CN116364292A - Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease - Google Patents

Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease Download PDF

Info

Publication number
CN116364292A
CN116364292A CN202310335253.3A CN202310335253A CN116364292A CN 116364292 A CN116364292 A CN 116364292A CN 202310335253 A CN202310335253 A CN 202310335253A CN 116364292 A CN116364292 A CN 116364292A
Authority
CN
China
Prior art keywords
data
detection
thyroid
feature vector
predicted
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
CN202310335253.3A
Other languages
Chinese (zh)
Inventor
周慧芳
宋雪霏
雷超宇
翟广涛
卞睿彤
谈子铭
孙瀚池
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.)
Ninth Peoples Hospital Shanghai Jiaotong University School of Medicine
Original Assignee
Ninth Peoples Hospital Shanghai Jiaotong University School of Medicine
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 Ninth Peoples Hospital Shanghai Jiaotong University School of Medicine filed Critical Ninth Peoples Hospital Shanghai Jiaotong University School of Medicine
Priority to CN202310335253.3A priority Critical patent/CN116364292A/en
Publication of CN116364292A publication Critical patent/CN116364292A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • 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
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
    • 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
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • 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
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Public Health (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Epidemiology (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Primary Health Care (AREA)
  • Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method for predicting prognosis of thyroid eye disease, which comprises the following steps: acquiring historical medical data and facial image data; preprocessing detection data, and performing feature embedding on the detection data with different data types to generate an embedded feature vector; generating time sequence data to be predicted by the embedded feature vector set obtained by a plurality of detection time points of the same object to be predicted; and constructing a prognosis prediction model based on a transducer, inputting the time series data to be predicted into the trained prognosis prediction model, and predicting thyroid function detection indexes in future detection. A large number of facial images and thyroid function detection data of TED patients are collected, abnormal conditions of thyroid functions are predicted noninvasively, the aim of predicting key function indexes of the patients is achieved by means of artificial intelligent neural network training, doctors and patients are assisted in assessing and monitoring prognosis eye health conditions, the aim of screening thyroid-related eye diseases is achieved, and a large amount of cost is saved.

Description

Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease
Technical Field
The invention relates to the technical field of information prediction, in particular to a method, a device, equipment and a storage medium for predicting prognosis of thyroid eye disease.
Background
Thyroid eye disease (thyroid eye disease, TED) is a blinding, teratogenic, disabling organ-specific autoimmune disease that is closely related to abnormal thyroid function, the main clinical features include abnormal eye position, congestion of the eyelid, etc., and most patients are accompanied by abnormal thyroid function. Related studies have shown that about 20% of TED patients ' eye disease occurs before thyroid dysfunction, about 40% of patients ' eye disease occurs simultaneously with thyroid dysfunction, about 40% of patients ' eye disease occurs after thyroid dysfunction, and thyroid dysfunction may be one of the risk factors for TED disease progression. Therefore, thyroid function related detection of TED patients is of great significance.
Traditional thyroid function detection is to perform blood drawing assay, and mainly detects TT3 (total triiodothyronine), TT4 (total thyroxine), FT3 (free triiodothyronine), FT4 (free thyroxine), TSH (thyroid stimulating hormone), thyroglobulin antibody and thyroperoxidase antibody. Although convenient, such methods still fall within the category of invasive methods. The multiple detection is easy to cause waste of time cost and economic cost for patients, waste of medical institution resources and no noninvasive detection method exists at present.
Disclosure of Invention
The invention aims to provide a prognosis prediction method, device, equipment and storage medium for thyroid eye diseases, which are used for collecting a large number of facial images and thyroid function detection data of TED patients, noninvasively predicting abnormal conditions of thyroid functions, achieving the purpose of predicting key function indexes by means of artificial intelligent neural network training, further assisting doctors and patients in evaluating and monitoring prognosis eye health conditions, achieving the purpose of screening thyroid-related eye diseases, and saving a large number of manpower and material resources.
The invention provides a method for predicting prognosis of thyroid eye diseases, which comprises the following steps:
acquiring historical medical data and facial image data, and generating detection data of an object to be predicted, wherein the historical medical data comprises thyroid function detection index data and basic information of the object to be predicted;
preprocessing the detection data, classifying and processing a plurality of detection data according to data types, performing feature embedding on the detection data of different data types of the same detection time point of the object to be predicted to generate an embedded feature vector, and encoding the embedded feature vector to obtain an encoded embedded feature vector, wherein the embedded feature vector comprises an image feature vector, a category feature vector and a numerical feature vector related to the object to be predicted;
Generating time sequence data to be predicted by using a plurality of encoded embedded feature vector sets obtained from a plurality of detection time points of the same object to be predicted;
and constructing a prognosis prediction model based on a transducer, inputting the time series data to be predicted into the trained prognosis prediction model, predicting thyroid function detection indexes in future detection, and outputting a prediction result.
Preferably, the historical medical data comprises basic information and thyroid function detection index data, wherein the thyroid function detection index data comprises free triiodothyronine, free tetraiodothyronine, total triiodothyronine, total tetraiodothyronine, thyroid stimulating hormone, thyroglobulin antibody and thyroid peroxidase antibody and historical test results corresponding to the indexes.
Preferably, the preprocessing the detection data, classifying and processing a plurality of detection data according to data types includes:
extracting facial image data with thyroid eye disease signs at different detection time points in the detection data;
determining pixel values of a target face image with pixel values in the face image based on a bilinear interpolation algorithm to scale the target face image to obtain the resized target face image;
And inputting the target face image with the adjusted size into a VGG-19 pre-training model to obtain an image feature vector corresponding to the target face image with the adjusted size.
Preferably, the preprocessing the detection data, classifying and processing a plurality of detection data according to data types includes:
extracting different types of category features in the detection data, wherein the category features at least comprise gender, thyroid disease history and eye symptoms;
converting the sex characteristic, the thyroid history characteristic and the ocular symptom characteristic into a first unique heat encoding characteristic, a second unique heat encoding characteristic and a third unique heat encoding characteristic, respectively;
performing fusion processing on the first single-hot coding feature, the second single-hot coding feature and the third single-hot coding feature to obtain a fusion feature vector as input of a current network input layer;
and multiplying the fusion feature vector by a preset weight matrix, and reducing the dimension to obtain a corresponding category feature vector output by the current network ebedding layer.
Preferably, the preprocessing the detection data, classifying and processing a plurality of detection data according to data types includes:
Extracting numerical characteristics in the detection data, wherein the numerical characteristics comprise thyroid function detection indexes, ages and smoking indexes;
and carrying out normalization processing on the numerical feature, converting the numerical feature into a preset range through linearization, and outputting a corresponding numerical feature vector.
Preferably, the constructing the prognosis prediction model based on the transducer includes:
configuring a main architecture into a transducer neural network comprising an encoder and a decoder;
inputting the time series data to be predicted into an encoder, outputting the embedded feature vector after encoding, using the encoded embedded feature vector as an input vector of a multi-layer sensor, mapping a current input vector to an output vector, and adopting softmax function calculation to predict each thyroid function detection index result in future examination;
inversely normalizing the predicted thyroid function detection index results in future examination to obtain corresponding dimensionalization prediction results;
during training, the embedded feature vector obtained by calculating a softmax function and an actual thyroid function detection index are input into the prognosis prediction model, a cross entropy loss function is calculated and minimized, and therefore the prognosis prediction model is trained;
If the future multiple detection results are predicted, the current prediction result is used as a new embedded feature vector to be imported into the trained prognosis prediction model, and the prediction result in the next detection is output.
Preferably, the method for training and evaluating the prognosis prediction model comprises the following steps:
dividing the time sequence data to be predicted into a training set, a verification set and a test set according to a preset proportion;
constructing the prognosis prediction model based on a Pytorch deep learning framework, inputting the training set and the verification set into the prognosis prediction model for training, and selecting a loss function of a current model as a minimum cross entropy loss function to obtain the trained prognosis prediction model;
optimizing and updating parameters of the trained prognosis prediction model based on back propagation of a gradient descent method to obtain the optimized prognosis prediction model;
and evaluating the performance of the optimized prognosis prediction model by adopting cross validation, and determining the prediction performance of the optimized prognosis prediction model in a data set formed by the time series data to be predicted by adopting the test set validation.
The invention provides a prognosis prediction device for thyroid eye diseases, which comprises the following components:
The data acquisition module is used for acquiring historical medical data and facial image data and generating detection data of an object to be predicted, wherein the historical medical data comprises thyroid function detection index data and basic information of the object to be predicted;
the data preprocessing module is used for preprocessing the detection data, classifying and processing a plurality of detection data according to data types, performing feature embedding on the detection data of different data types of the same detection time point of the object to be predicted to generate an embedded feature vector, and encoding the embedded feature vector to obtain an encoded embedded feature vector, wherein the embedded feature vector comprises an image feature vector, a category feature vector and a numerical feature vector related to the object to be predicted; generating time sequence data to be predicted by using a plurality of encoded embedded feature vector sets obtained from a plurality of detection time points of the same object to be predicted;
and the prognosis prediction module is used for constructing a prognosis prediction model based on a transducer, inputting the time series data to be predicted into the prognosis prediction model after training, predicting thyroid function detection indexes during future detection and outputting a prediction result.
The invention provides an electronic device, which is characterized by comprising:
a memory for storing a processing program;
and the processor is used for realizing the method for predicting the thyroid eye disease in advance when executing the processing program.
The invention provides a readable storage medium, which is characterized in that a processing program is stored on the readable storage medium, and the processing program realizes the method for predicting the thyroid eye disease according to the embodiment of the invention when being executed by a processor.
Aiming at the prior art, the invention has the following beneficial effects:
the invention provides a method for predicting the prognosis of thyroid eye disease, which predicts the progress of thyroid eye disease based on facial images and historical medical data, and carries out relevant processing and feature embedding on each data; predicting specific numerical values of seven thyroid function indexes by using a transducer-based prediction model, and outputting a result; training and evaluating the effect of the model; through neural network deep learning, facial images and historical medical data acquired at different time points are utilized, noninvasive thyroid disease detection is realized by means of a time sequence model, relevant function index values are obtained, the follow-up disease progress situation is predicted, medical resources can be effectively saved, pain of a patient is relieved, prognosis situation is improved, the purpose of screening the eye diseases related to the thyroid is finally achieved, a user can conveniently evaluate the eye health situation of the user, a large amount of economic cost and time cost are saved, meanwhile, the working efficiency of a medical institution on screening the eye diseases related to the thyroid can be effectively improved, the working cost is reduced, and further popularization and application of an efficient, convenient and low-cost eye orbit disease screening method in China are promoted.
The invention focuses on the strong correlation between the TED patient and the thyroid gland function abnormality, acquires a large number of images of the external eyes of the human face of the TED patient and thyroid gland function detection data, and achieves the aim of predicting the key hormone index level by means of artificial intelligent neural network training, thereby judging whether the thyroid gland function is abnormal or not, further assisting doctors in carrying out the disease control of the TED patient and improving the prognosis condition of the TED patient.
Drawings
FIG. 1 is a schematic diagram showing the steps of a method for prognosis prediction of thyroid eye disease according to an embodiment of the present invention;
FIG. 2 is a schematic overall flow chart of a method for predicting thyroid eye disease according to an embodiment of the present invention;
FIG. 3 is a diagram of a network architecture for feature vector fusion in accordance with one embodiment of the present invention;
FIG. 4 is a schematic diagram of an encoder in a transducer according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a device for prognosis prediction of thyroid eye disease according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The term "including" and variations thereof as used herein are intended to be open-ended, i.e., including, but not limited to. The term "based on" is based at least in part on. The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments. Related definitions of other terms will be given in the description below.
It should be noted that the terms "once," "second," and the like, as referred to in this disclosure, are used solely to distinguish one from another device, module, or unit, and are not intended to limit the order or interdependence of functions performed by such devices, modules, or units.
It should be noted that the references to "one" or "a plurality" in this disclosure are intended to be illustrative rather than limiting, and those of ordinary skill in the art will appreciate that "one or more" is intended to be construed as "one or more" unless the context clearly indicates otherwise.
Example 1
As shown in fig. 1 and 2, the present invention provides a method for predicting thyroid eye disease, comprising:
step S1, acquiring historical medical data and facial image data, and generating detection data of an object to be predicted, wherein the historical medical data comprises thyroid function detection index data and basic information related to the object to be predicted;
When the method is implemented, firstly, data are required to be collected, and facial images of TED patients, thyroid function detection index data and basic information related to the object to be predicted are collected. It can be understood that for each patient, its historical medical data is collected:
A. basic information related to the object to be predicted includes:
gender, age, smoking index, e.g. number of daily smoking years, is a number, e.g. 40 (years), with/without corresponding eye symptoms including protrusion of the eyeball, enlargement of the eye, reduced vision, eye pain, dysoculopathy, strabismus, double vision/visual ghost, lacrimation, eyelid swelling, photophobia or other symptoms, history of thyroid disorders including history of abnormal nail function (with/without), history of abnormal thyroid related antibody index (with/without), type of abnormal nail function (hyperthyroidism/hypothyroidism), I131 treatment history, i.e. whether I131 treatment was performed (yes/no).
B. Seven indicators of thyroid function detection are detected and recorded separately each time an examination is made. The historical medical data comprises basic information and thyroid function detection index data, wherein the thyroid function detection index data comprises free triiodothyronine, free tetraiodothyronine, total triiodothyronine, total tetraiodothyronine, thyroid stimulating hormone, thyroglobulin antibody, thyroperoxidase antibody and historical test results corresponding to the indexes.
Marking seven indexes for thyroid function detection according to the current clinical standard for subsequent model training. The reference table for the seven indexes of thyroid function detection is as follows:
Figure BDA0004156211120000081
Figure BDA0004156211120000091
C. this check is made as to whether or not a treatment prescription is being made, such as no prescription, intravenous hormone (4.5 g,12w usage), intravenous hormone (other usage), MTX, radiation therapy, oral hormone, orbital decompression, strabismus correction, eyelid retraction correction, other prescriptions, intravenous hormone (4.5 g,1m usage), eye drops, follow-up observations.
D. For each examination, the patient showed that the examination results at this time: thyroid function detects seven index values.
In this embodiment, facial image data is also collected for each patient:
E. each time an examination is made, a picture of the patient's face is taken and sequence data of its face picture is acquired.
Further, under the condition that the target data is missing, the detection data of the object to be predicted is supplemented with missing values or special marks are filled to shield corresponding vectors through interpolation, and uncertainty or other influences caused by the missing values are avoided.
It will be appreciated by those skilled in the art that the present invention is practiced assuming that sequential data of seven indices and facial photographs are collected for thyroid function of 10 examinations per patient over a period of time, and that the interval between each examination is substantially uniform. Or in actual situations, the problems that the examination times of different patients are different, the time between the two examinations is different, the data part of each examination of the patient is missing, and the like can be solved by a method of interpolation to supplement missing values or filling special mark shielding corresponding vectors when the method is applied. I.e. the corresponding dimensions are filled in by filling in 0's or interpolation methods.
Step S2, preprocessing the detection data, classifying and processing a plurality of detection data according to data types, performing feature embedding on the detection data of different data types of the same detection time point of the object to be predicted to generate an embedded feature vector, and encoding the embedded feature vector to obtain an encoded embedded feature vector, wherein the embedded feature vector comprises an image feature vector, a category feature vector and a numerical feature vector related to the object to be predicted;
for the collected historical medical data and facial image data, different processing modes are required according to different data types. After feature embedding, all data information of the patient at each time point is encoded into one vector for subsequent training of the model. Specifically, the collected data are divided into numerical data, including seven indexes of thyroid function detection, age and smoking index; category data including gender, ocular symptoms, thyroid history, I131 treatment history, treatment prescriptions; facial image data, i.e., a photograph of the patient's face. For each data category, the processing is as follows:
Specifically, the preprocessing the detection data in step S2, classifying and processing a plurality of detection data according to a data type includes:
step S210, extracting numerical characteristics in the detection data, wherein the numerical characteristics comprise thyroid function detection indexes, ages and smoking indexes;
step S211, performing normalization processing on the numerical feature, converting the numerical feature into a preset range through linearization, and outputting a corresponding numerical feature vector. Normalization (normalization) of data is to scale the data to fall within a small specified interval. The unit limitation of the data is removed, the data is converted into dimensionless pure numerical values, indexes of different units or orders can be compared and weighted conveniently, and the convergence of a training network is quickened. The most typical is the normalization of the data, i.e., the unified mapping of the data onto the [0,1] interval.
When the detected data is numerical data, normalization/standardization processing is performed to remove the dimension of the original data, to be a dimensionless expression, and the data is mapped to a range of 0 to 1. Therefore, model precision can be improved in subsequent training, and convergence speed can be improved. The original data is converted to the range of [0,1] by linearization using the method of linear function normalization (Min-Max scaling), the formula is as follows:
Figure BDA0004156211120000111
The method realizes the equal-proportion scaling of the original numerical data, wherein X_norm is normalized data, X is the original data, and X_max and X_min are the maximum value and the minimum value of the original data set respectively. The linear function normalization can preserve 0 in sparse features and can solve the data to the point where the variance of the features is small.
In the embodiment of the invention, for an application scene, for example, the numerical data is 11.18mg/L, the maximum value of all the data in the index is 12mg/L, the minimum value of all the data is 11mg/L, and the dimension is converted into 0.18 after the normalization operation; other numerical data are converted into 0.56,0.78, … through the normalization operation, and finally are spliced into numerical feature vectors such as [0.18,0.56,0.78, … ].
Specifically, the preprocessing the detection data in step S2, classifying and processing a plurality of detection data according to a data type includes:
step S220, extracting different types of category features in the detection data, wherein the category features at least comprise gender, thyroid disease history and eye symptoms;
step S221, performing single-heat coding on the sex characteristic, the thyroid disease history characteristic and the eye symptom characteristic to convert the sex characteristic, the thyroid disease history characteristic and the eye symptom characteristic into a first single-heat coding characteristic, a second single-heat coding characteristic and a third single-heat coding characteristic;
Step S222, carrying out fusion processing on the first single-hot coding feature, the second single-hot coding feature and the third single-hot coding feature to obtain a fusion feature vector as input of a current network input layer;
and S223, multiplying the fusion feature vector by a preset weight matrix, and reducing the dimension to obtain a corresponding category feature vector output by the current network ebedding layer.
When the detection data are category type data, considering that the number of categories is small, after each category feature is directly subjected to one-hot coding (one-hot), all the one-hot coding features are connected end to form an input vector, and then a weight matrix is multiplied to obtain an output. The reason that the weight matrix is needed to be multiplied is that the dimension of the feature after being connected end to end is higher, and dimension reduction is carried out by multiplying the weight matrix. Taking the three categories of sex, eye symptoms and thyroid history as examples, they need to go through the following steps:
1) The data of each type are separately monothermally encoded, i.e. the category characteristics include gender (male, female) for male expressed as [1,0], ocular symptoms such as protrusion of the eye, enlargement of the eye, vision loss, eye pain, eye movement disorder, strabismus, double vision/object ghost, lacrimation, eyelid swelling, photophobia or other symptoms, partial thyroid disorder external characteristic lesions concentrated on the change of the morphology of the orbitofacial surface, such as hyperthyroidism leading to agitation, dysphoria, insomnia, hyperhidrosis, emaciation, protruding eyes, etc. of the patient, hypothyroidism leading to slow expression, pale complexion, edema of the face and/or eyelid, dry skin, roughness, thick lips, large tongue, etc. of the patient. There were vision loss and strabismus expressed as [0,0,1,0,0,1,0,0,0,0,0], thyroid disease history (history of nail work abnormality, history of thyroid-related antibody index abnormality, type of nail work abnormality (hyperthyroidism/hypothyroidism)), active history of nail work abnormality, type of nail work abnormality expressed as nail hypothyroidism [1,0,1].
2) All the above-mentioned single-heat coding features are connected end to form new input vector x= [1,0,0,0,1,0,0,1,0,0,0,0,0,1,0,1].
3) And performing matrix multiplication operation on the input vector x and the weights to obtain the output of the current network layer.
Referring to fig. 3, the emmbedding layer is equivalent to first taking the form of the emmbedding vectors from the three input class features, respectively, and then summing the emmbedding vectors one by one. Where the weight matrix w16×k is the unbedding feature, and k is the dimension of the unbedding vector, which can be set arbitrarily. W16 x k where the first two lines correspond to the emmbedding signature of gender, lines 3 to 13 correspond to the emmbedding signature of the presence or absence of 11 ocular symptoms, and lines 14 to 16 correspond to the emmbedding signature of the thyroid history.
Specifically, the preprocessing the detection data in step S2, classifying and processing a plurality of detection data according to a data type includes:
step S230, extracting face source image data with thyroid eye disease signs at different detection time points in the detection data;
step S231, determining pixel values of a target face image by pixel values in the source image based on a bilinear interpolation algorithm so as to scale the target face image to obtain the resized target face image;
Step S232, inputting the target facial image with the adjusted size into a VGG-19 pre-training model to obtain an image feature vector corresponding to the target facial image with the adjusted size.
The face source image data collected in this embodiment is pixel data obtained from a face photo, each pixel has three specific RGB values, a link is established between the feature of the face photo and thyroid eye disease, and a Convolutional Neural Network (CNN) is input to obtain a feature vector. Specifically, a VGG19 pre-training model is adopted, all collected face photos of a patient are adjusted to be three-channel color pictures with 224 x 224 size through a bilinear interpolation method, and then the three-channel color pictures are input into the VGG19 pre-training model to obtain 1000-dimensional image feature vectors. The background of the face photo processed by the bilinear interpolation algorithm is based on the scaling of the picture, in the process of scaling the picture, the pixel value of the original image pixel matrix is essentially filled into the target image pixel matrix, and the target image pixel matrix can be larger than the original image pixel matrix or enlarged or smaller than the original image pixel matrix or reduced.
Finally, all the obtained numerical feature vectors, category feature vectors and image features are embedded into one vector. Specifically, all the feature vectors obtained as described above are sequentially arranged as inputs to a multi-layer perceptron (MLP), and thus can always be set to a fixed dimension, and set to 64 dimensions. The multi-layer perceptron is designed into two hidden layers, sigmoid is used as an activation function, the input is the sum of all feature vector dimensions, and the output is 64 dimensions.
Step S3, generating time sequence data to be predicted from a plurality of encoded embedded feature vector sets obtained from a plurality of detection time points of the same object to be predicted; the predicted time series data, namely the characteristic sequence of the detection data of the patient, comprises detection information of the patient at each detection time point, including facial images, basic information, thyroid function seven indexes and the like.
For each time of the patient, i.e. the detection data of each time point, feature embedding is carried out to obtain 64-dimension vector R i The method comprises the steps of carrying out a first treatment on the surface of the Then for each patient there are n detected data corresponding to the time series to be predicted U j ＝(R 1 ，R 2 ，…，R n )。
Obtaining a time sequence U to be predicted of each patient j And dividing the preprocessed data into a training set, a verification set and a test set. Suppose we will test all data R for the last time in all patient index sequences n As a test set, the next to last data R in the sequence n-1 As a validation set, the rest as a training set.
And S4, constructing a prognosis prediction model based on a transducer, inputting the time series data to be predicted into the trained prognosis prediction model, predicting thyroid function detection indexes in future detection, and outputting a prediction result.
Specifically, in the step S4, a prognosis prediction model is constructed based on the transducer, and the inputting the time series data to be predicted into the prognosis prediction model after training includes:
step S410, configuring a main architecture as a transducer neural network comprising an encoder and a decoder;
step S411, inputting the time series data to be predicted into an encoder, outputting the embedded feature vector after encoding, using the encoded embedded feature vector as an input vector of a multi-layer sensor, mapping the current input vector to an output vector, and calculating by adopting a softmax function to predict each thyroid function detection index result in future examination;
step S412, inversely normalizing the predicted thyroid function detection index results in future examination to obtain corresponding dimensionality prediction results;
step S413, inputting the embedded feature vector obtained through the softmax function and the actual thyroid function detection index into the prognosis prediction model during training, calculating a cross entropy loss function and minimizing the cross entropy loss function, thereby training the prognosis prediction model;
in step S414, if the future multiple detection results are predicted, the current prediction result is used as a new embedded feature vector to be imported into the trained prognosis prediction model, and the prediction result in the next detection is output.
Those skilled in the art will appreciate that a transducer neural network is used as the subject architecture in this implementation. The patient test data sequences collected in this embodiment are sequential data, and are typically processed using a cyclic neural network (RNN), LSTM, GRU, transducer, etc. The biggest difference between the transducer and the LSTM model is that the training process of the LSTM model is iterative and serial, requiring the input characters to be processed one by one. The training of the Transformer is parallel, namely, all characters are trained simultaneously, so that the calculation efficiency is greatly improved, and the method is more efficient in an actual application scene.
The complete transducer comprises an encoding part and a decoding part, and is mainly used for performing natural language processing work, such as machine translation and languageModeling, etc. This embodiment belongs to the category of sequence data classification and only uses the code part (encoder) of the transducer. Before inputting the above-obtained sequence into the model, each 64-dimensional vector R is required to be input i The position information of (2) is transmitted to the transducer, i.e. the sequence needs to be position-coded to obtain the sequence information. The complete coding part mainly comprises character embedding, position coding, self-attention mechanism, residual connection and full connection layers, and the whole structure is shown in fig. 4. The input of the encoder is divided into two branches, one enters a multi-head attention unit, the structure for giving the automatic adaptive weight according to the importance of the input characteristics, the obtained result is directly added with the input, namely, the direct branch from the input is called a 'shortcut', then the direct branch is input into a multi-layer perceptron (MLP), the MLP can be regarded as a unit consisting of linear transformation and nonlinear transformation, the characteristics can be further processed and extracted on the data, and the final output is used as the input of the next module. The output of the encoder is a column of vectors.
In particular to the application scene, the time sequence U to be predicted of all objects to be detected after embedding the previous features j After the position variable embedding (position embedding), the position variable is input to a transducer encoder to obtain an encoded embedded feature vector. The embedded feature vectors are subjected to the following different processes:
seven indexes are detected for the thyroid function after normalization/standardization; the embedded feature vector output by the encoder is designed into two hidden layers through a multi-layer perceptron, the input dimension is the dimension of the embedded feature vector output by the encoder, the output dimension is 7, namely the number of thyroid function detection indexes, each index in the next examination of prediction is calculated by using softmax after the embedded feature vector is obtained, and the result is between 0 and 1 after normalization. And when seven indexes are output, the results are inversely normalized, and the corresponding dimension output is added. And during training, calculating a cross entropy loss function by using the embedded feature vector obtained by softmax and seven indexes at the next examination after actual normalization, and minimizing the cross entropy loss function.
If the result obtained by multiple times of examination after prediction is required, the predicted data can be used as new Rn to be sent to a prognosis prediction model, and a round of calculation is required. The final output result of the prognosis prediction model is as follows: the next time the specific numerical values of the seven indexes of the Jiagong are checked.
Evaluation model in this example: five-fold cross-validation was used to evaluate the performance of the transducer model. In each validation, the dataset was randomly split into 5 aliquots: 4 aliquots of data were used as training data and the remaining 1 aliquot of data were used as test data. Ensuring that there is no overlap between the training data and the test data. The final validation results averaged five times the cross validation results. The evaluation indexes including accuracy (Acc), sensitivity (Sens), specificity, spec, precision (Prec) and Ma Xiusi correlation coefficients (Matthews correlation coefficient, MCC) are used: where TN represents true counterexamples, TP represents true examples, FN represents false counterexamples, and FP represents false positive examples. At the same time, ROC curves and AUCs were also used to evaluate performance.
Meanwhile, the quality evaluation is carried out on the results generated in the embodiment of the invention, such as:
1) analysis of absorptions
In order to evaluate the visibility of the prognosis prediction model to different areas of the image, the embodiment of the invention performs ablation analysis. Since both the image and the ocular anatomy are generally radially symmetric, with most pupils centered, this patent performs an "image visibility" analysis based on concentric circles. The central or peripheral portion of the image is masked during analysis. Under different degrees of masking, the two masks are compared, and the number of visible pixels is controlled at the same time, so that the visibility of the model to different areas is known.
2) Analysis of saliency
To improve the interpretability of the prognosis prediction model, embodiments of the present invention use the integral gradients to generate a map highlighting the significance of important areas in the image, thereby knowing the portion of the model that is most sensitive to the external eye image.
3) image resolution analysis
In order to understand the influence of image resolution on the model and generate an image most suitable for the prognosis prediction model, the embodiment of the invention performs image resolution analysis. And reducing the resolution of the image to a specified value, then increasing the resolution to an original value, and respectively outputting results.
Controlling thyroid function properly is critical to TED patients. The invention provides a novel thyroid eye disease prognosis prediction method, which can judge the thyroid function condition of a TED patient only by using a face image of the TED patient, saves the time for blood test, reduces the time and economic cost, and is efficient and quick.
The output result of the invention is the thyroid function related key hormone index level prediction result. For a TED patient with normal thyroid gland function, a doctor can know the current hormone level, and if the risk degree is high, the doctor can intervene in advance. For a TED patient with thyroid dysfunction, a doctor can judge the development stage of thyroid dysfunction according to the related hormone level, and formulate a corresponding treatment scheme to prevent the further development of the TED patient and improve the prognosis of the patient.
Example two
As shown in fig. 5, based on the same concept, the present invention provides a prognosis prediction apparatus for thyroid eye disease, comprising:
the data acquisition module 100 is configured to acquire historical medical data and sequence data of a facial image, and generate detection data of an object to be predicted, where the historical medical data includes thyroid function detection index data and basic information of the object to be predicted;
the data preprocessing module 200 is configured to preprocess the detection data, classify and process a plurality of detection data according to data types, perform feature embedding on the detection data of different data types of the same detection time point of the object to be predicted to generate an embedded feature vector, encode the embedded feature vector to obtain an encoded embedded feature vector, where the embedded feature vector includes an image feature vector, a category feature vector and a numerical feature vector related to the object to be predicted; generating time sequence data to be predicted by using a plurality of encoded embedded feature vector sets obtained from a plurality of detection time points of the same object to be predicted;
and the prognosis prediction module 300 is used for constructing a prognosis prediction model based on a transducer, inputting the time series data to be predicted into the prognosis prediction model after training, predicting thyroid function detection indexes during future detection, and outputting a prediction result.
It should be noted that, the division of each module in the embodiment of the apparatus/system is merely a division of a logic function, and may be fully or partially integrated into a physical entity or may be physically separated. And these modules may all be implemented in software in the form of calls by the processing element; or can be realized in hardware; the method can also be realized in a form that a part of units are called by processing elements to be software, and the other part of units are realized in a form of hardware.
The implementation principles of the data acquisition module 100, the data preprocessing module 200, and the prognosis prediction module 300 are described in the foregoing embodiments, and thus the description thereof is not repeated here.
Example III
Based on the same conception, in some embodiments of the present application, an electronic device is also provided. The electronic equipment comprises a memory and a processor, wherein the memory is used for storing a processing program, and the processor executes the processing program according to the instruction. The method for predicting thyroid eye disease in the foregoing embodiments is enabled when the processor executes the processing program.
In some embodiments of the present application, a readable storage medium is also provided, which may be a non-volatile readable storage medium or a volatile readable storage medium. The readable storage medium has instructions stored therein that, when executed on a computer, cause an electronic device comprising such readable storage medium to perform the aforementioned method of prognosis prediction of thyroid eye disease.
It will be appreciated that for the aforementioned method of prognosis of thyroid eye disease, if implemented as software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in essence or a part contributing to the prior art or all or part of the technical solution in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-only memory (ROM), a random access memory (Random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable storage medium may also be any readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Program code for carrying out the subject matter disclosed herein may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the C programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of remote computing devices, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., connected via the Internet using an Internet service provider).
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A method for prognosis prediction of thyroid eye disease, comprising:
acquiring historical medical data and facial image data, and generating detection data of an object to be predicted, wherein the historical medical data comprises thyroid function detection index data and basic information of the object to be predicted;
preprocessing the detection data, classifying and processing a plurality of detection data according to data types, performing feature embedding on the detection data of different data types of the same detection time point of the object to be predicted to generate an embedded feature vector, and encoding the embedded feature vector to obtain an encoded embedded feature vector, wherein the embedded feature vector comprises an image feature vector, a category feature vector and a numerical feature vector related to the object to be predicted;
generating time sequence data to be predicted by using a plurality of encoded embedded feature vector sets obtained from a plurality of detection time points of the same object to be predicted;
and constructing a prognosis prediction model based on a transducer, inputting the time series data to be predicted into the trained prognosis prediction model, predicting thyroid function detection indexes in future detection, and outputting a prediction result.
2. The method of claim 1, wherein the historical medical data comprises basic information and thyroid function detection index data, wherein the thyroid function detection index data comprises free triiodothyronine, free tetraiodothyronine, total triiodothyronine, total tetraiodothyronine, thyroid stimulating hormone, thyroglobulin antibody and thyroid peroxidase antibody, and historical test results for each index.
3. The method of claim 1, wherein preprocessing the test data, classifying and processing a plurality of the test data according to data type comprises:
extracting facial image data with thyroid eye disease signs at different detection time points in the detection data;
determining pixel values of a target face image with pixel values in the face image based on a bilinear interpolation algorithm to scale the target face image to obtain the resized target face image;
and inputting the target face image with the adjusted size into a VGG-19 pre-training model to obtain an image feature vector corresponding to the target face image with the adjusted size.
4. The method of claim 1, wherein preprocessing the test data, classifying and processing a plurality of the test data according to data type comprises:
extracting different types of category features in the detection data, wherein the category features at least comprise gender, thyroid disease history and eye symptoms;
converting the sex characteristic, the thyroid history characteristic and the ocular symptom characteristic into a first unique heat encoding characteristic, a second unique heat encoding characteristic and a third unique heat encoding characteristic, respectively;
performing fusion processing on the first single-hot coding feature, the second single-hot coding feature and the third single-hot coding feature to obtain a fusion feature vector as input of a current network input layer;
and multiplying the fusion feature vector by a preset weight matrix, and reducing the dimension to obtain a corresponding category feature vector output by the current network ebedding layer.
5. The method of claim 1, wherein preprocessing the test data, classifying and processing a plurality of the test data according to data type comprises:
Extracting numerical characteristics in the detection data, wherein the numerical characteristics comprise thyroid function detection indexes, ages and smoking indexes;
and carrying out normalization processing on the numerical feature, converting the numerical feature into a preset range through linearization, and outputting a corresponding numerical feature vector.
6. The method for prognosis prediction of thyroid eye disease according to claim 1, wherein the constructing a prognosis prediction model based on a transducer, inputting the time-series data to be predicted into the prognosis prediction model after training, comprises:
configuring a main architecture into a transducer neural network comprising an encoder and a decoder;
inputting the time series data to be predicted into an encoder, outputting the embedded feature vector after encoding, using the encoded embedded feature vector as an input vector of a multi-layer sensor, mapping a current input vector to an output vector, and adopting softmax function calculation to predict each thyroid function detection index result in future examination;
inversely normalizing the predicted thyroid function detection index results in future examination to obtain corresponding dimensionalization prediction results;
during training, the embedded feature vector obtained by calculating a softmax function and an actual thyroid function detection index are input into the prognosis prediction model, a cross entropy loss function is calculated and minimized, and therefore the prognosis prediction model is trained;
If the future multiple detection results are predicted, the current prediction result is used as a new embedded feature vector to be imported into the trained prognosis prediction model, and the prediction result in the next detection is output.
7. The method for prognosis prediction of thyroid-eye disease according to claim 1, characterized in that the method for training and evaluating prognosis prediction model comprises:
dividing the time sequence data to be predicted into a training set, a verification set and a test set according to a preset proportion;
constructing the prognosis prediction model based on a Pytorch deep learning framework, inputting the training set and the verification set into the prognosis prediction model for training, and selecting a loss function of a current model as a minimum cross entropy loss function to obtain the trained prognosis prediction model;
optimizing and updating parameters of the trained prognosis prediction model based on back propagation of a gradient descent method to obtain the optimized prognosis prediction model;
and evaluating the performance of the optimized prognosis prediction model by adopting cross validation, and determining the prediction performance of the optimized prognosis prediction model in a data set formed by the time series data to be predicted by adopting the test set validation.
8. A device for prognosis prediction of thyroid eye disease, comprising:
the data acquisition module is used for acquiring historical medical data and facial image data and generating detection data of an object to be predicted, wherein the historical medical data comprises thyroid function detection index data and basic information of the object to be predicted;
the data preprocessing module is used for preprocessing the detection data, classifying and processing a plurality of detection data according to data types, performing feature embedding on the detection data of different data types of the same detection time point of the object to be predicted to generate an embedded feature vector, and encoding the embedded feature vector to obtain an encoded embedded feature vector, wherein the embedded feature vector comprises an image feature vector, a category feature vector and a numerical feature vector related to the object to be predicted; generating time sequence data to be predicted by using a plurality of encoded embedded feature vector sets obtained from a plurality of detection time points of the same object to be predicted;
and the prognosis prediction module is used for constructing a prognosis prediction model based on a transducer, inputting the time series data to be predicted into the prognosis prediction model after training, predicting thyroid function detection indexes during future detection and outputting a prediction result.
9. An electronic device, comprising:
a memory for storing a processing program;
a processor which, when executing the processing program, implements the method for predicting a thyroid eye disease as defined in any one of claims 1 to 7.
10. A readable storage medium, wherein a processing program is stored on the readable storage medium, which when executed by a processor, implements the method for prognosis prediction of thyroid-eye disease according to any one of claims 1 to 7.
CN202310335253.3A 2023-03-31 2023-03-31 Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease Pending CN116364292A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310335253.3A CN116364292A (en) 2023-03-31 2023-03-31 Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310335253.3A CN116364292A (en) 2023-03-31 2023-03-31 Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease

Publications (1)

Publication Number Publication Date
CN116364292A true CN116364292A (en) 2023-06-30

Family

ID=86930189

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310335253.3A Pending CN116364292A (en) 2023-03-31 2023-03-31 Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease

Country Status (1)

Country Link
CN (1) CN116364292A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116936103A (en) * 2023-09-12 2023-10-24 神州医疗科技股份有限公司 User health prediction management method and system based on homodromous network
CN117556265A (en) * 2024-01-11 2024-02-13 中国医学科学院医学信息研究所 Detection index prediction method, device, equipment and storage medium
CN117612693A (en) * 2024-01-24 2024-02-27 四川互慧软件有限公司 Patient real-time monitoring and early warning method, device, computer and storage medium

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116936103A (en) * 2023-09-12 2023-10-24 神州医疗科技股份有限公司 User health prediction management method and system based on homodromous network
CN116936103B (en) * 2023-09-12 2023-12-15 神州医疗科技股份有限公司 User health prediction management method and system based on homodromous network
CN117556265A (en) * 2024-01-11 2024-02-13 中国医学科学院医学信息研究所 Detection index prediction method, device, equipment and storage medium
CN117612693A (en) * 2024-01-24 2024-02-27 四川互慧软件有限公司 Patient real-time monitoring and early warning method, device, computer and storage medium
CN117612693B (en) * 2024-01-24 2024-04-05 四川互慧软件有限公司 Patient real-time monitoring and early warning method, device, computer and storage medium

Similar Documents

Publication Publication Date Title
CN116364292A (en) Method, device, equipment and storage medium for predicting prognosis of thyroid eye disease
Chen et al. PCAT-UNet: UNet-like network fused convolution and transformer for retinal vessel segmentation
Wang et al. MsTGANet: Automatic drusen segmentation from retinal OCT images
Rajee et al. Gender classification on digital dental x-ray images using deep convolutional neural network
CN113240655B (en) Method, storage medium and device for automatically detecting type of fundus image
Hamwi et al. Development and integration of VGG and dense transfer-learning systems supported with diverse lung images for discovery of the Coronavirus identity
Pikulkaew et al. 2D facial expression and movement of motion for pain identification with deep learning methods
Thangavel et al. EAD-DNN: Early Alzheimer's disease prediction using deep neural networks
Soundrapandiyan et al. AI-based wavelet and stacked deep learning architecture for detecting coronavirus (COVID-19) from chest X-ray images
Ding et al. RCAR-UNet: Retinal vessel segmentation network algorithm via novel rough attention mechanism
CN116797572A (en) Rheumatoid arthritis activity grading device based on multi-mode data
CN114694234B (en) Emotion recognition method, system, electronic device and storage medium
Alshayeji et al. Two-stage framework for diabetic retinopathy diagnosis and disease stage screening with ensemble learning
WO2023108418A1 (en) Brain atlas construction and neural circuit detection method and related product
Shanthakumari et al. Deep Convolutional Neural Network Model for Classifying Alzheimer’s Disease
Ayus et al. A novel hybrid ensemble based Alzheimer’s identification system using deep learning technique
Goyal Novel computerised techniques for recognition and analysis of diabetic foot ulcers
Li et al. An Assisted Diagnosis of Alzheimer's Disease Incorporating Attention Mechanisms Med-3D Transfer Modeling.
CN113436743B (en) Representation learning-based multi-outcome efficacy prediction method, device and storage medium
Agrawal et al. EfficientNet-B3 and Image Processing for Monkeypox Detection using Skin Lesion Images
Bharani et al. Diabetes Mellitus Diagnosis based on Tongue Images using Machine Learning
US20230293094A1 (en) Method,apparatus and computer program for reading rotator cuff tear state or muscle fat degeneration disorder based on artificial intelligence
US20230420098A1 (en) Systems and methods for quantifying patient improvement through artificial intelligence
US20230178200A1 (en) Direct medical treatment predictions using artificial intelligence
Xin et al. Transformer guided self-adaptive network for multi-scale skin lesion image segmentation

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