CN110472694A - A kind of Lung Cancer Images pathological classification method and device - Google Patents

Abstract

The present application discloses a lung cancer image pathological classification method and device, including: firstly, extracting the color features of a preset RGB three-dimensional lesion image data set to obtain a feature vector set; The coefficient matrix is obtained by non-negative matrix decomposition; then, the coefficient matrix is divided into a training data set and a test data set as a sample data set, and the first real pathological label of each training data and the first true pathological label of each test data are respectively obtained. Two real pathological labels; then, use the training data set and the first real pathological label to classify the classifier to obtain a trained classifier; finally, input the test data set to the completed training Classification is carried out in the classifier, and the predicted pathological label and confidence distance are obtained. Resolved a technical issue with the long waiting time for a lung cancer diagnosis.

Description

A lung cancer image pathological classification method and device

technical field

The present application relates to the technical field of image processing, in particular to a method and device for pathological classification of lung cancer images.

Background technique

Lung cancer is the most common visceral malignant tumor with high morbidity and mortality. Early detection and diagnosis are crucial to the health of patients. Lung cancer can be divided into invasive carcinoma, microinvasive carcinoma, and carcinoma in situ according to subcategories. Microinvasive carcinoma and carcinoma in situ have the same surgical methods in the diagnosis and treatment process and can be classified as non-invasive carcinoma.

The traditional diagnosis method of lung cancer mainly relies on the preliminary examination and evaluation of chest imaging such as computed tomography (CT) by experienced doctors, and the determination of the subcategory of lung cancer must be obtained by pathologists observing the characteristics of living tissue samples under a microscope , after the doctor removes the patient's diseased tissue through surgery, it is usually sent to the pathology department for frozen section and paraffin section, but this method needs to wait for about 30 minutes to get the result, which is time-consuming.

Contents of the invention

This application provides a lung cancer image pathological classification method, device and equipment, which are used to solve the problem that the results can only be obtained from live frozen sections and paraffin sections made after doctors observe the lesion tissue obtained from the patient under a microscope, which leads to waiting Lung cancer diagnosis results in longer time technical issues.

In view of this, the present application provides a method for pathological classification of lung cancer images from the first aspect, including:

Extract the color features of the preset RGB three-dimensional lesion image data set to obtain a feature vector set;

Performing non-negative matrix decomposition on the matrix composed of the eigenvector set by column to obtain a coefficient matrix;

dividing the coefficient matrix as a sample data set into a training data set and a test data set, and obtaining the first real pathological label of each training data and the second real pathological label of each test data;

Using the training data set and the first true pathological label to perform classification training on a classifier to obtain a trained classifier;

The test data set is input into the trained classifier for classification, and a predicted pathological label and a confidence distance are obtained, wherein the confidence distance is a probability value corresponding to the predicted case label.

Preferably, the extraction of color features of preset RGB three-dimensional lesion data sets to obtain feature vectors includes:

Based on the color histogram, the color features of the preset RGB three-dimensional lesion data set are extracted to obtain the feature vector.

Preferably, there are multiple classifiers.

Preferably, it also includes:

AUC is calculated according to the second true pathological label, the predicted pathological label and the confidence distance, and the classification effects of different classifiers are evaluated.

Preferably, the extraction of the color features of the preset RGB three-dimensional lesion image data set to obtain the feature vector also includes:

The first screenshot processing is performed on the pathological region of lung cancer in the RGB three-dimensional lesion image sample to obtain the RGB three-dimensional lesion image data set.

Preferably, the extraction of the color features of the preset RGB three-dimensional lesion image data set to obtain the feature vector also includes:

A first scaling process is performed on the RGB three-dimensional lesion image samples to obtain the RGB three-dimensional lesion image data set.

Preferably, the extraction of the color features of the preset RGB three-dimensional lesion image data set to obtain the feature vector also includes:

After the second scaling process is performed on the RGB three-dimensional lesion image sample, a second screenshot process is performed on the lung cancer pathological area of the RGB three-dimensional lesion image sample after the second scaling process to obtain the RGB three-dimensional lesion image data set.

The present application provides a lung cancer image pathological classification device from the second aspect, including: a feature extraction module, a decomposition module, a processing module, a training module, and a testing module;

The feature extraction module is used to extract the color features of the preset RGB three-dimensional lesion image data set to obtain a feature vector set;

The decomposition module performs non-negative matrix decomposition on the matrix composed of the eigenvector set by column to obtain a coefficient matrix;

The processing module divides the coefficient matrix as a sample data set into a training data set and a test data set, and obtains the first real pathological label of each training data and the second real pathological label of each test data;

The training module uses the training data set and the first true label to classify and train the classifier to obtain a trained classifier;

The test module is configured to input the test data set into the trained classifier for classification to obtain the predicted pathological label and confidence distance, wherein the confidence distance is the probability value corresponding to the predicted case label.

Preferably, it also includes: an evaluation module;

The evaluation module calculates AUC according to the second real pathological label, the predicted pathological label and the confidence distance, and evaluates the classification effects of different classifiers.

Preferably, it also includes: an enhancement module;

The RGB three-dimensional lesion image sample is enhanced by at least one image enhancement method to obtain an RGB three-dimensional lesion image data set. The image enhancement method includes: first screenshot enhancement, first zoom enhancement, second zoomed second Screenshot enhancements.

As can be seen from the above technical solutions, the present application has the following advantages:

A method for pathological classification of lung cancer images provided in this application, first, extracting the color features of the preset RGB three-dimensional lesion image data set to obtain a feature vector set; secondly, performing a non-negative matrix on the matrix composed of the feature vector set by column Decompose to obtain a coefficient matrix; then, divide the coefficient matrix as a sample data set into a training data set and a test data set, and obtain the first real pathology label of each training data and the second real pathology label of each test data respectively label; then, use the training data set and the first true pathological label to classify the classifier to obtain a trained classifier; finally, input the test data set into the trained classifier Classification is performed to obtain the predicted pathological label and the confidence distance, where the confidence distance is the probability value corresponding to the predicted case label. The pathological classification method of lung cancer images provided by this application is more convenient and faster than extracting the texture and structural features of lung cancer images by extracting the color features of lung cancer images. The use of non-negative matrix factorization matrix can effectively reduce the dimension of the matrix, thereby reducing the amount of calculation. The classification speed of the trained classifier is relatively fast, so this application solves the existing problem of living frozen sections and paraffin sections made after the physician observes the diseased tissue obtained from the patient's body under a microscope in three aspects. Technical issues that lead to longer wait times for lung cancer diagnosis results.

Description of drawings

FIG. 1 is a schematic flow chart of Embodiment 1 of a lung cancer image pathological classification method of the present application;

FIG. 2 is a schematic flow diagram of Embodiment 2 of a lung cancer image pathological classification method of the present application;

FIG. 3 is a schematic structural diagram of an embodiment of a lung cancer image pathology classification device of the present application;

Detailed ways

In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiment of the application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the application. Obviously, the described embodiment is only It is a part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

This application provides a pathological classification method for lung cancer images. First, the color features of the preset RGB three-dimensional lesion image data set are extracted to obtain the feature vector set; secondly, the matrix composed of the feature vector set by column is subjected to non-negative matrix decomposition to obtain coefficient matrix; then, divide the coefficient matrix as a sample data set into a training data set and a test data set, and obtain the first real pathological label of each training data and the second real pathological label of each test data respectively; then, use The training data set and the first true pathological label are used to classify and train the classifier to obtain the trained classifier; finally, the test data set is input into the trained classifier for classification, and the predicted pathological label and confidence distance are obtained. Among them, The confidence distance is the probability value corresponding to the predicted case label. The pathological classification method of lung cancer images provided by this application is more convenient and faster than extracting the texture and structural features of lung cancer images by extracting the color features of lung cancer images. The use of non-negative matrix factorization matrix can effectively reduce the dimension of the matrix, thereby reducing the amount of calculation. The classification speed of the trained classifier is relatively fast, so this application solves the existing problem of living frozen sections and paraffin sections made after the physician observes the diseased tissue obtained from the patient's body under a microscope in three aspects. Technical issues that lead to longer wait times for lung cancer diagnosis results.

For ease of understanding, please refer to Figure 1. The present application provides a first embodiment of a method for pathological classification of lung cancer images, including:

Step 101, extracting color features of a preset RGB three-dimensional lesion image data set to obtain a set of feature vectors.

Among them, the color feature is one of the most representative features of RGB three-dimensional images, and the extracted features are in the form of vectors. Each RGB three-dimensional lesion image can extract a feature vector, and the extracted feature vectors are all 256×1 dimensional.

Step 102, perform non-negative matrix decomposition on the matrix composed of eigenvector sets by columns to obtain a coefficient matrix.

It should be noted that all the obtained feature vector sets are combined into a 256×N-dimensional matrix by column, where N is the total amount of RGB three-dimensional lesion images, and then the matrix is decomposed by the non-negative matrix decomposition method to obtain the coefficient matrix , the non-negative matrix factorization method can effectively reduce the dimension of the matrix by decomposing the matrix, thereby reducing the amount of calculation.

Step 103: Divide the coefficient matrix as a sample data set into a training data set and a test data set, and obtain the first real pathological label of each training data and the second real pathological label of each test data respectively.

It should be noted that the coefficient matrix is split by column as a sample data set, and each column is used as a sample data, and all sample data are divided into training data set and test data set in proportion, and the first training data set is made for each training data The real pathology label is to create a second real pathology label for each test data. The real pathology label in the embodiment of the present application refers to the real pathology category of the sample data.

Step 104: Use the training data set and the first true pathological label to perform classification training on the classifier to obtain a trained classifier.

It should be noted that the training data set is in vector format, and the first real pathological label is the real category of the training sample data. The training data set and the first real pathological label are input into the classifier, and the training classification makes the classification of the classifier more accurate. targeted.

Step 105, input the test data set into the trained classifier for classification, and obtain the predicted pathological label and confidence distance.

Among them, the confidence distance is the probability value corresponding to the predicted case label.

It should be noted that the test data set is in vector format, and the test data set is input into the trained classifier for classification, and the predicted pathological label can be obtained quickly.

Extracting the color features of lung cancer images is more convenient and faster than extracting the texture and structural features of lung cancer images. The use of non-negative matrix factorization matrix can effectively reduce the dimension of the matrix, thereby reducing the amount of calculation, and the classification speed of the trained classifier is faster. Fast, so, this application solves the existing problem that the results can only be obtained by living frozen sections and paraffin sections made by doctors observing the diseased tissues obtained from the patient under a microscope from three aspects, resulting in a long waiting time for the diagnosis of lung cancer. technical problem.

For ease of understanding, please refer to FIG. 2. The present application provides a second embodiment of a method for pathological classification of lung cancer images, including:

Step 201, using an image enhancement method to perform enhancement processing on the RGB three-dimensional lesion image samples to obtain an RGB three-dimensional lesion image data set.

It should be noted that, according to the quality and quantity of the acquired RGB three-dimensional lesion image, different image enhancement methods can be used to enhance the image, and the image enhancement method can be directly taking the first screenshot of the lesion area in the RGB three-dimensional lesion image; The first scaling method can be used to ensure the integrity of the RGB three-dimensional lesion image as much as possible, while taking into account the morphological specificity of the RGB three-dimensional lesion image lesion. At the same time, the second scaling can be further performed on the part of the RGB three-dimensional lesion image lesion. Two screenshots.

Step 202, extracting the color features of the RGB three-dimensional lesion image data set to obtain a feature vector set.

It should be noted that the color features of the RGB three-dimensional lesion data set are extracted based on the color histogram, and the feature vector is obtained. First, the RGB image is converted into a color composed of hue (H), saturation (S), and lightness (V). Space HSV; secondly, quantize the H component to 16 levels, quantize the S component and V component to 4 levels respectively; finally, synthesize the three color components into a one-dimensional feature vector, calculate its histogram distribution, and output a 256-dimensional column vector .

Step 203, perform non-negative matrix decomposition on the matrix composed of eigenvector sets by column to obtain a coefficient matrix.

It should be noted that all the column vectors representing the characteristics of the RGB three-dimensional lesion image are composed into a matrix by column, and then the matrix is decomposed by the non-negative matrix decomposition method to obtain the base matrix and coefficient matrix, and obtain the coefficient matrix.

Among them, the non-negative matrix factorization process is:

First, the mathematical formula for non-negative matrix factorization is known:

V≈WH

Among them, V is the original matrix, W is the base matrix, and H is the coefficient matrix;

The non-negative matrix factorization objective function can be obtained according to the non-negative matrix factorization mathematical formula:

The partial derivatives of the base matrix and the coefficient matrix are respectively calculated for the objective function, and then the iterative formula of the base matrix W and the coefficient matrix H is solved by the gradient descent method:

The output coefficient matrix can be obtained according to the preset number of iterations.

Step 204: Divide the coefficient matrix as a sample data set into a training data set and a testing data set, and obtain the first real pathological label of each training data and the second real pathological label of each testing data respectively.

It should be noted that, firstly, the coefficient matrix is split into column vectors by columns, and each column is used as a sample data, and all the sample data obtained are divided into training data sets and test data sets according to preset ratios, and for each sample data Obtain the sample label of the response, that is, the pathological category to which it belongs. At this time, the obtained label is the real label of the sample.

Step 205: Classify and train the SVM classifier with the training data set and the first true pathological label, and obtain a trained SVM classifier.

It should be noted that the training data set is a training column vector set, which can be input into the SVM for training, which can improve the recognition accuracy of the SVM classifier for pathological images, and can obtain a targeted SVM classifier.

Step 206, input the test data set into the trained SVM classifier for classification, and obtain the predicted pathological label and confidence distance.

Among them, the confidence distance is the probability value corresponding to the predicted case label.

It should be noted that, in addition to the SVM classifier, the classifier in this embodiment can also be other classifiers, which also does not affect the method to solve the technical problems raised in this application. Therefore, when executing step 201 of the second embodiment of this application After step 204, the training data set and the first true pathological label can be input into at least two classifiers, and the classifiers are trained to obtain different trained classifiers; then, the test data set is input into different training complete The classifier is used to classify, and multiple sets of predicted pathological labels and confidence distances are obtained. The second real pathological labels, predicted pathological labels, and confidence distances are used to calculate AUC (area under the ROC curve), and the classification effects of different classifiers are evaluated.

It should be noted that the sample category can be determined by using the confidence distance as the threshold. The sample categories mainly include: true positive (TP): judged as positive, but actually positive; false positive (FP): judged as negative, but actually positive; true positive (TP): Negative (TN): It is judged to be negative, but it is actually negative; False Negative (FN): It is judged to be negative, but it is actually positive; then the true positive probability and false positive probability can be obtained according to the second real pathological label and predicted pathological label; through The true positive probability and the false positive probability draw the ROC curve, so that the area under the ROC curve can be obtained, that is, AUC. As a value, AUC can clearly and intuitively reflect the classification effect of different classifiers, and at the same time objectively evaluate the classification accuracy of unbalanced samples. According to the evaluation of different classifiers by AUC, the classifier with better classification effect can be selected.

For ease of understanding, please refer to FIG. 3 , the present application provides a lung cancer image pathology classification device, including: an enhancement module 301, a feature extraction module 302, a decomposition module 303, a processing module 304, a training module 305, a testing module 306, and an evaluation module 307.

The enhancement module 301 uses an image enhancement method to perform enhancement processing on RGB three-dimensional lesion image samples, and obtains an RGB three-dimensional lesion image data set.

The feature extraction module 302 is configured to extract color features of a preset RGB three-dimensional lesion image data set to obtain a set of feature vectors.

Decomposition module 303 performs non-negative matrix decomposition on the matrix composed of eigenvector sets by columns to obtain a coefficient matrix.

The processing module 304 divides the coefficient matrix as a sample data set into a training data set and a test data set, and acquires the first real pathological label of each training data and the second real pathological label of each test data respectively.

The training module 305 uses the training data set and the first real label to perform classification training on the classifier to obtain a trained classifier.

The test module 306 is configured to input the test data set into the trained classifier for classification to obtain the confidence distance of the predicted pathological label, wherein the confidence distance is the probability value corresponding to the predicted case label.

The evaluation module 307 calculates AUC according to the second real pathological label, the predicted pathological label and the confidence distance, and evaluates the classification effect of different classifiers.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods may be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or modules may be in electrical, mechanical or other forms.

The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or may be distributed to multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional module in each embodiment of the present application may be integrated into one processing module, each module may exist separately physically, or two or more modules may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software function modules.

If the integrated modules are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions for executing all or part of the steps of the methods described in the various embodiments of the present application through a computer device (which may be a personal computer, a server, or a network device, etc.). The aforementioned storage media include: U disk, mobile hard disk, read-only memory (English full name: Read-OnlyMemory, English abbreviation: ROM), random access memory (English full name: Random Access Memory, English abbreviation: RAM), disk Or various media such as CDs that can store program codes.

As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still understand the foregoing The technical solutions described in each embodiment are modified, or some of the technical features are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the application.

Claims (10)

1. a kind of Lung Cancer Images pathological classification method characterized by comprising

The color character for extracting preset RGB three-dimensional lesions image data set, obtains set of eigenvectors；

Non-negative Matrix Factorization will be carried out by the matrix that column form by described eigenvector collection, obtains coefficient matrix；

It is divided into training dataset and test data set using the coefficient matrix as sample data set, and obtains each instruction respectively Practice the first true pathology label of data and the second true pathology label of each test data；

Classification based training is carried out to classifier with the training dataset and the described first true pathology label, obtains training completion Classifier；

The test data set is input in the classifier that the training is completed and is classified, prediction pathology label is obtained and is set Communication distance, wherein Confidence distance is the corresponding probability value of prediction case label.

2. Lung Cancer Images pathological classification method according to claim 1, which is characterized in that described to extract preset RGB three-dimensional The color character of lesion data collection, obtains feature vector, comprising:

The color character that preset RGB three-dimensional lesions data set is extracted based on color histogram, obtains feature vector.

3. Lung Cancer Images pathological classification method according to claim 1, which is characterized in that the quantity of the classifier is more It is a.

4. Lung Cancer Images pathological classification method according to claim 3, which is characterized in that further include:

AUC is calculated according to the described second true pathology label, the prediction pathology label and the Confidence distance, evaluation is different The classifying quality of classifier.

5. Lung Cancer Images pathological classification method according to claim 1, which is characterized in that described to extract preset RGB three-dimensional The color character of lesion image data set obtains feature vector, before further include:

First screenshot processing is carried out to the lung cancer pathological regions in RGB three-dimensional lesions image pattern, obtains the RGB three-dimensional lesions Image data set.

6. Lung Cancer Images pathological classification method according to claim 1, which is characterized in that described to extract preset RGB three-dimensional The color character of lesion image data set obtains feature vector, before further include:

First scaling processing is carried out to RGB three-dimensional lesions image pattern, obtains the RGB three-dimensional lesions image data set.

7. Lung Cancer Images pathological classification method according to claim 1, which is characterized in that described to extract preset RGB three-dimensional The color character of lesion image data set obtains feature vector, before further include:

It is three-dimensional to the RGB after second scaling processing after carrying out the second scaling processing to RGB three-dimensional lesions image pattern The lung cancer pathological regions of lesion image pattern carry out the second screenshot processing, obtain the RGB three-dimensional lesions image data set.

8. a kind of Lung Cancer Images pathological classification device characterized by comprising characteristic extracting module, decomposing module, processing mould Block, training module, test module；

The characteristic extracting module obtains feature vector for extracting the color character of preset RGB three-dimensional lesions image data set Collection；

The decomposing module will carry out Non-negative Matrix Factorization by the matrix that column form by described eigenvector collection, obtain coefficient square Battle array；

The coefficient matrix is divided into training dataset and test data set by the processing module, and The first true pathology label of each training data and the second true pathology label of each test data are obtained respectively；

The training module carries out classification based training to classifier with the training dataset and first true tag, obtains The classifier that training is completed；

The test module is classified for the test data set to be input in the classifier that the training is completed, is obtained To prediction pathology label and Confidence distance, wherein Confidence distance is the corresponding probability value of prediction case label.

9. Lung Cancer Images pathological classification device according to claim 8, which is characterized in that further include: evaluation module；

The evaluation module, according to the described second true pathology label, the prediction pathology label and the Confidence distance meter AUC is calculated, the classifying quality of different classifications device is evaluated.

10. Lung Cancer Images pathological classification device according to claim 8, which is characterized in that further include: enhancing module；

Sample, which carries out enhancing processing, to be want to RGB three-dimensional lesions figure by least one image enchancing method, obtains RGB three-dimensional lesions Image data set, described image Enhancement Method include: second section after the enhancing of the first screenshot, the first scaling enhancing, the second scaling Figure enhancing.