CN113658110A - Medical image identification method based on dynamic field adaptive learning - Google Patents
Medical image identification method based on dynamic field adaptive learning Download PDFInfo
- Publication number
- CN113658110A CN113658110A CN202110832315.2A CN202110832315A CN113658110A CN 113658110 A CN113658110 A CN 113658110A CN 202110832315 A CN202110832315 A CN 202110832315A CN 113658110 A CN113658110 A CN 113658110A
- Authority
- CN
- China
- Prior art keywords
- adaptive
- loss
- domain
- medical image
- individual
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 26
- 210000004556 brain Anatomy 0.000 claims abstract description 20
- 238000012360 testing method Methods 0.000 claims abstract description 20
- 230000006870 function Effects 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 3
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 9
- 238000013459 approach Methods 0.000 claims description 2
- 230000015572 biosynthetic process Effects 0.000 claims description 2
- 238000002156 mixing Methods 0.000 claims description 2
- 238000003786 synthesis reaction Methods 0.000 claims description 2
- 238000013526 transfer learning Methods 0.000 abstract description 10
- 238000002610 neuroimaging Methods 0.000 abstract description 3
- 230000006978 adaptation Effects 0.000 description 17
- 238000009826 distribution Methods 0.000 description 9
- 230000009191 jumping Effects 0.000 description 5
- 238000013508 migration Methods 0.000 description 5
- 230000005012 migration Effects 0.000 description 5
- 102000001554 Hemoglobins Human genes 0.000 description 4
- 108010054147 Hemoglobins Proteins 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 238000010191 image analysis Methods 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000002059 diagnostic imaging Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 102000008186 Collagen Human genes 0.000 description 1
- 108010035532 Collagen Proteins 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007177 brain activity Effects 0.000 description 1
- 230000036995 brain health Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 229920001436 collagen Polymers 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002599 functional magnetic resonance imaging Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000002329 infrared spectrum Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000004066 metabolic change Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 102000004169 proteins and genes Human genes 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30016—Brain
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a medical image identification method based on dynamic field adaptive learning, relates to the technical field of transfer learning, and solves the problem of high acquisition cost of labeled data in near-infrared brain imaging. The invention comprises the following steps: preprocessing is carried out on the basis of the ImageNet data set to obtain a source domain data set, and meanwhile, a near-infrared brain imager is used for collecting a small amount of brain medical image data to obtain a target domain data set; defining a convolutional neural network model for image recognition based on a source domain data set, training the convolutional neural network by using a back propagation algorithm, updating parameters until the network converges, and finishing the training; and substituting the target domain data set into the trained convolutional neural network to obtain the recognition output result of the brain medical image data, and testing. The invention realizes the function of dynamically and automatically searching the optimal opportunity for transfer learning, and obviously improves the classification precision and the convergence rate of the network.
Description
Technical Field
The invention belongs to the field of transfer learning, and relates to a medical image identification method based on dynamic field adaptive learning.
Background
With the continuous development and progress of medical imaging technology and computer technology, medical image analysis has become an indispensable tool and technological means in medical research, clinical disease diagnosis and treatment. In recent years, Deep Learning (DL), in particular, a Deep Convolutional Neural Network (CNN), has rapidly developed into a research hotspot of medical image analysis, which can automatically diagnose a disease hidden in a special area from medical image big data. In the field of medical imaging, a physician or researcher usually needs to know some detailed information about a certain internal tissue organ in order to make a correct treatment decision when performing quantitative analysis, real-time monitoring and treatment planning on such tissue organ. Biomedical imaging has become an integral part of, and increasingly important in, disease diagnosis and treatment. However, due to the particularity of medical image data, it is extremely difficult to acquire a large amount of labeled medical data, and the neural network is often trained only by a small amount of labeled data. It is therefore of great interest to study how to use small amounts of tagged medical data for medical image analysis.
Transfer learning (Transfer learning) refers to a learning process in which a model learned in an old field is applied to a new field by using similarities between data, tasks, and models. In the conventional classification learning, in order to ensure that the classification model obtained by training has accuracy and high reliability, there are two basic assumptions: (1) the training sample for learning and the new test sample satisfy independent same distribution; (2) there must be enough training samples available to learn a good classification model. However, in practical applications, we find that these two conditions are often not satisfied. How to use a small amount of labeled training samples or source domain data to establish a reliable model to predict target domains with different data distributions. This is the problem to be solved by the transfer learning.
Domain Adaptation (Domain Adaptation), which is a representative method in transfer learning, refers to using information-rich source Domain samples to improve the performance of a target Domain model. Two crucial concepts in the domain adaptation problem: a source domain (source domain) represents a different domain from the test sample, but has rich supervisory information; the target domain represents the domain where the test sample is located, with no or only a few labels. The source domain and the target domain tend to belong to the same class of tasks, but are distributed differently. According to different types of target domains and source domains, the domain adaptive problem has four different scenes: unsupervised, supervised, heterogeneous distribution and multiple source domain problems. By performing domain adaptation at different stages, researchers have proposed four different domain adaptation methods:
1) and (4) sample self-adaptation, namely performing weighted resampling on the source domain samples so as to approximate the distribution of the target domain.
2) Feature adaptation, projecting the source domain and the target domain into a common feature subspace.
3) And (4) model self-adaptation, namely modifying the error function of the source domain and considering the error of the target domain.
4) And (4) performing relation self-adaptation, mining and carrying out analogy migration by utilizing the relation between the source domain and the target domain.
The maximum Mean difference mmd (maximum Mean Di spread), which is the most frequently used metric in migration learning, measures the distance between two distributions in the regenerated hilbert space, and is a nuclear learning method. Many domain adaptive methods use maximum mean difference to measure the distribution difference between the source domain and the target domain and perform knowledge migration by reducing the distribution difference. The mathematical formula is as follows:
domain Adaptation (Domain Adaptation) is a representative method in migration learning, and refers to using information-rich source Domain samples to improve the performance of a target Domain model. Two crucial concepts in the domain adaptation problem: a source domain (source domain) represents a different domain from the test sample, but has rich supervisory information; the target domain represents the domain where the test sample is located, with no or only a few labels. The source domain and the target domain tend to belong to the same class of tasks, but are distributed differently.
Deep network Adaptation, many deep learning methods develop an Adaptation Layer (Adaptation Layer) to perform the Adaptation of the source domain and target domain data. The self-adaptation can make the data distribution of the source domain and the target domain closer, thereby making the effect of the network better. Most important in deep networks is the definition of network loss, and most deep migration learning methods adopt the following loss definition mode:
L＝Lclass+λLA
where L represents the ultimate loss of the network, LclassRepresents the regular classification loss of the network over the source domain, LAWhich represents the loss of adaptation of the network, is usually represented using the maximum mean difference MMD. And λ is a weighting parameter that trades off the two components.
As a newly developed functional brain imaging device, compared with the existing brain imaging and detection devices, the functional near infrared spectrum imaging (fNIRS) has the advantages that the instrument is noiseless during operation, the requirements of users on the conventional imaging space and time resolution are met, the anti-interference performance is good, the manufacturing cost is low, and the instrument is portable, wearable and noninvasive, and is suitable for popularization and application in places such as communities, families, schools and the like. The types of devices for which fNIRS can be used for brain health monitoring include functional magnetic resonance imaging, in particular, fNIRS uses two different wavelengths of near infrared light to measure changes in the concentration of oxygenated hemoglobin (HbO or O2Hb), deoxygenated hemoglobin (HbR or HHb), total hemoglobin (tHbO) in the corresponding brain regions during brain activity, thereby monitoring metabolic changes in neural activity within the brain. fNIRS is a method of measuring the change in the concentration of HbO and HbR by absorbing near infrared light diffusely reflected back through tissue. Near infrared light is chosen because of its relatively low absorption by substances such as water, hemoglobin, proteins, collagen, and fat.
Disclosure of Invention
The invention aims to: the invention provides a medical image recognition method based on self-adaptive learning in the dynamic field, which solves the technical scheme adopted by the invention as follows:
1. a medical image recognition method based on dynamic field adaptive learning is characterized in that: the method comprises the following steps:
s101: preprocessing is carried out on the basis of the image data set to obtain a source domain data set, and meanwhile, a near-infrared brain imager is used for collecting a small amount of brain medical image data to obtain a target domain data set;
s102: defining a convolutional neural network model for image recognition based on a source domain dataset, initializing each level parameter and training round number epochmaxAdding a domain self-adaptive layer before an output layer of a full connection layer in the convolutional neural network;
s103: calculating the domain-adaptive loss L of a convolutional neural networkMMDAnd a classification loss Lclassification；
S104: defining an initial population lambda1To lambdaiAnd searching the optimal position of the adaptive parameter lambda based on the differential evolution algorithm, and based on the optimal adaptive weight parameter lambda and the domain adaptive loss LMMDAnd a classification loss LclassificationTraining the convolutional neural network by using a back propagation algorithm, updating parameters until the network converges, and finishing the training;
s105: and substituting the target domain data set into the trained convolutional neural network to obtain the recognition output result of the brain medical image data, and testing.
Further, in S102, defining a convolutional neural network model for image recognition based on the source domain data set specifically includes: and defining the total number of the neural network layers, including a convolutional layer and a fully-connected layer, defining the parameters of each layer of the convolutional layer and the parameters of the fully-connected layer, adding an adaptive layer after the last-but-one fully-connected layer, and calculating the maximum mean difference MMD between a source domain and a target domain through the output of the adaptive layer.
Further, in S104, an adaptive loss L is constructed based on the adaptive weight parameter λMMDAdjusting parameters of each layer of the convolutional neural network specifically comprises: round number epoch based on preset population and population evolution stagesAcquiring an optimal adaptive weight parameter lambda by using a differential evolution algorithm;
the loss function is defined as: l ═ Lclassification+λLMMDWherein L represents the net ultimate loss LclassificationRepresents the regular classification loss of the network over the source domain, LMMDRepresents the adaptive loss of the network, here represented by the maximum mean difference MMD;
round number epoch of training phase based on preset neural networktAnd performing iterative training on the convolutional neural network by using a back propagation algorithm until the final loss L is converged, and stopping training.
Further, the method for obtaining the adaptive weight parameter λ in the evolution phase specifically includes: initializing number and position [ lambda ] of original population1,λ2,λ3,…,λi]And copying each individual once to form an individual pair [ lambda ]1,λ1,λ2,λ2,λ3,λ3,…,λi,λi]；
Performing network iterative training by using each individual pair in turn to obtain the same individual lambdaiThe deceleration speed of the two adjacent iteration loss functions is used as an individual adaptive value to evaluate the current individual lambdaiPerformance of (d);
realizing individual variation through a differential evolution algorithm, taking a difference vector of two individuals randomly selected from a population as a random variation source of a third individual, weighting the difference vector, and carrying out vector synthesis with the third individual to generate a variant individual;
randomly selecting individuals through crossing, and carrying out parameter mixing on the variant individuals and a predetermined target individual to generate test individuals; if the adaptive value of the test individual is better than that of the target individual, replacing the target individual with the test individual in the next generation, otherwise, still storing the target individual;
and (4) performing iterative training, reserving good individuals in each generation, eliminating poor individuals, and guiding the search process to approach to the global optimal solution to obtain the optimal adaptive weight parameter lambda.
Further, in S104, the variation function of the differential evolution algorithm is defined as:
wherein r is1，r1And r1Is three random numbers with interval of [1, i]F is a scaling factor which is a determined constant; g represents the g-th.
Further, in the step S104, L is lost through domain adaptationMMDAnd a classification loss LclassificationThe value of (2) completes the training, and the method comprises the following steps:
calculating the current loss L, i.e. the classification loss LclassificationMaximum mean difference MMD domain adaptive loss L of optimal adaptive weight parameter lambdaMMDSumming;
and when the loss is greater than the expected value, iterating by using a back propagation algorithm, and adjusting the neural network parameters until the loss value is stable and the network model converges.
Further, in S105, the target domain data set is a brain medical image data set, which specifically includes: storing the trained network model parameters; and (4) carrying out precision test on the network model by using the brain medical image data set, and outputting a test result.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. the invention provides an effective dynamic field adaptation method of a medical image to solve the problem of 'when to transfer'. The method automatically searches for a proper transfer learning opportunity in the training process, and is not a traditional fixed-domain adaptive parameter lambda method.
2. The method uses the self-adaptive weight parameter lambda to optimize the loss function, and compared with the accuracy of the traditional transfer learning method, the method not only improves the convergence speed of the algorithm, but also improves the test precision.
3. The invention provides a dynamic field self-adaptive method applicable to medical image recognition, which can assist in training a small number of marked medical image domains by means of a trained domain and has great application value.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and that for those skilled in the art, other relevant drawings can be obtained according to the drawings without inventive effort, wherein:
FIG. 1 is a flow chart of a dynamic domain adaptation method proposed by the present invention;
FIG. 2 is a schematic diagram of an ImageNet image dataset;
FIG. 3 is a schematic diagram of a medical brain image dataset acquired by the FNI RS;
FIG. 4 is a flow chart for training a neural network using a back propagation algorithm.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the detailed description and specific examples, while indicating the preferred embodiment of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
It is noted that relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The features and properties of the present invention are described in further detail below with reference to examples.
Example 1
Referring to fig. 1, the present embodiment provides a medical image recognition method based on dynamic domain adaptive learning, including the following steps:
s101: the data is preprocessed based on the ImagNet dataset, an example of an image in the dataset, as shown in fig. 2, is defined as the source domain dataset. Acquiring medical image data of human brain by using a near-infrared brain imager, as shown in fig. 3, and manually marking a small amount of data to define the data as a target domain data set;
s102: defining the structure of the convolutional neural network, defining how to obtain output from the input of the neural network, using a classical convolutional neural network model such as AlexNet or VGGNet, and adding a domain adaptive layer before the output layer of the full-connection layer in the convolutional neural network;
s103: domain adaptive loss L defining convolutional neural networkMMDAnd a classification loss Lclassification；
Where the domain adaptation loses LMMDIs defined as:
wherein XSAnd XTRepresenting the source domain data set and the target domain data set, respectively, phi (—) is a kernel function that can map the original distribution to a reproducible hilbert space.
Loss of classification LclassificationUsing cross-entropy representation, defined as:
wherein
And
the label value and the predicted value of the source domain.
S104: defining an initial population lambda1To lambda1And copying each individual in the population once to form an individual pair lambda1,λ1,λ2,λ2,λ3,λ3,…,λi,λi]Performing network iterative training using each individual pair in turn to obtain the same individual lambdaiThe deceleration speed of the two adjacent iteration loss functions is used as the fitness value of the individual to evaluate the current individual lambdaiPerformance of (d); searching the optimal position of the adaptive parameter lambda based on a differential evolution algorithm, and searching the optimal position of the adaptive parameter lambda based on the optimal adaptive weight parameter lambda and the domain adaptive loss LMMDAnd a classification loss LclassificationDefining a loss function as：
L＝Lclassification+λLMMD
Training the convolutional neural network by using a back propagation algorithm, updating parameters until the network converges, and finishing the training;
s105: and storing the trained network, and loading the target domain data set of the human brain medical image to a network model for testing.
According to the above process, after the algorithm is finished, the classification accuracy rate higher than that of the method using the static weight parameters is obtained, the classification and identification accuracy rate of the brain medical image is improved by about 5%, and the method is greatly improved in the field of transfer learning.
As mentioned above in S104, the back propagation algorithm training neural network is also an important process, and the back propagation algorithm training neural network is described in detail below.
Example 2
On the basis of embodiment 1, embodiment 2 of the present invention provides a method flow for training a neural network by using a back propagation algorithm. Referring to fig. 4, the method flow includes the following steps:
s201, initializing a variable of training times to 0;
s202, selecting a part of training data in a source domain data set, namely Batch processing (Batch);
s203, obtaining a predicted value of output through forward propagation;
s204, calculating loss, and updating various parameters of the neural network through a back propagation algorithm;
s205, judging whether the training expectation is reached, if so, jumping to S208, and if not, jumping to S201;
s206, judging whether the set training times are reached, if so, jumping to S208, and if not, jumping to S207;
s207, performing 1 adding operation on the training times, and jumping to S202;
and S208, finishing the training and finishing the iteration.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents and improvements made by those skilled in the art within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (7)
1. A medical image recognition method based on dynamic field adaptive learning is characterized in that: the method comprises the following steps:
s101: preprocessing is carried out on the basis of the image data set to obtain a source domain data set, and meanwhile, a near-infrared brain imager is used for collecting a small amount of brain medical image data to obtain a target domain data set;
s102: defining a convolutional neural network model for image recognition based on a source domain dataset, initializing each level parameter and training round number epochmaxAdding a domain self-adaptive layer before an output layer of a full connection layer in the convolutional neural network;
s103: calculating the domain-adaptive loss L of a convolutional neural networkMMDAnd a classification loss Lclassification；
S104: defining an initial population lambda1To lambdaiAnd searching the optimal position of the adaptive parameter lambda based on the differential evolution algorithm, and based on the optimal adaptive weight parameter lambda and the domain adaptive loss LMMDAnd a classification loss LclassificationTraining the convolutional neural network by using a back propagation algorithm, updating parameters until the network converges, and finishing the training;
s105: and substituting the target domain data set into the trained convolutional neural network to obtain the recognition output result of the brain medical image data, and testing.
2. The medical image recognition method based on dynamic domain adaptive learning according to claim 1, characterized in that: in S102, defining a convolutional neural network model for image recognition based on a source domain dataset includes:
and defining the total number of the neural network layers, including a convolutional layer and a fully-connected layer, defining the parameters of each layer of the convolutional layer and the parameters of the fully-connected layer, adding an adaptive layer after the last-but-one fully-connected layer, and calculating the maximum mean difference MMD between a source domain and a target domain through the output of the adaptive layer.
3. The medical image recognition method based on dynamic domain adaptive learning according to claim 1, characterized in that: in the S104, an adaptive loss L is constructed based on an adaptive weight parameter lambdaMMDAdjusting parameters of each layer of the convolutional neural network specifically comprises:
round number epoch based on preset population and population evolution stagesAcquiring an optimal adaptive weight parameter lambda by using a differential evolution algorithm;
the loss function is defined as: l ═ Lclassification+λLMMDWherein L represents the net ultimate loss LclassificationRepresents the regular classification loss of the network over the source domain, LMMDRepresents the adaptive loss of the network, here represented by the maximum mean difference MMD;
round number epoch of training phase based on preset neural networktAnd performing iterative training on the convolutional neural network by using a back propagation algorithm until the final loss L is converged, and stopping training.
4. The medical image recognition method based on dynamic domain adaptive learning according to claim 3, characterized in that: the method for acquiring the self-adaptive weight parameter lambda in the evolution stage specifically comprises the following steps:
initializing number and position [ lambda ] of original population1，λ2，λ3，…，λi]And copying each individual once to form an individual pair [ lambda ]1，λ1，λ2，λ2，λ3，λ3，…，λi，λi]；
Performing network iterative training by using each individual pair in turn to obtain the same individual lambdaiThe deceleration speed of the two adjacent iteration loss functions is used as an individual adaptive value to evaluate the current individual lambdaiPerformance of (d);
realizing individual variation through a differential evolution algorithm, taking a difference vector of two individuals randomly selected from a population as a random variation source of a third individual, weighting the difference vector, and carrying out vector synthesis with the third individual to generate a variant individual;
randomly selecting individuals through crossing, and carrying out parameter mixing on the variant individuals and a predetermined target individual to generate test individuals; if the adaptive value of the test individual is better than that of the target individual, replacing the target individual with the test individual in the next generation, otherwise, still storing the target individual;
and (4) performing iterative training, reserving good individuals in each generation, eliminating poor individuals, and guiding the search process to approach to the global optimal solution to obtain the optimal adaptive weight parameter lambda.
5. The medical image recognition method based on dynamic domain adaptive learning according to claim 1, characterized in that: in S104, the variation function of the differential evolution algorithm is defined as:
wherein r is1，r1And r1Is three random numbers with interval of [1, i]F is a scaling factor which is a determined constant; g represents the g-th.
6. The medical image recognition method based on dynamic domain adaptive learning according to claim 1, characterized in that: in S104, loss L is adaptively determined through the domainMMDAnd a classification loss LclassificationThe value of (2) completes the training, and the method comprises the following steps:
calculating the current loss L, i.e. the classification loss LclassificationMaximum mean difference MMD domain adaptive loss L of optimal adaptive weight parameter lambdaMMDSumming;
and when the loss is greater than the expected value, iterating by using a back propagation algorithm, and adjusting the neural network parameters until the loss value is stable and the network model converges.
7. The medical image recognition method based on dynamic domain adaptive learning according to claim 1, characterized in that: in S105, the target domain data set is a brain medical image data set, which specifically includes: storing the trained network model parameters; and (4) carrying out precision test on the network model by using the brain medical image data set, and outputting a test result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110832315.2A CN113658110A (en) | 2021-07-22 | 2021-07-22 | Medical image identification method based on dynamic field adaptive learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110832315.2A CN113658110A (en) | 2021-07-22 | 2021-07-22 | Medical image identification method based on dynamic field adaptive learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113658110A true CN113658110A (en) | 2021-11-16 |
Family
ID=78477685
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110832315.2A Pending CN113658110A (en) | 2021-07-22 | 2021-07-22 | Medical image identification method based on dynamic field adaptive learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113658110A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114494339A (en) * | 2021-12-29 | 2022-05-13 | 南京信息工程大学 | Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm |
| CN116524256A (en) * | 2023-04-24 | 2023-08-01 | 南通大学 | Brain MRI image classification method based on residual neural network and evolutionary algorithm |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106960002A (en) * | 2017-02-13 | 2017-07-18 | 上海大学 | A kind of cross-cutting information extraction method of feature based model |
| CN108414226A (en) * | 2017-12-25 | 2018-08-17 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning |
| CN109118446A (en) * | 2018-07-30 | 2019-01-01 | 西南财经大学 | A kind of underwater image restoration and denoising method |
| CN109655901A (en) * | 2017-10-11 | 2019-04-19 | 中国石油化工股份有限公司 | A kind of adaptive angle of polarization calculation method of frequency domain and system |
| CN110135510A (en) * | 2019-05-22 | 2019-08-16 | 电子科技大学中山学院 | Dynamic domain self-adaptive method, equipment and computer readable storage medium |
| AU2020103905A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning |
| CN112381178A (en) * | 2020-12-07 | 2021-02-19 | 西安交通大学 | Medical image classification method based on multi-loss feature learning |
| CN112509017A (en) * | 2020-11-18 | 2021-03-16 | 西北工业大学 | Remote sensing image change detection method based on learnable difference algorithm |
| CN112784879A (en) * | 2020-12-31 | 2021-05-11 | 前线智能科技(南京)有限公司 | Medical image segmentation or classification method based on small sample domain self-adaption |
-
2021
- 2021-07-22 CN CN202110832315.2A patent/CN113658110A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106960002A (en) * | 2017-02-13 | 2017-07-18 | 上海大学 | A kind of cross-cutting information extraction method of feature based model |
| CN109655901A (en) * | 2017-10-11 | 2019-04-19 | 中国石油化工股份有限公司 | A kind of adaptive angle of polarization calculation method of frequency domain and system |
| CN108414226A (en) * | 2017-12-25 | 2018-08-17 | 哈尔滨理工大学 | Fault Diagnosis of Roller Bearings under the variable working condition of feature based transfer learning |
| CN109118446A (en) * | 2018-07-30 | 2019-01-01 | 西南财经大学 | A kind of underwater image restoration and denoising method |
| CN110135510A (en) * | 2019-05-22 | 2019-08-16 | 电子科技大学中山学院 | Dynamic domain self-adaptive method, equipment and computer readable storage medium |
| CN112509017A (en) * | 2020-11-18 | 2021-03-16 | 西北工业大学 | Remote sensing image change detection method based on learnable difference algorithm |
| AU2020103905A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning |
| CN112381178A (en) * | 2020-12-07 | 2021-02-19 | 西安交通大学 | Medical image classification method based on multi-loss feature learning |
| CN112784879A (en) * | 2020-12-31 | 2021-05-11 | 前线智能科技(南京)有限公司 | Medical image segmentation or classification method based on small sample domain self-adaption |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114494339A (en) * | 2021-12-29 | 2022-05-13 | 南京信息工程大学 | Unmanned aerial vehicle target tracking method based on DAMDNet-EKF algorithm |
| CN116524256A (en) * | 2023-04-24 | 2023-08-01 | 南通大学 | Brain MRI image classification method based on residual neural network and evolutionary algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111967495B (en) | Classification recognition model construction method | |
| US11742076B2 (en) | Machine learning systems for generating multi-modal data archetypes | |
| CN111180068A (en) | Chronic disease prediction system based on multi-task learning model | |
| CN112381178B (en) | Medical image classification method based on multi-loss feature learning | |
| CN113658110A (en) | Medical image identification method based on dynamic field adaptive learning | |
| CN115147376A (en) | Skin lesion intelligent identification method based on deep Bayesian distillation network | |
| CN116884623B (en) | Medical rehabilitation prediction system based on laser scanning imaging | |
| Popkes et al. | Interpretable outcome prediction with sparse Bayesian neural networks in intensive care | |
| Alshari et al. | Machine learning model to diagnose diabetes type 2 based on health behavior | |
| CN115115038B (en) | Model construction method based on single lead electrocardiosignal and gender identification method | |
| CN116350234A (en) | ECG arrhythmia classification method and system based on GCNN-LSTM model | |
| CN116524284A (en) | Nuclear magnetic resonance image analysis method, system, device and storage medium | |
| CN113283465B (en) | Diffusion tensor imaging data analysis method and device | |
| Cao et al. | Alzheimer’s Disease Stage Detection Method Based on Convolutional Neural Network | |
| Zhang et al. | Pneumonia Detection Method based on Convolutional Neural Network | |
| Wyns et al. | Prediction of arthritis using a modified Kohonen mapping and case based reasoning | |
| CN114255234B (en) | Method for training model for identifying cardiovascular and cerebrovascular risks based on fundus images and related product | |
| Nafi'udin et al. | EFFICIENCY AND ACCURACY OF CONVOLUTIONAL AND FOURIER TRANSFORM LAYERS IN NEURAL NETWORKS FOR MEDICAL IMAGE CLASSIFICATION | |
| CN118820815A (en) | Indirect estimation method and system for reference interval of child inspection index | |
| CN118609814A (en) | Machine learning-based blood glucose concentration prediction method, prediction system, terminal equipment and storage medium | |
| Stainton et al. | On the application of quantization for mobile optimized convolutional neural networks as a predictor of realtime ageing biomarkers | |
| Naveena et al. | Effective skin cancer prediction using hybrid Xception and XGBoost approach tuned by honey badger optimization | |
| Tian et al. | Optimization of Bayesian algorithms for multi-threshold image segmentation | |
| CN114566241A (en) | Medical data processing method, device, equipment and medium | |
| CN117152517A (en) | Identification method and device for gestational diabetes based on fundus image |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20211116 |