CN113205484B - Mammary tissue classification and identification method based on transfer learning - Google Patents
Mammary tissue classification and identification method based on transfer learning Download PDFInfo
- Publication number
- CN113205484B CN113205484B CN202110367206.8A CN202110367206A CN113205484B CN 113205484 B CN113205484 B CN 113205484B CN 202110367206 A CN202110367206 A CN 202110367206A CN 113205484 B CN113205484 B CN 113205484B
- Authority
- CN
- China
- Prior art keywords
- pathological
- breast tissue
- image
- tissue section
- preprocessed
- 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
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
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine 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
- G06T2207/20081—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
- G06T2207/30068—Mammography; Breast
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Probability & Statistics with Applications (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention provides a breast tissue classification and identification method based on transfer learning. Firstly, obtaining pathological images of breast tissue slices of suspected patients through tissue slices, obtaining the preprocessed pathological images of the breast tissue slices through format adjustment and a series of image processing, and manually marking the pathological categories of the preprocessed images; constructing a migration learning mammary tissue classification model, sequentially inputting the preprocessed images into the model, predicting pathological categories through the model, constructing a loss function model by combining with actual pathological categories of the images, and performing optimization training to obtain an optimized migration learning mammary tissue classification model; and predicting the pathological category of the suspected patient through the optimized model, and combining the pathological category with the heat map generator to obtain a thermodynamic diagram corresponding to the evidence area. The method has the advantages of being suitable for an application environment with a small data set, enhancing robustness in training and improving reliability of results in prediction.
Description
Technical Field
The invention belongs to the field of medical image data processing, and particularly relates to a breast tissue classification and identification method based on transfer learning.
Background
The incidence of breast cancer is on the rise from the end of the 70 s in the 20 th century, China is not a high-incidence country of breast cancer, but the situation is still optimistic, and in recent decades, breast cancer gradually occupies the first place of the incidence of female cancer. According to the metastasis of cancer tissues in peripheral tissues, breast cancer is mainly classified into carcinoma in situ in which cancer cells have not spread and invasive carcinoma in which cancer cells have spread. Therefore, if the breast tissue can be accurately classified and identified, the breast cancer diagnosis method can help to judge whether the breast cancer is suffered from the breast cancer and the type of the breast cancer, so that early diagnosis and correct treatment can be achieved as far as possible, and the death rate of the breast cancer can be reduced to a certain extent.
The classification and identification work of breast tissues in the current medical detection is highly dependent on the experience judgment of imaging doctors, and the method has high accuracy but low efficiency. In response to such a problem, in recent years, many scholars have proposed different computer detection methods, such as machine detection methods based on ant colony algorithm, annealing algorithm, SVM algorithm, etc., which improve the efficiency of breast tissue detection to some extent. However, these methods have a narrow range of application, poor mobility, and strong dependency on the data amount, and cannot be used to advantage in an operation environment where the data set is small. In medical image processing, however, obtaining a database of large samples requires extremely high time and economic costs. In addition, these methods cannot indicate the basis of the obtained results, only can give predicted values, and are not strong in persuasion, so that these machine detection methods also have great limitations.
Disclosure of Invention
The invention aims to provide a breast tissue classification detection method based on transfer learning. In training, images are enhanced and augmented to expand the data set, inhibit overfitting, and thereby enhance robustness; in the prediction, the input image is subjected to image enhancement, after a result is obtained, a Grad-CAM algorithm is applied to generate a thermodynamic diagram, and a suspicious region (focus) is visualized, so that the reliability of the result is enhanced.
The technical scheme adopted by the invention is as follows: a breast tissue classification and identification method based on transfer learning comprises the following steps:
step 1: acquiring an original pathological image of a breast tissue section, and processing the original pathological image of the breast tissue section by a bilinear interpolation method to obtain a pathological image after the format of the breast tissue section is adjusted; carrying out image enhancement processing and image amplification processing on the pathological image after the breast tissue section format is adjusted to obtain a pathological image after the breast tissue section is preprocessed, and manually marking the pathological type of the pathological image after the breast tissue section is preprocessed;
step 2: constructing a migration learning mammary tissue classification model, sequentially inputting pathological images after pretreatment of mammary tissue slices as samples into the migration learning mammary tissue classification model, predicting through the migration learning mammary tissue classification model to obtain a predicted pathological category of the pathological images after pretreatment of the mammary tissue slices, constructing a loss function model by combining the pathological category of the pathological images after pretreatment of the mammary tissue slices, and further performing optimization training to obtain an optimized migration learning mammary tissue classification model;
and step 3: and (2) performing tissue slicing on the pathological image of the suspected patient to obtain a pathological image of the breast tissue slice of the suspected patient, performing image enhancement treatment and image amplification treatment in the step (1) to obtain a pathological image of the breast tissue slice of the suspected patient after preprocessing, further predicting a pathological category of the suspected patient through an optimized migration learning breast tissue classification model, and combining the pathological category of the suspected patient with a thermodynamic diagram corresponding to an evidence area obtained by constructing a heat map generator based on a Grad-CAM algorithm.
The pathological image after the breast tissue section format adjustment in the step 1 is a pathological image after the breast tissue section format adjustment in a certain standard format, and the label of the pathological image uses unique heat coding;
the image enhancement in the step 1 uses a Laplacian operator of 4 channels;
the limiting conditions of the image augmentation process in step 1 are as follows:
-π/2≤θ≤π/2
sin|θ|+cos|θ|≤r≤2
wherein, theta represents the rotation angle of the original pixel point relative to the original point during the rotation transformation in the image augmentation processing, and r represents the scaling size of the scaling transformation in the image augmentation processing;
the pathological image after the breast tissue section pretreatment in the step 1 is as follows: x is the number of k ，k∈[1,n]；
Wherein x is k Shows the k-th milkPathological images after the gland tissue section is preprocessed, wherein n is the number of the pathological images after the breast tissue section is preprocessed;
the pathological classification of the pathological image after the breast tissue section pretreatment in the step 1 is as follows: p is a radical of k ，k∈[1,n]；
Wherein p is k Representing the pathological category of the pathological images after the k piece of breast tissue section is preprocessed, wherein n is the number of the pathological images after the breast tissue section is preprocessed;
preferably, the migration learning breast tissue classification model in the step 2 is formed by sequentially cascading a feature extraction module, a full connection layer module and a classification module;
the feature extraction module is formed by sequentially cascading a plurality of convolution pooling modules;
the convolution pooling module is composed of a convolution layer and a pooling layer;
the characteristic extraction module is used for extracting the characteristics of the pathological image after the k piece of breast tissue slice is preprocessed;
the fully-connected layer module is used for carrying out nonlinear fitting processing on the characteristics of the pathological images after the k breast tissue slices are preprocessed to obtain the characteristic value of the fixed number of the pathological images after the k breast tissue slices are preprocessed, and the activating function of the fully-connected layer module adopts a Leaky ReLU function;
the classification module receives the characteristic value of the pathological image after the pretreatment of the kth mammary tissue slice of the full-connection layer module, and obtains a predicted initial value a of the fixed digit of the pathological image after the pretreatment of the kth mammary tissue slice by using linear regression k Then the predicted initial value a of the pathological image after the k-th breast tissue section is preprocessed k Bringing the predicted pathological category y into a softmax function to obtain a predicted initial value of the fixed digit of the pathological image after the pretreatment of the kth breast tissue section k The softmax function is expressed as follows:
wherein, a k Shows the pretreatment of the kth breast tissue sectionThe predicted initial value of the post-treatment pathological image, n is the total number of the predicted types, y k t And (4) the t element in the prediction vector of the pathological image after the k breast tissue section is preprocessed.
Step 2, the loss function model is as follows:
wherein x is k Representing the pathological images after the pretreatment of the kth mammary tissue section, n is the total number of the pathological images after the pretreatment of the mammary tissue section, and p k Representing the pathology category of the pathology image after the k-th breast tissue slice has been preprocessed, q k For the predicted pathology category of the pathology image after the k-th breast tissue slice preprocessing, c is the constant number [ 100 ] indicating the in situ tissue]；
Step 2, the method for further optimizing training to obtain the optimized migration learning mammary tissue classification model comprises the following steps:
and (4) obtaining an optimized migration learning mammary tissue classification model by using minimization of the loss function model as an optimization target through an Adam optimization algorithm.
Preferably, the step 3 of combining the pathological category of the suspected patient with the construction of the heatmap generator based on the Grad-CAM algorithm to obtain the thermodynamic diagram corresponding to the evidence area specifically includes:
calculating partial derivatives of prediction vectors output by the Softmax function of the optimized migration learning mammary tissue classification model in the step 2 on all pixels of the last convolution pooling layer of the optimized migration learning mammary tissue classification model, namely calculating partial derivatives
Wherein A is k And selecting a convolution pooling layer corresponding to the pathological image after the k piece of breast tissue section preprocessing. Meanwhile, for the gradient information obtained by the process, a global average is taken once for the partial derivative of each pixel of the feature map in each channel dimensionAll, we get the neuron importance weights as follows:
wherein Z is the total number of elements of the selected convolution pooling layer;
and (3) weighting the features of the last convolution pooling layer in the optimized migration learning mammary tissue classification model in the step (2) by using the neuron importance weight to obtain an initial thermodynamic diagram, and performing superposition operation to obtain a final thermodynamic diagram.
Compared with the prior art, the invention has the advantages that:
the invention uses the pre-training model for transfer learning, thereby effectively reducing the difficulty of constructing and training the neural network. In addition, the convolution-pooling layer of the pre-training model is used as a feature extractor for transfer learning, so that the morphological feature of the image can be captured conveniently, and the final accuracy is improved.
The invention carries out preprocessing on the input image, and uses image enhancement and image augmentation in the preprocessing process. In the image enhancement, a Laplacian operator of 4 channels is used for processing, so that the image characteristics are highlighted; in image augmentation, a series of restrictions are set on the scaling coefficient r and the central rotation coefficient theta, so that the augmented image has no black area and is not easy to lose the characteristic area.
The method reconstructs the loss function in the neural network training process, and effectively improves the accuracy of the model.
The invention performs visual analysis on the output result, provides the basis of the prediction result, ensures that the prediction result is more convincing, and fills the blank of visualization of the evidence area in the field of breast tissue classification prediction.
Drawings
FIG. 1: the invention relates to a flow chart of a breast tissue classification and identification system based on transfer learning.
FIG. 2: the invention is a schematic diagram of the neural network structure based on the transfer learning.
FIG. 3: is the image enhancement effect diagram of the invention.
FIG. 4: the image augmentation effect of the invention is compared with that of the traditional image augmentation effect.
FIG. 5: the effect of the method is compared with that of the traditional loss function on the basis of the VGG19 model in the training process.
FIG. 6: is the output thermodynamic diagram of the present invention.
Detailed Description
The invention will now be described in further detail with reference to the accompanying figures 1-6 and specific examples:
as shown in fig. 1, the implementation process of the present invention is as follows: a breast tissue classification and identification method based on transfer learning comprises the following steps:
step 1: acquiring an original pathological image of a breast tissue section, and processing the original pathological image of the breast tissue section by a bilinear interpolation method to obtain a pathological image after the format of the breast tissue section is adjusted; carrying out image enhancement processing and image amplification processing on the pathological image after the breast tissue section format is adjusted to obtain a pathological image after the breast tissue section is preprocessed, and manually marking the pathological type of the pathological image after the breast tissue section is preprocessed;
the pathological image after the breast tissue section format adjustment in the step 1 is a pathological image after the breast tissue section format adjustment in a certain standard format, and the label of the pathological image uses unique heat coding;
the image enhancement in step 1 uses a laplacian operator with 4 channels, and the enhancement effect is shown in fig. 2;
the augmentation result of the image augmentation process in step 1 is shown in fig. 3, and the limiting conditions are as follows:
-π/2≤θ≤π/2
sin|θ|+cos|θ|≤r≤2
wherein, θ ═ 1 denotes that the rotation angle of the original pixel point with respect to the origin is 1(rad) at the time of rotation conversion in the image augmentation processing, and r ═ 1 denotes that the scaling size of scaling conversion in the image augmentation processing is 1;
the pathological image after the breast tissue section pretreatment in the step 1 is as follows: x is the number of k ，k∈[1,n]；
Wherein x is k Representing the pathological images after the pretreatment of the kth mammary tissue section, wherein n is the number of the pathological images after the pretreatment of the mammary tissue section;
the pathological classification of the pathological image after the breast tissue section pretreatment in the step 1 is as follows: p is a radical of k ，k∈[1,n]；
Wherein p is k Representing the pathological category of the pathological images after the pretreatment of the kth breast tissue section, wherein n is the number of the pathological images after the pretreatment of the breast tissue section;
step 2: constructing a migration learning mammary tissue classification model, sequentially inputting pathological images after pretreatment of mammary tissue slices as samples into the migration learning mammary tissue classification model, predicting through the migration learning mammary tissue classification model to obtain a predicted pathological category of the pathological images after pretreatment of the mammary tissue slices, constructing a loss function model by combining the pathological category of the pathological images after pretreatment of the mammary tissue slices, and further performing optimization training to obtain an optimized migration learning mammary tissue classification model;
as shown in fig. 4, the migration learning breast tissue classification model in step 2 is formed by sequentially cascading a feature extraction module, a full connection layer module and a classification module;
the feature extraction module is formed by sequentially cascading a plurality of convolution pooling modules;
the convolution pooling module is composed of a convolution layer and a pooling layer;
the characteristic extraction module is used for extracting the characteristics of the pathological image after the k piece of breast tissue slice is preprocessed;
the fully-connected layer module is used for carrying out nonlinear fitting processing on the characteristics of the pathological images after the k breast tissue slices are preprocessed to obtain the characteristic value of the fixed number of the pathological images after the k breast tissue slices are preprocessed, and the activating function of the fully-connected layer module adopts a Leaky ReLU function;
the classification module receives the characteristic value of the pathological image after the pretreatment of the kth mammary tissue slice of the full-connection layer module, and the prediction initial of the fixed digit of the pathological image after the pretreatment of the kth mammary tissue slice is obtained by linear regressionValue a k Then the predicted initial value a of the pathological image after the k-th breast tissue section is preprocessed k Bringing the predicted pathological category y into a softmax function to obtain a predicted initial value of the fixed digit of the pathological image after the pretreatment of the kth breast tissue section k The softmax function is expressed as follows:
wherein, a k Representing the predicted initial value of the pathological image after the k-th breast tissue section is preprocessed, n is the total number of the predicted types, y k t And (4) the t element in the prediction vector of the pathological image after the k breast tissue section is preprocessed.
Step 2, the loss function model is as follows:
wherein x is k Representing the pathological images after the pretreatment of the kth mammary tissue section, n is the total number of the pathological images after the pretreatment of the mammary tissue section, and p k Representing the pathology category of the pathology image after the k-th breast tissue slice has been preprocessed, q k For the predicted pathology category of the pathology image after the k-th breast tissue slice preprocessing, c is the constant number [ 100 ] indicating the in situ tissue]；
Step 2, the method for further optimizing training to obtain the optimized migration learning mammary tissue classification model comprises the following steps:
and (4) obtaining an optimized migration learning mammary tissue classification model by using minimization of the loss function model as an optimization target through an Adam optimization algorithm. The model behaves during training as shown in SVGG in fig. 5 (left), the original transfer learning model behaves during training as shown in VGG + in fig. 5 (left), and fig. 5 (right) shows the variance levels of the two models respectively on the training set and the test set.
And step 3: the pathological image of the suspected patient is subjected to tissue slicing to obtain a pathological image of the breast tissue slice of the suspected patient, the pathological image of the breast tissue slice of the suspected patient after preprocessing is obtained through the image enhancement processing and the image amplification processing in the step 1, the pathological category of the suspected patient is further obtained through predicting a migration learning breast tissue classification model after optimization, the pathological category of the suspected patient is combined with a thermodynamic diagram generator constructed based on a Grad-CAM algorithm to obtain a thermodynamic diagram corresponding to an evidence area, and the effect is shown in FIG. 6.
Step 3, combining the pathological category of the suspected patient with a heatmap generator constructed based on a Grad-CAM algorithm to obtain a heatmap corresponding to the evidence area, specifically:
calculating partial derivatives of prediction vectors output by the Softmax function of the optimized migration learning mammary tissue classification model in the step 2 on all pixels of the last convolution pooling layer of the optimized migration learning mammary tissue classification model, namely calculating partial derivatives
Wherein, A k And selecting a convolution pooling layer corresponding to the pathological image after the k piece of breast tissue section preprocessing. Meanwhile, for the gradient information obtained in the process, taking a global average of partial derivatives of each pixel of the feature map on each channel dimension to obtain neuron importance weight, wherein the weight is as follows:
wherein Z is the total number of elements of the selected convolution pooling layer;
and (3) weighting the features of the last convolution pooling layer in the optimized migration learning mammary tissue classification model in the step (2) by using the neuron importance weight to obtain an initial thermodynamic diagram, and performing superposition operation to obtain a final thermodynamic diagram.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (3)
1. A breast tissue classification and identification method based on transfer learning is characterized by comprising the following steps:
step 1: acquiring an original pathological image of a breast tissue section, and processing the original pathological image of the breast tissue section by a bilinear interpolation method to obtain a pathological image after the format of the breast tissue section is adjusted; carrying out image enhancement processing and image amplification processing on the pathological image after the breast tissue section format is adjusted to obtain a pathological image after the breast tissue section is preprocessed, and manually marking the pathological type of the pathological image after the breast tissue section is preprocessed;
step 2: constructing a migration learning mammary tissue classification model, sequentially inputting pathological images after pretreatment of mammary tissue slices as samples into the migration learning mammary tissue classification model, predicting through the migration learning mammary tissue classification model to obtain a predicted pathological category of the pathological images after pretreatment of the mammary tissue slices, constructing a loss function model by combining the pathological category of the pathological images after pretreatment of the mammary tissue slices, and further performing optimization training to obtain an optimized migration learning mammary tissue classification model;
and step 3: performing tissue slicing on the pathological image of the suspected patient to obtain a pathological image of the breast tissue slice of the suspected patient, obtaining the pathological image of the breast tissue slice of the suspected patient after preprocessing by the image enhancement processing and the image amplification processing in the step 1, further obtaining the pathological category of the suspected patient by predicting a migration learning breast tissue classification model after optimization, and combining the pathological category of the suspected patient with a thermodynamic diagram corresponding to an evidence area obtained by constructing a heat map generator based on a Grad-CAM algorithm;
step 2, the transfer learning mammary tissue classification model is formed by sequentially cascading a feature extraction module, a full-connection layer module and a classification module;
the feature extraction module is formed by sequentially cascading a plurality of convolution pooling modules;
the convolution pooling module is composed of a convolution layer and a pooling layer;
the characteristic extraction module is used for extracting the characteristics of the pathological image after the k piece of breast tissue slice is preprocessed;
the fully-connected layer module is used for carrying out nonlinear fitting processing on the characteristics of the pathological images after the k breast tissue slices are preprocessed to obtain the characteristic value of the fixed number of the pathological images after the k breast tissue slices are preprocessed, and the activating function of the fully-connected layer module adopts a Leaky ReLU function;
the classification module receives the characteristic value of the pathological image after the pretreatment of the kth mammary tissue slice of the full-connection layer module, and obtains a predicted initial value a of the fixed digit of the pathological image after the pretreatment of the kth mammary tissue slice by using linear regression k Then the predicted initial value a of the pathological image after the k-th breast tissue section is preprocessed k Bringing the predicted pathological category y into a softmax function to obtain a predicted initial value of the fixed digit of the pathological image after the pretreatment of the kth breast tissue section k The softmax function is expressed as follows:
wherein, a k Representing the predicted initial value of the pathological image after the k-th breast tissue section is preprocessed, n' is the number of the predicted types, y k t The t element in the prediction vector of the pathological image after the k breast tissue section is preprocessed is taken as the element;
step 2, the loss function model is as follows:
wherein n is the total number of pathological images after the pretreatment of the breast tissue section, p k Representing the pathological image after the k-th breast tissue section preprocessingPathological class of (1), p k T Is p k Transpose of (q) k Predicting the pathological type of the pathological image after the k-th breast tissue section is preprocessed, wherein c is a constant number indicating in-situ tissue;
step 2, the method for further optimizing training to obtain the optimized migration learning mammary tissue classification model comprises the following steps:
and (4) obtaining an optimized migration learning mammary tissue classification model by using minimization of the loss function model as an optimization target through an Adam optimization algorithm.
2. The breast tissue classification and identification method based on transfer learning according to claim 1,
the pathological image after the breast tissue section format adjustment in the step 1 is a pathological image after the breast tissue section format adjustment in a certain standard format, and the label of the pathological image uses unique heat coding;
the image enhancement in the step 1 uses a Laplacian operator of 4 channels;
the limiting conditions of the image augmentation process in step 1 are as follows:
-π/2≤θ≤π/2
sin|θ|+cos|θ|≤r≤2
wherein, theta represents the rotation angle of the original pixel point relative to the original point during the rotation transformation in the image augmentation processing, and r represents the scaling size of the scaling transformation in the image augmentation processing;
the pathological image after the breast tissue section pretreatment in the step 1 is as follows: x is the number of k ，k∈[1,n]；
Wherein x is k Representing the pathological images after the pretreatment of the kth mammary tissue section, wherein n is the number of the pathological images after the pretreatment of the mammary tissue section;
the pathological classification of the pathological image after the breast tissue section pretreatment in the step 1 is as follows: p is a radical of k ，k∈[1,n]；
Wherein p is k Representing the pathological category of the pathological images after the k-th breast tissue section is preprocessed, and n is the number of the pathological images after the breast tissue section is preprocessed.
3. The breast tissue classification and identification method based on transfer learning according to claim 1,
step 3, combining the pathological category of the suspected patient with a heatmap generator constructed based on a Grad-CAM algorithm to obtain a thermodynamic diagram corresponding to the evidence area, specifically:
calculating partial derivatives of prediction vectors output by the Softmax function of the optimized migration learning mammary tissue classification model in the step 2 on all pixels of the last convolution pooling layer of the optimized migration learning mammary tissue classification model, namely calculating partial derivatives
Wherein A is k Selecting a convolution pooling layer corresponding to the pathological image after the k-th breast tissue slice preprocessing; meanwhile, for the gradient information obtained in the process, taking a global average of partial derivatives of each pixel of the feature map on each channel dimension to obtain neuron importance weight, wherein the weight is as follows:
wherein Z is the total number of elements of the selected convolution pooling layer;
and (3) weighting the features of the last convolution pooling layer in the optimized migration learning mammary tissue classification model in the step (2) by using the neuron importance weight to obtain an initial thermodynamic diagram, and performing superposition operation to obtain a final thermodynamic diagram.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110367206.8A CN113205484B (en) | 2021-04-06 | 2021-04-06 | Mammary tissue classification and identification method based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110367206.8A CN113205484B (en) | 2021-04-06 | 2021-04-06 | Mammary tissue classification and identification method based on transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113205484A CN113205484A (en) | 2021-08-03 |
| CN113205484B true CN113205484B (en) | 2022-08-05 |
Family
ID=77026233
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110367206.8A Active CN113205484B (en) | 2021-04-06 | 2021-04-06 | Mammary tissue classification and identification method based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113205484B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113838558B (en) * | 2021-08-16 | 2023-04-18 | 电子科技大学 | Method and device for analyzing breast cancer pathological image based on convolutional neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635835A (en) * | 2018-11-08 | 2019-04-16 | 深圳蓝韵医学影像有限公司 | A kind of breast lesion method for detecting area based on deep learning and transfer learning |
| CN109766277A (en) * | 2019-01-02 | 2019-05-17 | 北京航空航天大学 | A kind of software fault diagnosis method based on transfer learning and DNN |
| WO2021016131A1 (en) * | 2019-07-19 | 2021-01-28 | The Jackson Laboratory | Convolutional neural networks for classification of cancer histological images |
-
2021
- 2021-04-06 CN CN202110367206.8A patent/CN113205484B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635835A (en) * | 2018-11-08 | 2019-04-16 | 深圳蓝韵医学影像有限公司 | A kind of breast lesion method for detecting area based on deep learning and transfer learning |
| CN109766277A (en) * | 2019-01-02 | 2019-05-17 | 北京航空航天大学 | A kind of software fault diagnosis method based on transfer learning and DNN |
| WO2021016131A1 (en) * | 2019-07-19 | 2021-01-28 | The Jackson Laboratory | Convolutional neural networks for classification of cancer histological images |
Non-Patent Citations (2)
| Title |
|---|
| 《Multi-label transfer learning for the early diagnosis of breast cancer》;Hiba Chougrad，et al;《Neurocomputing》;20200607;第392卷;第168-180页 * |
| 《基于迁移学习的多视角乳腺肿块和钙化簇分类方法》;肖禾，等;《计算机应用》;20200510;第40卷(第5期);第1460-1464页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113205484A (en) | 2021-08-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107886514B (en) | Mammary gland molybdenum target image lump semantic segmentation method based on depth residual error network | |
| CN110390351B (en) | Epileptic focus three-dimensional automatic positioning system based on deep learning | |
| CN109886986B (en) | Dermatoscope image segmentation method based on multi-branch convolutional neural network | |
| CN109191476A (en) | The automatic segmentation of Biomedical Image based on U-net network structure | |
| CN112733950A (en) | Power equipment fault diagnosis method based on combination of image fusion and target detection | |
| CN111951288B (en) | Skin cancer lesion segmentation method based on deep learning | |
| CN111325750B (en) | Medical image segmentation method based on multi-scale fusion U-shaped chain neural network | |
| CN113378791B (en) | Cervical cell classification method based on double-attention mechanism and multi-scale feature fusion | |
| CN112001894B (en) | Thyroid gland boundary smoothness detection device | |
| CN111161271A (en) | Ultrasonic image segmentation method | |
| CN112085742B (en) | NAFLD ultrasonic video diagnosis method based on context attention | |
| CN115965602A (en) | Abnormal cell detection method based on improved YOLOv7 and Swin-Unet | |
| CN113782184A (en) | Cerebral apoplexy auxiliary evaluation system based on facial key point and feature pre-learning | |
| CN112614093A (en) | Breast pathology image classification method based on multi-scale space attention network | |
| CN113205484B (en) | Mammary tissue classification and identification method based on transfer learning | |
| CN110766678A (en) | Ovarian cancer molecular typing prediction system | |
| Zhang et al. | Multi-scale aggregation networks with flexible receptive fields for melanoma segmentation | |
| CN116704188A (en) | Wheat grain image segmentation algorithm with different volume weights based on improved U-Net network | |
| CN111554384A (en) | Adenocarcinoma pathological image analysis method based on prior perception and multitask learning | |
| CN113870194B (en) | Breast tumor ultrasonic image processing device with fusion of deep layer characteristics and shallow layer LBP characteristics | |
| CN112001896B (en) | Thyroid gland border irregularity detection device | |
| CN113192076B (en) | MRI brain tumor image segmentation method combining classification prediction and multi-scale feature extraction | |
| CN115578400A (en) | Image processing method, and training method and device of image segmentation network | |
| CN117765410B (en) | Remote sensing image double-branch feature fusion solid waste identification method and system and electronic equipment | |
| CN117975170B (en) | Medical information processing method and system based on big data |
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 |