preliminary image to obtain a second patient image group, comprises: Performing rash disease identification analysis on the feature information of the patient preliminary image, and if the distance changes of the first marked region, the second marked region and the abnormal region change periodically, marking the feature information of the patient preliminary image as a first rash disease category; If the range of the first marked region and the second marked region gradually increases, the distance of the abnormal region gradually decreases, and the range of the first marked region and the second marked region and the distance of the abnormal region do not change periodically, marking the feature information of the patient preliminary image as a second rash disease category; If the first marked region, the second marked region and the abnormal region have no obvious changes, marking the feature information of the patient preliminary image as a third rash disease category; Binding the patient image analysis result with the patient preliminary image, and performing a second classification process on the patient preliminary image, classifying the patient preliminary image according to the first rash disease category, the second rash disease category and the third rash disease category according to the different markings, and obtaining a second patient image group.

8. The rash disease prediction method based on a convolutional neural network according to claim 7, characterized in that: The rash prediction tree diagram is a main point of personal information of each patient, and the first branch diffusion is performed downward from the main point as a core, the first branch diffusion being the patient preliminary image of different parts of the patient, and the second branch diffusion is extended downward according to the first branch diffusion; The second branch diffusion is the first patient image group and the second patient image group corresponding to the part, and the third branch diffusion is extended according to the second branch, the third branch diffusion being a rash disease identification analysis condition, the rash disease identification analysis condition including a current rash disease identification analysis result and a rash disease identification analysis prediction result; The current rash disease identification analysis result is the feature information of the patient preliminary image corresponding to the first rash disease category, the second rash disease category and the third rash disease category; The rash disease identification analysis prediction result is obtained by inputting the monitoring result of the feature information of the patient preliminary image into a big data prediction model. 9.The rash disease prediction method based on convolutional neural network according to claim 8, characterized in that, The step of establishing a rash feature vector according to the feature information of the patient preliminary image corresponding to the image in the second patient image group, comprises: Establishing a rash feature vector according to the feature information of the patient preliminary image corresponding to the image in the second patient image group; According to the first rash disease category, the second rash disease category and the third rash disease category to which the patient preliminary image belongs, a rash disease probability gradient is obtained, the rash disease probability of the first rash disease category is 40%, the rash disease probability of the second rash disease category is 70%, and the rash disease probability of the third rash disease category is 20%; The rash feature vector includes the corresponding rash disease probability gradient to which the patient preliminary image belongs, the first number of labeled regions and the second number of labeled regions. 10.A rash disease prediction system based on a convolutional neural network, characterized by, The system comprises: An image acquisition module: acquiring photographs of various parts of a patient, the photographs of various parts of the patient including images under different light conditions, to obtain patient preliminary images; A convolutional neural network training module: labeling the patient preliminary images to obtain a patient preliminary image dataset; and training a patient image convolutional neural network according to the patient preliminary image dataset; A feature extraction module: performing a first classification process on the patient preliminary images through the patient image convolutional neural network to obtain a first patient image group, and extracting feature information of the patient preliminary images; A rash disease recognition and analysis module: performing rash disease recognition and analysis on the feature information of the patient preliminary images to obtain a patient image analysis result, binding the patient image analysis result with the patient preliminary images, and performing a second classification process on the patient preliminary images to obtain a second patient image group; A rash disease estimation module: establishing a rash estimation tree diagram according to the second patient image group, and establishing a rash feature vector according to the feature information of the patient preliminary images corresponding to the images in the second patient image group; and performing patient rash condition estimation according to the change state of the rash estimation tree diagram and the rash feature vector, to establish a multi-dimensional rash disease estimation diagram.

Images