CN113705632A - Rectal cancer MRI image classification method and device, electronic equipment and storage medium - Google Patents
Rectal cancer MRI image classification method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN113705632A CN113705632A CN202110918371.8A CN202110918371A CN113705632A CN 113705632 A CN113705632 A CN 113705632A CN 202110918371 A CN202110918371 A CN 202110918371A CN 113705632 A CN113705632 A CN 113705632A
- Authority
- CN
- China
- Prior art keywords
- rectal cancer
- image
- mri image
- model
- classification
- 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
Images
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
Abstract
The invention discloses a rectal cancer MRI image classification method, a rectal cancer MRI image classification device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring a rectal cancer MRI image, preprocessing the MRI image, calculating the similarity between any pictures to form an N x N characteristic matrix, extracting 10-dimensional characteristics through a self-coding structure to form an N x 10 characteristic matrix, and dividing an image characteristic matrix set into a training set and a testing set; constructing an SVM model, and taking a training set as the input of the SVM model to obtain a test model; and inputting the test data in the test set into the test model to obtain a classification result. The method is high in classification accuracy, and can accurately predict the image classification result and assist doctors in performing T stage diagnosis of the rectal cancer.
Description
Technical Field
The present invention relates to an image classification method, an apparatus, an electronic device and a storage medium, and in particular, to an MRI image classification method, an apparatus, an electronic device and a storage medium for rectal cancer.
Background
Magnetic resonance imaging relies on the magnetic properties of the nuclei and their interaction with an externally applied magnetic field. The nuclei are composed of positively charged protons and uncharged neutrons. Electrons outside the core are negatively charged, forming an electron cloud shell. The nucleus of nuclear magnetic resonance does not mean any one nucleus, but a nucleus having an MR active element. The protons are excited with radio wave pulses of the correct frequency to resonate them, disrupting the magnetic alignment. The excited protons release the absorbed energy in the form of a radio frequency signal and the emissions are received by a receive coil on the scanner. The radio frequency causing proton resonance depends on the strength of the magnetic field. By applying different frequencies in sequence, MRI images can be obtained. The MRI image can reflect not only the anatomical morphology of the human body but also physiological function information such as the blood flow and cell metabolism of the human body.
Before medical image classification algorithms appeared, doctors completely relied on manual diagnosis when analyzing myocardial MRI images, and the manual diagnosis mode is not only tedious and inefficient, but also the accuracy of diagnosis results completely depends on personal experience and professional level of doctors. In the prior art, Liaoxian and the like judge the preoperative T stage of rectal cancer based on high-resolution T2WI imaging omics, the image segmentation is carried out on ITK-SNAP (Version 3.60, www.itksnap.org) software, the whole tumor is delineated on preoperative high-resolution T2WI to be an ROI, 14 imaging omics characteristics are extracted through general linear LASSO analysis, and finally, the classification is realized by using a random forest algorithm; this classification relies on manual delineation of the lesion and extraction of features, which, although accurate, is inefficient.
Although the existing medical image classification algorithm can also classify the MRI images, most of the classification results are not satisfactory, and doctors still play a leading role in the image analysis process. In the treatment of rectal cancer, T-phase diagnosis based on Magnetic Resonance Imaging (MRI) is a critical step. T staging is related to the extent of involvement of cancer cells in the organ. Accurate T stage identification can help doctors to judge the disease development stage and implement reasonable treatment means.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a rectal cancer MRI image classification method with high accuracy; another object of the present invention is to provide an MRI image classification apparatus for rectal cancer; another object of the present invention is to provide an electronic device; it is another object of the invention to provide a non-transitory computer-readable storage medium.
The technical scheme is as follows: the rectal cancer MRI image classification method comprises the steps of collecting rectal cancer MRI images, preprocessing the MRI images, calculating the similarity between any images to form an N x N characteristic matrix, extracting 10-dimensional characteristics through a self-coding structure to form an N x 10 characteristic matrix, and dividing an image characteristic matrix set into a training set and a test set; constructing an SVM model, and taking a training set as the input of the SVM model to obtain a test model; and inputting the test data in the test set into the test model to obtain a classification result.
Further, the MRI image preprocessing method is to scale all image sizes to 96 × 96 size by resize function of the deep learning library pytorch.
Further, the method for calculating the similarity between any two pictures is a complex wavelet structure similarity index calculation method:
further, the self-coding structure is as follows: n → 64 → 10 ← 64 ← N, where N represents the number of samples.
The device applying the rectal cancer MRI image classification method comprises an image acquisition module, an image feature extraction module and an SVM model classification module; the image acquisition module acquires a rectal cancer MRI image and sends the image to the image feature extraction module, the image feature extraction module calculates the similarity between any pictures through a complex wavelet structure similarity index to form an N x N feature matrix, then extracts new features through a self-coding structure to form an N x 10 feature matrix, the image feature extraction module divides the extracted feature matrix into a training set and a testing set and sends the training set to the SVM model classification module, the SVM model classification module inputs the training set for model training, and then test data is input into a trained model to obtain a classification result.
Furthermore, the image feature extraction module calculates the similarity between any pictures by a complex wavelet structure similarity index calculation method to form an N-N feature matrix.
Further, the N x N feature matrix forms an N x 10 feature matrix by extracting new features from the encoded structure.
In another aspect, an electronic device is provided, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to implement the MRI image classification method for rectal cancer.
In another aspect, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, implements the rectal cancer MRI image classification method as described above.
Has the advantages that: compared with the prior art, the invention has the following remarkable advantages: the method is high in classification accuracy, and can accurately predict the image classification result and assist doctors in performing T stage diagnosis of the rectal cancer.
Drawings
FIG. 1 is a schematic diagram of the apparatus of the present invention;
FIG. 2 is a schematic flow chart of the method of the present invention.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
Examples
As shown in FIG. 1, the MRI image classification device for rectal cancer of the invention comprises an image acquisition module, an image feature extraction module and an SVM model classification module; the image acquisition module acquires a rectal cancer MRI image, the image is sent to the image feature extraction module, the image feature extraction module calculates the similarity between any images through a complex wavelet structure similarity index to form an N x N feature matrix, new features are extracted through a self-coding structure to form an N x 10 feature matrix, the image feature extraction module divides the extracted feature matrix into a training set and a testing set and sends the training set to the SVM model classification module, the SVM model classification module performs model training through the input training set, and then test data are input into the trained model to obtain a classification result.
The general method for calculating the similarity index of the complex wavelet structure comprises the following steps:
the invention is applied to the similarity calculation of the medical MRI images, the signal-to-noise ratio of the MRI images is higher, so that the parameter K is not needed, the CW-SSIM is improved, and a new image similarity calculation mode is formed:
the obtained ICW-SSIM is used for calculating the image similarity.
The embodiment also relates to a feature extraction method: the invention relates to self-encoding (Autoencoder), which adjusts a frame therein according to experience and experimental research and designs a self-encoding structure suitable for medical images: n → 64 → 10 ← 64 ← N, where N represents the number of samples.
In the image classification stage, a classical Support Vector Machine (SVM) classifier is used. The support vector machine is a generalized linear classifier for binary classification of data in a supervised learning mode, and a decision boundary of the support vector machine is a maximum margin hyperplane for solving learning samples.
As shown in fig. 2, the image classification method of the present embodiment is as follows:
(1) collecting rectal cancer MRI image samples, and scaling all the samples into 96 × 96 size images by using a resize function of a deep learning library pytorch, namely the number of pixels of the images is 96 × 96;
(2) calculating the image similarity between any images of the preprocessed MRI images by using ICW-SSIM to form an N x N characteristic matrix;
(3) extracting 10-dimensional new features from the N x N feature matrix through a 5-layer automatic encoder to form an N x 10 feature matrix;
(4) the image characteristic matrix set is divided into a training set and a testing set; randomly dividing a training set and a testing set: 80% of the data were used as training set and 20% as test set.
(5) Constructing an SVM model;
the SVM model training method comprises the following steps:
(501) the SVM model adopts a python built-in model.
(502) And in the training stage, a 5-fold cross validation method is adopted, and the model with the best effect is selected as a final model.
(6) And inputting the test data in the test set into the test model to obtain a classification result.
The MRI image generally has higher signal-to-noise ratio and higher robustness, and better characteristics of the original nuclear magnetic resonance image can be obtained by modifying a CW-SSIM model into a complex wavelet structure similarity index (ICW-SSIM). The SVM processes small data sets very well because its objective function is convex and the best parameter values for the processing can be obtained by convex optimization. To obtain better characteristics, a 5-layer auto-encoder is used to obtain the underlying features of the original data.
Comparative example
All image samples were scaled to 96 × 96 using the same samples as in the example, and then the similarity was calculated to obtain N × N feature matrices, and then self-encoding was used to obtain N × 10 features. All features are also in accordance with 8: 2, randomly dividing the training set and the testing set for training and testing the comparison model.
The comparison models comprise CNNs, Alexnet, Resnet18, Resnet50, capsule network, random forest + RF features, FE-SVM + RF features, random forest + deployed features, and the accuracy of the classification results is calculated.
The evaluation indexes of the image classification accuracy rate mainly include three types: ACC, AUC and F-score. The AUC indicator is calculated as follows:
ACC=(#TP+#TN)/N
where # TP is the number of positive samples predicted to be positive, # TN is the number of negative samples predicted to be negative, and N is the total number of samples. The AUC indicator is calculated as follows:
wherein rankiRepresents xiThe rank position (rank), M and N distributions of the prediction probabilities represent the number of positive samples and the number of negative samples.The F-score index is calculated as follows:
F-score=2*Precision*Recall/(Precision+Recall)
Precision=#TP/(#TP+#FP)
Recall=#TP/(#TP+#FN)
where # FP represents the number of false positive samples and # FN represents the number of false negative samples.
TABLE 1 comparison of the prediction accuracy effect of the examples with other advanced models at present (mean. + -. variance)
TABLE 2 comparison of the prediction accuracy effectiveness of the examples with other advanced models at present (maximum)
As shown in table 1, a total of 100 experiments were performed. The model proposed by the examples is in full contrast to other advanced models. The SVM + deployed features model adopted in the embodiment is better in performance on three accuracy indexes (ACC, AUC and F-score) compared with the most advanced model and the like at present. In particular, the method is compared with the Features (RF Features) extracted by Random Forest, namely, the Random Forest + RF Features, the SVM + RF Features and the SVM + RF Features have good effect and have larger difference without an SVM + deployed Features model. Only when the proposed two-step feature (image similarity + self-encoding) is adopted, the Random Forest + deployed Features model effect is obviously proposed, and ACC reaches 0.7372 +/-0.0324 and is very close to the SVM + deployed Features model.
As shown in table 2, the maximum values of the three indexes in 100 experiments are extracted and compared, and the maximum value of the three indexes is obtained by the proposed SVM + deployed Features model. The effectiveness of the SVM + deployed features model adopted by the embodiment is fully illustrated.
Claims (7)
1. The rectal cancer MRI image classification method is characterized in that rectal cancer MRI images are collected, the MRI images are preprocessed, the similarity between any images is calculated to form an N x N characteristic matrix, 10-dimensional characteristics are extracted through a self-coding structure to form an N x 10 characteristic matrix, and an image characteristic matrix set is divided into a training set and a testing set; constructing an SVM model, and taking a training set as the input of the SVM model to obtain a test model; and inputting the test data in the test set into the test model to obtain a classification result.
2. The MRI image classification method for rectal cancer according to claim 1, characterized in that the MRI image preprocessing method is to scale all image sizes by resize function of deep learning library pytorch.
4. the MRI image classification method for rectal cancer according to claim 1, characterized in that the self-coding structure is: n → 64 → 10 ← 64 ← N, where N represents the number of samples.
5. The rectal cancer MRI image classification device is characterized by comprising an image acquisition module, an image feature extraction module and an SVM model classification module; the image acquisition module acquires a rectal cancer MRI image and sends the image to the image feature extraction module, the image feature extraction module calculates the similarity between any pictures through a complex wavelet structure similarity index to form an N x N feature matrix, then extracts new features through a self-coding structure to form an N x 10 feature matrix, the image feature extraction module divides the extracted feature matrix into a training set and a testing set and sends the training set to the SVM model classification module, the SVM model classification module inputs the training set for model training, and then test data is input into a trained model to obtain a classification result.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the MRI image classification method for rectal cancer according to any one of claims 1-4.
7. A non-transitory computer-readable storage medium, having stored thereon a computer program which, when executed by a processor, implements the MRI image classification method of rectal cancer according to any one of claims 1 to 4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110918371.8A CN113705632A (en) | 2021-08-11 | 2021-08-11 | Rectal cancer MRI image classification method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110918371.8A CN113705632A (en) | 2021-08-11 | 2021-08-11 | Rectal cancer MRI image classification method and device, electronic equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113705632A true CN113705632A (en) | 2021-11-26 |
Family
ID=78652265
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110918371.8A Pending CN113705632A (en) | 2021-08-11 | 2021-08-11 | Rectal cancer MRI image classification method and device, electronic equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113705632A (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020062901A1 (en) * | 2018-09-28 | 2020-04-02 | 深圳大学 | Method and system for analyzing image quality of super-resolution image |
US20200111210A1 (en) * | 2018-10-09 | 2020-04-09 | General Electric Company | System and method for assessing image quality |
CN112149717A (en) * | 2020-09-03 | 2020-12-29 | 清华大学 | Confidence weighting-based graph neural network training method and device |
-
2021
- 2021-08-11 CN CN202110918371.8A patent/CN113705632A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020062901A1 (en) * | 2018-09-28 | 2020-04-02 | 深圳大学 | Method and system for analyzing image quality of super-resolution image |
US20200111210A1 (en) * | 2018-10-09 | 2020-04-09 | General Electric Company | System and method for assessing image quality |
CN112149717A (en) * | 2020-09-03 | 2020-12-29 | 清华大学 | Confidence weighting-based graph neural network training method and device |
Non-Patent Citations (2)
Title |
---|
YIZHANG WANG ET AL.: ""A feature extraction based support vector machine model for rectal cancer T-stage prediction using MRI images"", 《MULTIMEDIA TOOLS AND APPLICATIONS》 * |
徐肖攀等: ""基于多模态MRI影像组学策略构建膀胱癌肌层浸润预测模型研究"", 《中国医学装备》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Intelligent scanning: Automated standard plane selection and biometric measurement of early gestational sac in routine ultrasound examination | |
George et al. | Computer assisted recognition of breast cancer in biopsy images via fusion of nucleus-guided deep convolutional features | |
JP2017076193A (en) | Brain activity analysis device, brain activity analysis method and brain activity analysis program | |
Lamrani et al. | Brain tumor detection using mri images and convolutional neural network | |
Mahapatra et al. | Active learning based segmentation of Crohns disease from abdominal MRI | |
Mahapatra et al. | Active learning based segmentation of Crohn's disease using principles of visual saliency | |
CN111080658A (en) | Cervical MRI image segmentation method based on deformable registration and DCNN | |
Karimzadeh et al. | A novel shape-based loss function for machine learning-based seminal organ segmentation in medical imaging | |
Ansari et al. | Multiple sclerosis lesion segmentation in brain MRI using inception modules embedded in a convolutional neural network | |
CN115136189A (en) | Automated detection of tumors based on image processing | |
Razzaq et al. | Brain tumor detection from mri images using bag of features and deep neural network | |
Tandon et al. | Sequential convolutional neural network for automatic breast cancer image classification using histopathological images | |
CN103268494A (en) | Parasite egg identifying method based on sparse representation | |
Tong et al. | Automatic lumen border detection in IVUS images using dictionary learning and kernel sparse representation | |
CN116030063B (en) | Classification diagnosis system, method, electronic device and medium for MRI image | |
CN115861716B (en) | Glioma classification method and device based on twin neural network and image histology | |
Zhang et al. | Factorized Omnidirectional Representation based Vision GNN for Anisotropic 3D Multimodal MR Image Segmentation | |
CN113705632A (en) | Rectal cancer MRI image classification method and device, electronic equipment and storage medium | |
CN111598144B (en) | Training method and device for image recognition model | |
Lin et al. | Feature selection algorithm for classification of multispectral MR images using constrained energy minimization | |
CN113222887A (en) | Deep learning-based nano-iron labeled neural stem cell tracing method | |
Kothari et al. | An overview on automated brain tumor segmentation techniques | |
Ben-Cohen et al. | Sparsity-based liver metastases detection using learned dictionaries | |
Chang et al. | Texture analysis method for shape-based segmentation in medical image | |
CN112562031B (en) | Nuclear magnetic resonance image clustering method based on sample distance reconstruction |
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 |