CN115239993A - Human body alopecia type and stage identification system based on cross-domain semi-supervised learning - Google Patents

Human body alopecia type and stage identification system based on cross-domain semi-supervised learning Download PDF

Info

Publication number
CN115239993A
CN115239993A CN202210798275.9A CN202210798275A CN115239993A CN 115239993 A CN115239993 A CN 115239993A CN 202210798275 A CN202210798275 A CN 202210798275A CN 115239993 A CN115239993 A CN 115239993A
Authority
CN
China
Prior art keywords
training
diagnosis
supervised learning
classification
classification model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210798275.9A
Other languages
Chinese (zh)
Inventor
阳行意
张传
包纪元
迟筠航
牛思慧
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Siyue Technology Co ltd
Original Assignee
Hangzhou Siyue Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Hangzhou Siyue Technology Co ltd filed Critical Hangzhou Siyue Technology Co ltd
Priority to CN202210798275.9A priority Critical patent/CN115239993A/en
Publication of CN115239993A publication Critical patent/CN115239993A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Public Health (AREA)
  • Pathology (AREA)
  • Data Mining & Analysis (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a human body alopecia type and stage identification system based on cross-domain semi-supervised learning, which comprises: the classification module is used for classifying the input images; the analysis module is used for receiving the classification result, the clinic history sequence and the disease knowledge map of the classification module and comprehensively analyzing and outputting the analysis result through the trained analysis model; the method for training the classification model comprises the following steps: collecting training image data, wherein the training image data are a plurality of images with different alopecia degrees; training the classification model by adopting a semi-supervised learning method; training the classification model by adopting a multi-source field generalization method; and training the classification model by adopting a course learning method. The human body alopecia type and stage identification system based on cross-domain semi-supervised learning can be used for identifying the alopecia stage by utilizing a small amount of marked data and a large amount of multi-source unlabeled data, and accurately and comprehensively analyzing patients by combining professional knowledge of medical skin diagnosis.

Description

Human body alopecia type and stage identification system based on cross-domain semi-supervised learning
Technical Field
The invention relates to a human body alopecia type and stage identification system based on cross-domain semi-supervised learning.
Background
The problem of hair loss is now a widespread health problem facing the urban population. The existing diagnosis method is mainly artificial alopecia diagnosis based on inquiry. The pathological diagnosis of the traditional alopecia problem strongly depends on the judgment of a professional physician, and a patient must go to a special medical institution and obtain a diagnosis and treatment suggestion through a series of complicated diagnosis and treatment processes. In addition, the traditional diagnosis method for alopecia by inquiry makes it difficult for general patients to pay the cost of professional medical care and enjoy the services. Meanwhile, a series of initial alopecia diseases cannot be discovered in the initial stage due to the complicated diagnosis and treatment process, and the disease condition is worsened due to the fact that the diagnosis and treatment are not timely performed.
Disclosure of Invention
The invention provides a human body alopecia type and stage identification system based on cross-domain semi-supervised learning, which solves the technical problems and specifically adopts the following technical scheme:
a human body alopecia type and stage identification system based on cross-domain semi-supervised learning comprises:
the classification module is used for receiving the picture to be analyzed and classifying the input image through the trained classification model;
the analysis module is used for receiving the classification result, the clinic history sequence and the disease knowledge map of the classification module and comprehensively analyzing and outputting the analysis result through the trained analysis model;
the method for training the classification model comprises the following steps:
collecting training image data, wherein the training image data are a plurality of images with different alopecia degrees;
training the classification model by adopting a semi-supervised learning method;
training the classification model by adopting a multi-source field generalization method;
and training the classification model by adopting a course learning method.
Further, the training image data contains P data sets
Figure BDA0003733006730000011
Each data set contains data having N p Labeled data set of individual samples
Figure BDA0003733006730000012
And has M p Label-free data set of individual samples
Figure BDA0003733006730000013
Wherein x i Represents the ith image, y i Representing the label corresponding to the ith image, x j Representing the jth image.
Further, the label is the hair loss condition and severity corresponding to the image.
Further, a specific method for training the classification model by adopting a semi-supervised learning method comprises the following steps:
supervised training is performed on labeled samples, for x i Forward prediction is performed to obtain the probability p = f (x) of each class i (ii) a θ) and predict value p thereof i With the true label y i Comparing, and correcting the prediction error through back propagation;
for samples without labels, in the training stage, strong random enhancement transformation and weak random transformation are carried out on the samples,
Figure BDA0003733006730000021
Figure BDA0003733006730000022
forward predicting the data after twice conversion to obtain the probability of each category
Figure BDA0003733006730000023
And
Figure BDA0003733006730000024
Figure BDA0003733006730000025
and if the prediction confidence of the weak transformation is higher than the threshold tau, taking the prediction result as a pseudo label, and correcting the strong transformation prediction result by using the pseudo label.
Further, in the process of training the classification model by adopting a semi-supervised learning method, cross entropy is used as a loss function of training.
Further, a specific method for training the classification model by adopting a multi-source domain generalization method is as follows:
performing combined training on the classification models on different data sources;
and carrying out a regularization operation on the classification model.
Further, a concrete method for training the classification model by adopting a course learning method is as follows:
and taking the confidence coefficient of the sample as a judgment basis for the difficulty of the sample, wherein the higher the confidence coefficient is, the easier the sample is, the lower the confidence coefficient is, the harder the sample is, the higher the confidence coefficient is, and the greater the weight of the sample is.
Further, the specific method for training the analysis model is as follows:
collecting the text records of the historical diagnosis of the patient, and screening out the diagnosis result, diagnosis and treatment opinions and the diagnosis time in the diagnosis history of the patient;
converting the diagnosis history into a triple sequence with the length of L
Figure BDA0003733006730000026
Wherein each element is a triplet s = (time of visit, result of visit, opinion of diagnosis and treatment);
constructing a disease knowledge graph G related to alopecia diseases, wherein each node on the graph represents one alopecia disease type, and each edge represents the mutual transfer probability between two diseases;
constructing an analysis model, wherein the analysis model comprises two graph convolution neural networks GNN hist And GNN G In which GNN hist Triple in charge of handling visit historyAccording to S, outputting a history information characteristic h hist ，GNN G Responsible for processing the disease knowledge map G and outputting the disease knowledge characteristics h G ；
Characterizing the historical information by h hist Knowledge of disease characteristics h G And classification result p of classification model i Merging the vectors to output a unified vector h i ＝concat([h hist ,h G ,p i ]) Predicting the final analysis diagnosis result through a linear layer;
and (4) training the analysis model by using the real diagnosis label, and updating the parameters by a gradient descent method to obtain the trained analysis model.
Further, the specific method for screening out the diagnosis result, diagnosis and treatment opinions and the diagnosis time in the patient's diagnosis history comprises the following steps:
constructing a medical keyword dictionary;
and matching keywords meeting the keyword requirements in the text records of the historical diagnosis through a regular expression technology.
The system has the advantages that the system for identifying the type and the stage of the alopecia of the human body based on cross-domain semi-supervised learning can identify the alopecia stage by utilizing a small amount of marked data and a large amount of multi-source non-marked data, and accurately and comprehensively analyze patients by combining professional knowledge of medical skin diagnosis.
Drawings
Fig. 1 is a schematic diagram of a human body alopecia type and stage identification system based on cross-domain semi-supervised learning according to the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and the embodiments.
Fig. 1 shows a cross-domain semi-supervised learning-based human alopecia type and stage identification system according to the present application. The method mainly comprises the following steps: a classification module and an analysis module.
The classification module is used for receiving the pictures to be analyzed and classifying the input images through the trained classification model. The analysis module is used for receiving the classification result, the treatment history sequence and the disease knowledge map of the classification module and comprehensively analyzing and outputting the analysis result through the trained analysis model.
In the application, in order to improve the classification efficiency of the classification model, a comprehensive method is adopted to train the classification model.
Specifically, the method for training the classification model comprises the following steps:
training image data is collected, and the training image data are images with different alopecia degrees.
And training the classification model by adopting a semi-supervised learning method.
And training the classification model by adopting a multi-source domain generalization method.
And training the classification model by adopting a course learning method.
As a preferred embodiment, the training image data comprises P data sets
Figure BDA0003733006730000031
Each data set contains data having N p Labeled data set of individual samples
Figure BDA0003733006730000032
And has M p Label-free data set of individual samples
Figure BDA0003733006730000033
Wherein x i Represents the ith image, y i Representing the label corresponding to the ith image, x j Representing the jth image. Wherein, the label is the hair loss condition and the severity degree corresponding to the image. It can be understood that in the application, the labeled data are truly and accurately derived from past desensitization patient diagnosis pictures of a plurality of social hair transplantation institutions and diagnosis and treatment pictures of patients with diagnosis information of a plurality of hospitals. Meanwhile, the non-annotation data is derived from hair and face data collected by an internet crawler and hair and face data derived from a public data set.
It is an object of the present application to utilize labeled and unlabeled exemplars in multiple datasetsTraining to obtain a model with better classification effect
Figure BDA0003733006730000034
Where θ is the model parameter, g is the feature extractor of the model, and t is the classifier.
As a preferred embodiment, the specific method for training the classification model by using the semi-supervised learning method is as follows:
supervised training is performed on labeled samples, and the model firstly performs on x i Forward prediction is carried out to obtain the probability p of each category i ＝f(x i (ii) a θ) and predict the value p i With the true label y i And comparing, and correcting the prediction error through back propagation.
During the training process, cross entropy (cross) is used as a loss function of the training,
L sup (p i ,y i )＝-E y [logp i ]
the gradient value of the loss for each parameter can be found by a gradient descent algorithm. The parameters are then modified by gradients, so that the loss function is minimized,
Figure BDA0003733006730000041
η in the equation represents the learning rate of the parameter update.
For a sample without a label, in a training stage, two data transformations are required to be performed on the sample, including a strong stochastic enhancement transformation and a weak stochastic transformation, which are respectively expressed as follows,
Figure BDA0003733006730000042
Figure BDA0003733006730000043
for the data after two conversions respectivelyForward prediction is performed to obtain the probability of each class
Figure BDA0003733006730000044
And
Figure BDA0003733006730000045
Figure BDA0003733006730000046
if the prediction confidence of the weak transform is higher than the threshold τ, the prediction result is taken as a pseudo label, and the strong transform prediction result is corrected using the pseudo label, as follows,
Figure BDA0003733006730000047
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003733006730000048
is a threshold function, i.e. when
Figure BDA0003733006730000049
Output 1 is output, otherwise output 0. The process mainly uses a prediction value with higher confidence coefficient as a pseudo label to train a classification model, and meanwhile, the prediction consistency after different transformations is ensured. At this point, we again use the loss function described above to perform a gradient descent, and then modify the parameters by gradient, so that the loss function is minimized,
Figure BDA00037330067300000410
and after each iteration is finished, updating the classification model by using a weighted average mode. Given a weighted average parameter y, the parameters after the gradient descent and the parameters of the previous step are used for weighted summation, as shown below,
θ new ＝γ θ +(1-γ)θ new
through the processing, the generalization capability of the classification model is improved.
As a preferred embodiment, a specific method for training the classification model by using a multi-source domain generalization method is as follows:
and training the classification models on different data sources in a combined manner. This process is equivalent to the supervised training process described above, except that joint training on multiple different data is required.
In particular, given K different data sources, tagged data is obtained
Figure BDA00037330067300000411
Supervised training of the model is performed using the joint dataset. In each training turn, K samples are sampled from the joint data set and input into the model to obtain a prediction probability
Figure BDA00037330067300000412
Wherein one sample is sampled per data set
Figure BDA00037330067300000413
During the training process, the model is updated again using the sum of the cross entropies (Crossentropy) of the multiple samples as a loss function of the training,
Figure BDA0003733006730000051
the gradient value of the loss for each parameter can be found by a gradient descent algorithm. The parameters are then modified by gradients, such that the loss function is minimized,
Figure BDA0003733006730000052
η in the equation represents the learning rate of the parameter update.
And carrying out regularization operation on the classification model. The specific method for carrying out regularization operation on the classification model comprises the following steps:
and for the same type data of different sources, ensuring that the feature spaces of the data belong to the same distribution. For the sample data, there is a certain difference between the data from different data sources. It is desirable to learn a uniform feature space to reduce the data gap caused by different data sources.
Figure BDA0003733006730000053
Figure BDA0003733006730000054
Two homogeneous data x for different sources i And x j The L2 distance of its feature space is reduced.
As a preferred embodiment, the specific method for training the classification model by using the curriculum learning method is as follows:
and taking the confidence coefficient of the sample as a judgment basis for the difficulty of the sample, wherein the higher the confidence coefficient is, the easier the sample is, the lower the confidence coefficient is, the harder the sample is, the higher the confidence coefficient is, and the training weight of the sample is larger. The process uses the same supervised learning approach. Using the prediction probability p i Is weighted for each sample, the loss function is as follows,
Figure BDA0003733006730000055
such a loss function design can ensure that simpler samples can be learned first. Therefore, samples are scored and learned easily and difficultly, and low-quality data are screened out, so that the model cannot be excessively concerned with noise samples or low-quality samples.
Based on a given image prediction, it is desirable to combine historical visit data for a patient with professional dermatologic medical knowledge to make a comprehensive diagnosis of the patient. In the process, an analysis model f' (p) needs to be obtained through training i S, G; Π) it predicts the probability p of the image i Diagnosis History sequence S and alopecia medical knowledge map G as inputAnd outputting a current comprehensive analysis and diagnosis result. Π being the parameters of the model. The alopecia diagnosis problem is converted into a conditional probabilistic inference problem.
The analytical model was obtained as follows:
firstly, collecting the text records of the historical diagnosis of the patient, and firstly screening out the diagnosis result, diagnosis and treatment opinions and the diagnosis time in the diagnosis history of the patient by a keyword detection technology. It is understood that there is a strong correlation between different alopecia diseases. Alopecia exists in certain stages, and the degree is changed from mild to severe. Different types of hair loss may also be interconvertible, for example, it may be possible to switch from M-type hair loss to generalized hair loss.
In the process, firstly, a medical keyword dictionary is constructed by communicating with doctors, and then keywords meeting the keyword requirements are found in medical texts through a regular expression technology. The diagnosis result refers to a character string for typing judgment of the current disease condition, such as severe/normal/mild/no alopecia and the like. The diagnosis and treatment suggestion refers to a character string of a treatment means proposed for the current disease condition, such as taking a certain medicine/performing a certain operation treatment/strengthening exercise. The visit time refers to the date of the day when the diagnosis and treatment judgment is made, such as a certain day of a certain month in a certain year.
Then, the diagnosis history is converted into a triple sequence with the length L
Figure BDA0003733006730000061
I.e. a history sequence of treatment, wherein each element is a triplet s = (time of treatment, result of treatment, opinion). The sequence summarizes the patient's historical situation.
Then, with the help of expert doctors, a disease knowledge graph G related to the alopecia disease is constructed, wherein each node on the graph represents one alopecia disease type, and each edge represents the mutual transition probability between two diseases. The map is constructed by a medical professional.
The analytical model comprises two graph convolutional neural networks GNN hist And GNN G . Wherein GNN hist Responsible for processing the three-tuple data S of the visit history and outputting a historyInformation characteristic h hist 。GNN G Responsible for processing the disease knowledge map G and outputting the disease knowledge characteristics h G
h hist ＝GNN hist (S),h G ＝GNN G (G)
We characterize the historical information by h hist Knowledge of disease characteristics h G And the classification result of the classification model, i.e. the output probability p i Merging the vectors to output a unified vector h i ＝concat([h hist ,h G ,p i ]). The final analysis diagnosis result is predicted through a linear layer,
Figure BDA0003733006730000062
and finally, training the analysis model by using a real diagnosis label, and updating the parameters by a gradient descent method to obtain the trained analysis model.
Figure BDA0003733006730000063
Figure BDA0003733006730000064
The foregoing shows and describes the general principles, principal features and advantages of the invention. It should be understood by those skilled in the art that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by using equivalents or equivalent changes fall within the protection scope of the present invention.

Claims (9)

1. A human body alopecia type and stage identification system based on cross-domain semi-supervised learning is characterized by comprising:
the classification module is used for receiving the pictures to be analyzed and classifying the input images through the trained classification model;
the analysis module is used for receiving the classification result, the clinic history sequence and the disease knowledge map of the classification module and comprehensively analyzing and outputting the analysis result through the trained analysis model;
the method for training the classification model comprises the following steps:
collecting training image data, wherein the training image data are images with different alopecia degrees;
training the classification model by adopting a semi-supervised learning method;
training the classification model by adopting a multi-source domain generalization method;
and training the classification model by adopting a course learning method.
2. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 1,
the training image data comprises P data sets
Figure FDA0003733006720000011
Each data set contains data having N p Labeled data set of individual samples
Figure FDA0003733006720000012
And has M p Unmarked dataset of individual samples
Figure FDA0003733006720000013
Wherein x i Represents the ith image, y i Representing the label corresponding to the ith image, x j Representing the jth image.
3. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 2,
the label corresponding to the image is the hair loss condition and severity corresponding to the image.
4. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 2,
the specific method for training the classification model by adopting the semi-supervised learning method comprises the following steps:
supervised training is performed on labeled samples, for x i Forward prediction is carried out to obtain the probability p of each category i ＝f(x i (ii) a θ) and predict the value p i With the true label y i Comparing, and correcting the prediction error through back propagation;
for samples without labels, in the training stage, strong random enhancement transformation and weak random transformation are carried out on the samples,
Figure FDA0003733006720000014
Figure FDA0003733006720000015
forward predicting the data after twice conversion to obtain the probability of each category
Figure FDA0003733006720000016
And
Figure FDA0003733006720000017
and if the prediction confidence of the weak transformation is higher than the threshold tau, taking the prediction result as a pseudo label, and correcting the strong transformation prediction result by using the pseudo label.
5. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 4,
in the process of training the classification model by adopting a semi-supervised learning method, cross entropy is used as a loss function of training.
6. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 2,
the specific method for training the classification model by adopting the multi-source field generalization method is as follows:
jointly training the classification models on different data sources;
and carrying out a regularization operation on the classification model.
7. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 2,
the specific method for training the classification model by adopting the course learning method comprises the following steps:
and taking the confidence coefficient of the sample as a difficult judgment basis of the sample, wherein the higher the confidence coefficient is, the easier the sample is, the lower the confidence coefficient is, the harder the sample is, the higher the confidence coefficient is, and the greater the weight of the sample is.
8. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 2,
the specific method for training the analysis model comprises the following steps:
collecting text records of historical diagnosis of a patient, and screening out diagnosis results, diagnosis and treatment opinions and diagnosis time in the diagnosis history of the patient;
converting the diagnosis history into a triple sequence with the length of L
Figure FDA0003733006720000021
Wherein each element is a triplet s = (time of visit, result of visit, opinion of diagnosis and treatment);
constructing a disease knowledge graph G related to alopecia diseases, wherein each node on the graph represents one alopecia disease type, and each edge represents the mutual transfer probability between two diseases;
constructing the analysis model, wherein the analysis model comprises two graphsConvolutional neural network GNN hist And GNN G In which GNN hist Is responsible for processing the three-tuple data S of the history of the treatment and outputting a history information characteristic h hist ，GNN G Responsible for processing the disease knowledge map G and outputting the disease knowledge characteristics h G ；
Characterizing the historical information by h hist Knowledge of disease characteristics h G And the classification result p of the classification model i Merging the vectors to output a unified vector h i ＝concat([h hist ,h G ,p i ]) Predicting the final analysis diagnosis result through a linear layer;
and training the analysis model by using a real diagnosis label, and updating parameters by a gradient descent method to obtain the trained analysis model.
9. The system for identifying types and stages of human hair loss based on cross-domain semi-supervised learning of claim 8,
the specific method for screening out the diagnosis result, diagnosis and treatment opinions and the diagnosis time in the patient diagnosis history comprises the following steps:
constructing a medical keyword dictionary;
and matching keywords meeting the keyword requirements in the text records of the historical diagnosis through a regular expression technology.
CN202210798275.9A 2022-07-06 2022-07-06 Human body alopecia type and stage identification system based on cross-domain semi-supervised learning Pending CN115239993A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210798275.9A CN115239993A (en) 2022-07-06 2022-07-06 Human body alopecia type and stage identification system based on cross-domain semi-supervised learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210798275.9A CN115239993A (en) 2022-07-06 2022-07-06 Human body alopecia type and stage identification system based on cross-domain semi-supervised learning

Publications (1)

Publication Number Publication Date
CN115239993A true CN115239993A (en) 2022-10-25

Family

ID=83671449

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210798275.9A Pending CN115239993A (en) 2022-07-06 2022-07-06 Human body alopecia type and stage identification system based on cross-domain semi-supervised learning

Country Status (1)

Country Link
CN (1) CN115239993A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115631386A (en) * 2022-12-19 2023-01-20 天津医之本医疗科技有限公司 Pathological image classification method and system based on machine learning
CN116386857A (en) * 2023-06-07 2023-07-04 深圳市森盈智能科技有限公司 Pathological analysis system and method

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115631386A (en) * 2022-12-19 2023-01-20 天津医之本医疗科技有限公司 Pathological image classification method and system based on machine learning
CN116386857A (en) * 2023-06-07 2023-07-04 深圳市森盈智能科技有限公司 Pathological analysis system and method
CN116386857B (en) * 2023-06-07 2023-11-10 深圳市森盈智能科技有限公司 Pathological analysis system and method

Similar Documents

Publication Publication Date Title
CN109949936B (en) Re-hospitalization risk prediction method based on deep learning mixed model
Verma et al. Temporal deep learning architecture for prediction of COVID-19 cases in India
CN115239993A (en) Human body alopecia type and stage identification system based on cross-domain semi-supervised learning
CN113241135A (en) Disease risk prediction method and system based on multi-mode fusion
CN112766355A (en) Electroencephalogram signal emotion recognition method under label noise
CN112529063B (en) Depth domain adaptive classification method suitable for Parkinson voice data set
CN112766376A (en) Multi-label eye fundus image identification method based on GACNN
CN114743037A (en) Deep medical image clustering method based on multi-scale structure learning
CN115841607A (en) Brain network structure and similarity joint learning method based on graph attention network
Zhao et al. Deeply supervised active learning for finger bones segmentation
Shirazi et al. Deep learning in the healthcare industry: theory and applications
CN115083599A (en) Knowledge graph-based preliminary diagnosis and treatment method for disease state
Thangavel et al. EAD-DNN: Early Alzheimer's disease prediction using deep neural networks
Cabezas et al. Hierarchical clustering: Visualization, feature importance and model selection
Sirigineedi et al. The Early Detection Of Alzheimer's Illness Using Machine Learning And Deep Learning Algorithms
CN114417836A (en) Deep learning-based Chinese electronic medical record text semantic segmentation method
Shahin et al. A robust deep neural network framework for the detection of diabetes
Li et al. Bayesian nested latent class models for cause-of-death assignment using verbal autopsies across multiple domains
Sun et al. TSRNet: Diagnosis of COVID-19 based on self-supervised learning and hybrid ensemble model
Zhang et al. Exploring unsupervised multivariate time series representation learning for chronic disease diagnosis
JP2024061599A (en) A system for identifying abnormalities in the course of medical treatment based on a hierarchical neural network
CN116797817A (en) Autism disease prediction technology based on self-supervision graph convolution model
Qiu et al. Predictive modeling of the total joint replacement surgery risk: a deep learning based approach with claims data
Ma et al. Semi-parametric Bayes regression with network-valued covariates
CN116072298A (en) Disease prediction system based on hierarchical marker distribution learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB03 Change of inventor or designer information
CB03 Change of inventor or designer information

Inventor after: Yang Xingyi

Inventor after: Zhang Chuan

Inventor after: Bao Jiyuan

Inventor before: Yang Xingyi

Inventor before: Zhang Chuan

Inventor before: Bao Jiyuan

Inventor before: Chi Junhang

Inventor before: Niu Sihui