CN111430025B - Disease diagnosis model training method based on medical image data augmentation - Google Patents
Disease diagnosis model training method based on medical image data augmentation Download PDFInfo
- Publication number
- CN111430025B CN111430025B CN202010161921.1A CN202010161921A CN111430025B CN 111430025 B CN111430025 B CN 111430025B CN 202010161921 A CN202010161921 A CN 202010161921A CN 111430025 B CN111430025 B CN 111430025B
- Authority
- CN
- China
- Prior art keywords
- image data
- neural network
- medical image
- disease
- data
- 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.)
- Active
Links
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
- G16H50/20—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
-
- 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
-
- 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
-
- 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
-
- 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
Abstract
The invention relates to a disease diagnosis method based on medical image data amplification, and belongs to the technical field of disease diagnosis methods. Firstly, mapping original image data into a depth feature space, and extracting features with highly-linearized semantic information; acquiring a characteristic covariance matrix for data amplification according to the distribution of data corresponding to different categories in the medical image in a characteristic space; and calculating a data amplification loss function, and obtaining a model with stronger feature extraction capability by continuously optimizing the loss. The data amplification method is only used when a model is trained, and does not introduce more calculation amount and calculation time when the model is used for predicting the medical image data. The method effectively expands the fixed medical image data, remarkably reduces the quantity requirement of the training of the deep neural network on the marked patient data, effectively relieves the problems of difficult acquisition of the medical image data and high marking cost, and improves the accuracy rate of disease diagnosis.
Description
Technical Field
The invention relates to a disease diagnosis model training method based on medical image data amplification, and belongs to the technical field of disease diagnosis methods.
Background
Medical images are image data of internal tissues, organs and the like of a human body, which are acquired by using modes such as x-ray projection and the like, can provide a large amount of visual medical information, and are increasingly used in modern clinical diagnosis and medical treatment. With the advent of artificial intelligence, computer vision-oriented deep learning technology was used as a rapid, accurate and effective image analysis method, which was widely used for automatic analysis of medical images, and can effectively reduce medical risks such as misdiagnosis caused by factors such as insufficient subjective experience and cognitive fluctuation of doctors. However, deployment of deep learning algorithms requires sufficient labeling data, and some cases of disorders are scarce, often making it difficult to obtain sufficient valid data. On the other hand, medical image labeling requires a large number of experienced radiologists, has extremely high labor cost, and is difficult to obtain large-scale labeling samples. Therefore, how to use a small amount of labeling samples to obtain a deep learning model as good as possible has a key meaning.
Aiming at the problem of data scarcity, the data amplification method is an effective method for amplifying training data based on a small number of samples. The existing method mainly starts from two aspects: (1) The method adopts a natural picture amplification mode to perform simple geometric transformation (translation, scaling and the like) of pixel space and image attribute transformation (brightness, contrast and the like) on an original image so as to obtain more training samples, but the method has the risk of losing focus information contained in the image in consideration of the specificity of the medical image, and the obtained practical effect is limited; (2) Training depth generation networks or self-encoders, such as Wasserstein Generation Antagonism Networks (WGAN), variational self-encoders (VAE), etc., randomly generates similar medical images by introducing random variables, repeatedly reasoning, but the implementation process of the method is very complex and has only a small improvement in effect compared with the former.
Disclosure of Invention
The invention aims to provide a disease diagnosis model training method based on medical image data amplification, which improves the existing disease diagnosis method, amplifies training data based on a small amount of samples by adopting a data amplification method, obtains a more accurate disease diagnosis model through short-time efficient training, and improves the accuracy of disease diagnosis by using the model.
The invention provides a disease diagnosis model training method based on medical image data amplification, which comprises the following steps:
(1) Acquiring medical image data of a plurality of patients from a hospital or related unit, wherein the medical image data of the ith patient is denoted as x i ,x i Is a three-dimensional matrix of AXHXW, A is the channel number of the image data, H and W are the height and width of the image data respectively, each x i With a diseased label y i Correspondingly, y i Is an integer with a value between 0 and K, which is given by a specialist for indicating what kind of disease the patient suffers from, wherein K is the number of disease types to be diagnosed, 0 represents no disease, i.e. y i =k represents that the patient has the kth disease, y i =0 represents that the patient is not ill;
(2) Establishing a neural network which is formed by arranging a plurality of function layers according to ResNet rules or DenseNet rules, setting parameters of the neural network as Θ, and obtaining medical image data x of the ith patient in the step (1) i As an input to the neural network, the output of the upper layer in the neural network is used as the input of the lower layer until the last layer, and a feature f is output from the neural network i Is marked as
f i =g(x i ;Θ)
Where g is the transfer function of the neural network, f i Is a vector of F×1, F is a feature F i Is determined by the structure of the neural network;
(3) For each type of disease from 0 to K, an overall feature covariance matrix Σ for data augmentation is defined separately k And overall feature mean vectorWherein k is disease category number, Σ k Is F x F,)>Is of dimension Fx 1, will be k Initializing to unity matrix, will ∈ ->Initializing to zero vector, F is characteristic F in step (2) i Dimension number of (a);
(4) Initializing the neural network parameters theta obtained in the step (2) with standard normal distribution, and training the neural network parameters theta to obtain the neural network for disease diagnosis, wherein the method comprises the following steps of:
(4-1) setting training algebra t=0 and setting total training algebra E at the time of initialization;
(4-2) randomly extracting B data from the medical image data of the step (1), and recording asThe corresponding diseased label is->B pieces of medical image data are input into the neural network in the step (2), and B pieces of network output characteristics corresponding to the medical image data are obtained according to the following formula>
Wherein g and Θ are the neural network transfer function and neural network parameters defined in step (2), respectively;
(4-3) the B features obtained according to step (4-2)Calculating the eigenvalue vector of each disease category using the following formula>Sum-feature covariance matrix sigma k ′:
Wherein the superscript T is a matrix transposition operation, k is a disease category number, b=1, 2, …, B,a class discriminant function defined for:
(4-4) updating the ensemble feature covariance matrix Σk and the ensemble feature mean vector of step (3) using the following expression
Wherein n isThe number of patients with the kth disease contained in the medicine, m is the total number of the patients with the kth disease calculated by the method so far;
(4-5) performing the following form of linear operation on the results obtained in (4-2) to obtain discrimination scores
Wherein W is a scoring matrix, the dimension of W is (K+1) x F, C is a scoring bias vector, the dimension of C is (K+1) x 1, K is the disease type in step (1), and F is the number of feature dimensions in step (2);
(4-6) obtaining a data amplification loss L by using the following data amplification loss function:
wherein lambda is a set data amplification factor,the value range is 0.5-1, w k For the kth row, c of the scoring matrix w k For scoring the kth element of the bias vector c,for the b-th medical image data +.>The corresponding diseased tag is a constant between 0 and K, < >>And->Respectively represent +.>Line and scoring bias vector C +.>The superscript T is transposition operation;
(4-7) reversely deriving the data amplification loss L to obtain the partial derivatives of the parameters theta, the scoring matrix W and the scoring bias vector C of the neural network by LAnd->Update Θ, W and C with the following formula:
η is the learning rate calculated from the following formula:
wherein t and E are the current training algebra and the total training algebra defined in step (4-1), cos represents the cosine function, η 0 Is the initial learning rate;
(4-8) making training algebra t=t+1, judging t, if t is less than or equal to E, returning to (4-2), and if t is more than E, obtaining a neural network for disease diagnosis corresponding to the parameters Θ, the scoring matrix W and the scoring bias vector C in the step (4-7);
(5) Medical imaging of a patient to be diagnosedInputting the characteristic +.A.A.in the neural network for disease diagnosis of the step (4-8)>For characteristics->The following operation is performed to obtain a discrimination score +.>
Wherein W and C are the scoring matrix and the scoring bias vector obtained in the step (4-8), respectively, to be compared withSerial number corresponding to the element with the largest value in +.>As a diagnostic result of the disease type or not, the +.>And the value is one of 0 to K, so that disease diagnosis based on medical image data amplification is realized.
The disease diagnosis model training method based on medical image data amplification provided by the invention is characterized in that:
the disease diagnosis model training method based on medical image data augmentation of the invention aims at the defects existing in the existing automatic disease diagnosis technology based on medical image data, for example, the prior art is required to rely on a large amount of marked patient data to train a deep neural network, however, the marking of the medical image data requires a large amount of experienced radiologists, consumes a long time to finish, and has higher time cost and economic cost. Firstly, mapping original image data into a depth feature space, and extracting features with highly-linearized semantic information; secondly, according to the distribution of data corresponding to different categories in the medical image in the feature space, a feature covariance matrix for data amplification is obtained; finally, a data amplification loss function is calculated, and a model with stronger feature extraction capability is obtained by continuously optimizing the loss. The data amplification method is only used when a model is trained, and does not introduce more calculation amount and calculation time when the model is used for predicting the medical image data. According to the method, the fixed medical image data are effectively expanded, the quantity requirement of training of the deep neural network on the marked patient data is remarkably reduced, and therefore the problems of difficult acquisition of the medical image data and high marking cost are effectively solved; in addition, the training samples are subjected to data amplification, so that the neural network model can receive more information in the training process, the prediction accuracy, the robustness and the prediction stability of the obtained disease diagnosis model can be effectively improved, and the disease diagnosis effect is effectively improved; finally, aiming at the characteristics of medical influence, the invention adopts a semantic data amplification method, so that on one hand, the risk of losing effective focus information of an image is reduced, a more reliable prediction model can be obtained, and on the other hand, the convergence speed is accelerated, so that the training process is more efficient.
The method utilizes a training sample data amplification mode, so that the deep neural network can receive more information in the training process, the robustness of the deep neural network is enhanced, and the diagnosis precision and the prediction stability of the obtained disease diagnosis model are remarkably improved. The method adopts a semantic data amplification mode, and effectively solves the problem that focus information is possibly damaged in the traditional image-based data amplification method; on the other hand, the problems of slow model convergence and high calculation cost caused by repeated input of a large amount of reinforced medical image data in the traditional method are avoided, the convergence speed is greatly increased, and the training process can be completed in a short time and high efficiency.
Detailed Description
The invention provides a disease diagnosis model training method based on medical image data amplification, which comprises the following steps:
acquiring medical image data of a plurality of patients from a hospital or related unit, wherein the medical image data of the ith patient is denoted as x i ,x i Is a three-dimensional matrix of AXHXW, A is the channel number of the image data, H and W are the height and width of the image data respectively, each x i With a diseased label y i Correspondingly, y i Is an integer with a value between 0 and K, which is given by a specialist for indicating what kind of disease the patient suffers from, wherein K is the number of disease types to be diagnosed, 0 represents no disease, i.e. y i =k represents that the patient has the kth disease, y i =0 represents that the patient is not ill;
(2) Establishing a neural network which is formed by arranging a plurality of function layers according to ResNet rules or DenseNet rules, setting parameters of the neural network as Θ, and obtaining medical image data x of the ith patient in the step (1) i As an input to the neural network, the output of the upper layer in the neural network is used as the input of the lower layer until the last layer, and a feature f is output from the neural network i Is marked as
f i =g(x i ;Θ)
Where g is the transfer function of the neural network, f i Is a vector of F×1, F is a feature F i Is determined by the structure of the neural network;
(3) For each type of disease from 0 to K, an overall feature covariance matrix Σ for data augmentation is defined separately k And overall feature mean vectorWherein k is disease category number, Σ k Is F x F,)>Is of dimension Fx 1, will be k Initializing to unity matrix, will ∈ ->Initializing to zero vector, F is characteristic F in step (2) i Dimension number of (a);
(4) Initializing the neural network parameters theta obtained in the step (2) with standard normal distribution, and training the neural network parameters theta to obtain the neural network for disease diagnosis, wherein the method comprises the following steps of:
(4-1) setting training algebra t=0 and setting total training algebra E at the time of initialization;
(4-2) randomly extracting B data from the medical image data of the step (1), and recording asThe corresponding diseased label is->B pieces of medical image data are input into the neural network in the step (2), and B pieces of network output characteristics corresponding to the medical image data are obtained according to the following formula>
Wherein g and Θ are the neural network transfer function and neural network parameters defined in step (2), respectively;
(4-3) the B features obtained according to step (4-2)Calculating the eigenvalue vector of each disease category using the following formula>Sum-feature covariance matrix sigma k ′:
Wherein the superscript T is a matrix transposition operation, k is a disease category number, b=1, 2, …, B,a class discriminant function defined for:
(4-4) updating the ensemble feature covariance matrix Σk and the ensemble feature mean vector of step (3) using the following expression
Wherein n isThe number of patients with the kth disease contained in the medicine, m is the total number of the patients with the kth disease calculated by the method so far;
(4-5) performing the following form of linear operation on the results obtained in (4-2) to obtain discrimination scores
Wherein W is a scoring matrix, the dimension of W is (K+1) x F, C is a scoring bias vector, the dimension of C is (K+1) x 1, K is the disease type in step (1), and F is the number of feature dimensions in step (2);
(4-6) obtaining a data amplification loss L by using the following data amplification loss function:
wherein lambda is a set data amplification coefficient, and the value range is 0.5-1, w k For the kth row, c of the scoring matrix w k For scoring the kth element of the bias vector c,for the b-th medical image data +.>The corresponding diseased tag is a constant between 0 and K, < >>And->Respectively represent +.>Line and scoring bias vector C +.>The superscript T is transposition operation;
(4-7) reversely deriving the data amplification loss L to obtain the partial derivatives of the parameters theta, the scoring matrix W and the scoring bias vector C of the neural network by LAnd->Update Θ, W and C with the following formula:
η is the learning rate calculated from the following formula:
wherein t and E are the current training algebra and the total training algebra defined in step (4-1)The number cos represents the cosine function, eta 0 Is the initial learning rate;
(4-8) making training algebra t=t+1, judging t, if t is less than or equal to E, returning to (4-2), and if t is more than E, obtaining a neural network for disease diagnosis corresponding to the parameters Θ, the scoring matrix W and the scoring bias vector C in the step (4-7);
(5) Medical imaging of a patient to be diagnosedInputting the characteristic +.A.A.in the neural network for disease diagnosis of the step (4-8)>For characteristics->The following operation is performed to obtain a discrimination score +.>
Wherein W and C are the scoring matrix and the scoring bias vector obtained in the step (4-8), respectively, to be compared withSerial number corresponding to the element with the largest value in +.>As a diagnostic result of the disease type or not, the +.>And the value is one of 0 to K, so that disease diagnosis based on medical image data amplification is realized.
The above examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principle of the present invention should be made in the equivalent manner, and the embodiments are included in the protection scope of the present invention.
Claims (1)
1. A disease diagnosis model training method based on medical image data augmentation is characterized by comprising the following steps:
(1) Acquiring medical image data of a plurality of patients from a hospital or related unit, wherein the medical image data of the ith patient is denoted as x i ,x i Is a three-dimensional matrix of AXHXW, A is the channel number of the image data, H and W are the height and width of the image data respectively, each x i With a diseased label y i Correspondingly, y i Is an integer with a value between 0 and K, which is given by a specialist for indicating what kind of disease the patient suffers from, wherein K is the number of disease types to be diagnosed, 0 represents no disease, i.e. y i =k represents that the patient has the kth disease, y i =0 represents that the patient is not ill;
(2) Establishing a neural network which is formed by arranging a plurality of function layers according to ResNet rules or DenseNet rules, setting parameters of the neural network as Θ, and obtaining medical image data x of the ith patient in the step (1) i As an input to the neural network, the output of the upper layer in the neural network is used as the input of the lower layer until the last layer, and a feature f is output from the neural network i Denoted as f i =g(x i ;Θ)
Where g is the transfer function of the neural network, f i Is a vector of F×1, F is a feature F i Is determined by the structure of the neural network;
(3) For each type of disease from 0 to K, an overall feature covariance matrix Σ for data augmentation is defined separately k And overall feature mean vectorWherein k is disease category number, Σ k Dimension of (2)Degree is F×F->Is of dimension Fx 1, will be k Initializing to unity matrix, will ∈ ->Initializing to zero vector, F is characteristic F in step (2) i Dimension number of (a);
(4) Initializing the neural network parameters theta obtained in the step (2) with standard normal distribution, and training the neural network parameters theta to obtain the neural network for disease diagnosis, wherein the method comprises the following steps of:
(4-1) setting training algebra t=0 and setting total training algebra E at the time of initialization;
(4-2) randomly extracting B data from the medical image data of the step (1), and recording asThe corresponding diseased label is->B pieces of medical image data are input into the neural network in the step (2), and B pieces of network output characteristics corresponding to the medical image data are obtained according to the following formula>
Wherein g and Θ are the neural network transfer function and neural network parameters defined in step (2), respectively;
(4-3) the B features obtained according to step (4-2)The following formula is usedCalculating the eigenvalue vector of each disease category +.>Sum-feature covariance matrix sigma k ′:
Wherein the superscript T is a matrix transposition operation, k is a disease category number, b=1, 2, …, B,a class discriminant function defined for:
(4-4) updating the ensemble feature covariance matrix Σk and the ensemble feature mean vector of step (3) using the following expression
Wherein n isComprises the following componentsM is the total number of patients with the kth disease calculated so far;
(4-5) performing the following form of linear operation on the results obtained in (4-2) to obtain discrimination scores
Wherein W is a scoring matrix, the dimension of W is (K+1) x F, C is a scoring bias vector, the dimension of C is (K+1) x 1, K is the disease type in step (1), and F is the number of feature dimensions in step (2);
(4-6) obtaining a data amplification loss L by using the following data amplification loss function:
wherein lambda is a set data amplification coefficient, and the value range is 0.5-1, w k For the kth row, c of the scoring matrix w k For scoring the kth element of the bias vector c,for the b-th medical image data +.>The corresponding diseased tag is a constant between 0 and K, < >>And->Respectively represent +.>Line and scoring bias vector C +.>The superscript T is transposition operation;
(4-7) reversely deriving the data amplification loss L to obtain the partial derivatives of the parameters theta, the scoring matrix W and the scoring bias vector C of the neural network by LAnd->Update Θ, W and C with the following formula:
η is the learning rate calculated from the following formula:
wherein t and E are the current training algebra and the total training algebra defined in step (4-1), cos represents the cosine function, η 0 Is the initial learning rate;
(4-8) making training algebra t=t+1, judging t, if t is less than or equal to E, returning to (4-2), and if t is more than E, obtaining a neural network for disease diagnosis corresponding to the parameters Θ, the scoring matrix W and the scoring bias vector C in the step (4-7);
(5) Medical imaging of a patient to be diagnosedInputting the characteristic +.A.A.in the neural network for disease diagnosis of the step (4-8)>For characteristics->The following operation is performed to obtain a discrimination score +.>
Wherein W and C are the scoring matrix and the scoring bias vector obtained in the step (4-8), respectively, to be compared withSerial number corresponding to the element with the largest value in +.>As a diagnostic result of the disease type or not, the +.>And the value is one of 0 to K, so that disease diagnosis based on medical image data amplification is realized.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010161921.1A CN111430025B (en) | 2020-03-10 | 2020-03-10 | Disease diagnosis model training method based on medical image data augmentation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010161921.1A CN111430025B (en) | 2020-03-10 | 2020-03-10 | Disease diagnosis model training method based on medical image data augmentation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111430025A CN111430025A (en) | 2020-07-17 |
CN111430025B true CN111430025B (en) | 2023-07-25 |
Family
ID=71547679
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010161921.1A Active CN111430025B (en) | 2020-03-10 | 2020-03-10 | Disease diagnosis model training method based on medical image data augmentation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111430025B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112183723B (en) * | 2020-09-17 | 2022-07-05 | 西北工业大学 | Data processing method for clinical detection data missing problem |
CN112581458B (en) * | 2020-12-24 | 2024-03-26 | 清华大学 | Image processing method and device |
CN112750530A (en) * | 2021-01-05 | 2021-05-04 | 上海梅斯医药科技有限公司 | Model training method, terminal device and storage medium |
CN115719647B (en) * | 2023-01-09 | 2023-04-11 | 之江实验室 | Hemodialysis-concurrent cardiovascular disease prediction system integrating active learning and contrast learning |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647272A (en) * | 2018-04-28 | 2018-10-12 | 江南大学 | A kind of small sample extending method based on data distribution |
CN110021022A (en) * | 2019-02-21 | 2019-07-16 | 哈尔滨理工大学 | A kind of thyroid gland nuclear medical image diagnostic method based on deep learning |
CN110400610A (en) * | 2019-06-19 | 2019-11-01 | 西安电子科技大学 | Small sample clinical data classification method and system based on multichannel random forest |
EP3576020A1 (en) * | 2018-05-30 | 2019-12-04 | Siemens Healthcare GmbH | Methods for generating synthetic training data and for training deep learning algorithms for tumor lesion characterization, method and system for tumor lesion characterization, computer program and electronically readable storage medium |
CN110689089A (en) * | 2019-10-12 | 2020-01-14 | 电子科技大学 | Active incremental training method for deep learning of multi-class medical image classification |
-
2020
- 2020-03-10 CN CN202010161921.1A patent/CN111430025B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647272A (en) * | 2018-04-28 | 2018-10-12 | 江南大学 | A kind of small sample extending method based on data distribution |
EP3576020A1 (en) * | 2018-05-30 | 2019-12-04 | Siemens Healthcare GmbH | Methods for generating synthetic training data and for training deep learning algorithms for tumor lesion characterization, method and system for tumor lesion characterization, computer program and electronically readable storage medium |
CN110021022A (en) * | 2019-02-21 | 2019-07-16 | 哈尔滨理工大学 | A kind of thyroid gland nuclear medical image diagnostic method based on deep learning |
CN110400610A (en) * | 2019-06-19 | 2019-11-01 | 西安电子科技大学 | Small sample clinical data classification method and system based on multichannel random forest |
CN110689089A (en) * | 2019-10-12 | 2020-01-14 | 电子科技大学 | Active incremental training method for deep learning of multi-class medical image classification |
Non-Patent Citations (1)
Title |
---|
余行 等.利用卷积神经网络对CT图像进行定位的可行性研究.中国医疗器械杂志.2019,第43卷(第06期),第454-458页. * |
Also Published As
Publication number | Publication date |
---|---|
CN111430025A (en) | 2020-07-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111430025B (en) | Disease diagnosis model training method based on medical image data augmentation | |
Zhuang et al. | An Effective WSSENet-Based Similarity Retrieval Method of Large Lung CT Image Databases. | |
Salido et al. | Using deep learning to detect melanoma in dermoscopy images | |
CN108898160B (en) | Breast cancer histopathology grading method based on CNN and imaging omics feature fusion | |
CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
CN107833629A (en) | Aided diagnosis method and system based on deep learning | |
CN108573490A (en) | A kind of intelligent read tablet system for tumor imaging data | |
CN111767952B (en) | Interpretable lung nodule benign and malignant classification method | |
CN113674281A (en) | Liver CT automatic segmentation method based on deep shape learning | |
CN112784879A (en) | Medical image segmentation or classification method based on small sample domain self-adaption | |
CN110610489B (en) | Optical laryngoscope image lesion area marking method based on attention mechanism | |
CN109448822A (en) | CT image pulmonary emphysema automatic marking method based on image report | |
CN111783792B (en) | Method for extracting significant texture features of B-ultrasonic image and application thereof | |
CN108877923A (en) | A method of the tongue fur based on deep learning generates prescriptions of traditional Chinese medicine | |
CN112102276A (en) | Low-field-intensity MR stomach segmentation method based on transfer learning image enhancement | |
CN114565613B (en) | Pancreas postoperative diabetes prediction system based on there is study of supervision degree of depth subspace | |
CN111784639A (en) | Oral panoramic film dental caries depth identification method based on deep learning | |
JP2024035070A (en) | Multi-mode medical data fusion system based on multi-view subspace clustering | |
CN113159223A (en) | Carotid artery ultrasonic image identification method based on self-supervision learning | |
CN113643261B (en) | Lung disease diagnosis method based on frequency attention network | |
Dong et al. | Supervised learning-based retinal vascular segmentation by m-unet full convolutional neural network | |
CN110136113A (en) | A kind of vagina pathology image classification method based on convolutional neural networks | |
CN113627564A (en) | Deep learning-based CT medical image processing model training method and diagnosis and treatment system | |
CN109949299A (en) | A kind of cardiologic medical image automatic segmentation method |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |