CN113361612A - Magnetocardiogram classification method based on deep learning - Google Patents
Magnetocardiogram classification method based on deep learning Download PDFInfo
- Publication number
- CN113361612A CN113361612A CN202110653261.3A CN202110653261A CN113361612A CN 113361612 A CN113361612 A CN 113361612A CN 202110653261 A CN202110653261 A CN 202110653261A CN 113361612 A CN113361612 A CN 113361612A
- Authority
- CN
- China
- Prior art keywords
- network
- magnetocardiogram
- level
- training
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 18
- 238000013135 deep learning Methods 0.000 title claims abstract description 7
- 238000010586 diagram Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 19
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 238000013210 evaluation model Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 description 3
- 208000019622 heart disease Diseases 0.000 description 3
- 238000013526 transfer learning Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 238000005086 pumping Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- 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
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
一种基于深度学习的心磁图分级方法，包括以下步骤：首先将测得的心磁信号进行预处理，得到所需的电流密度矢量图，然后对样本进行适当的扩充，再通过扩充后的样本对网络进行训练得到较优模型，用于心磁图的分级。本发明方法实施简单，只需要少量的心磁图电流密度矢量图对网络进行训练就可以达到很高的分级准确率。
A deep learning-based magnetocardiogram grading method includes the following steps: firstly, preprocess the measured magnetocardiogram to obtain the required current density vector diagram, then appropriately expand the sample, and then pass the expanded The sample trains the network to obtain a better model, which is used for the grading of the magnetocardiogram. The method of the invention is simple to implement, and only needs a small amount of magnetocardiogram current density vector diagrams to train the network to achieve high classification accuracy.
Description
Technical Field
The invention relates to a magnetocardiogram grading method.
Background
Since the change of the magnetic permeability of human tissues is small and can be considered as a constant, the propagation of magnetic signals in the body is hardly influenced, so that the detection of the biological phenomenon by the MCG is more reliable, and the accuracy of diagnosis of various heart diseases can be improved.
The classification of the current density vector diagram of the magnetocardiogram is an important index for clinical magnetocardiogram diagnosis. At present, related researches on the current density vector diagram of the magnetocardiogram are few, most of the known methods are classified by using traditional machine learning, although the methods have effects, researchers need to design experiments to extract the characteristics of the current density vector diagram of the magnetocardiogram, and are time-consuming and labor-consuming, so that a new method needs to be developed for simply and efficiently classifying the current density vector diagram of the magnetocardiogram.
Disclosure of Invention
The invention overcomes the defects in the prior art and provides a high-efficiency and high-accuracy grading method for a current density vector diagram of a small sample magnetocardiogram. Has great practical significance for clinical auxiliary diagnosis of heart diseases.
The invention provides a novel and noninvasive heart disease detection technology, wherein the classification of a current density vector diagram of a magnetocardiogram is an important evaluation index, and the invention provides a classification method of the current density vector diagram of the magnetocardiogram based on ResNet18 and transfer learning in view of the characteristics that a convolutional neural network has excellent performance in the aspect of image classification and transfer learning is suitable for small sample classification.
The invention discloses a magnetocardiogram grading method based on deep learning, which comprises the following steps:
step 1, collecting magnetocardiogram data of a subject, generating a current density vector diagram after processing, then sending all samples to experts for labeling, and finally obtaining 5 types (0 level, 1 level, 2 level, 3 level and 4 level) of labeled data.
And 2, expanding the samples marked in the step 1 (changing contrast, changing chroma, changing brightness, adding random noise and sharpening).
And 3, mixing the expanded data set with the original data set, and dividing the data set into a training set, a verification set and a test set according to the ratio of 6:2: 2.
Step 4, loading a ResNet18 network pre-trained on the ImageNet data set, removing the softmax output layer with the final node of 1000, and adding an output layer with 5 nodes according to the magnetocardiogram classification task; inputting the divided training set and verification set into a network for training to obtain a better network model and storing network weight for subsequent identification and classification; because the bottom layer features of the pictures extracted by the convolutional neural network are all the same and different, the first layers of ResNet are frozen during training, and the images are not subjected to back propagation of errors.
And 5, randomly extracting pictures in the test set by using the trained network model, inputting the pictures into the network model, outputting the prediction labels and the corresponding probabilities corresponding to the pictures by the network model, and outputting the evaluation indexes of the evaluation model.
The invention utilizes the convolutional neural network to automatically extract the characteristic of the image, does not need a large amount of training images for transfer learning, combines the convolutional neural network and the training images, only needs to put a small amount of samples with labels into the network for training, then adjusts related parameters, and obtains good grading accuracy when the network achieves better effect.
The invention has the advantages that: the implementation is simple, and the high grading accuracy rate can be achieved only by training the network by a small amount of current density vector diagrams of the magnetocardiogram.
Drawings
FIG. 1 is an overall flow chart of the method of the present invention.
Fig. 2 is a network structure diagram of the method of the present invention.
FIG. 3 is a loss curve for the network training of the present invention.
FIG. 4 is a graph of network training accuracy and validation accuracy in accordance with the present invention.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
The invention discloses a magnetocardiogram grading method based on deep learning, which comprises the following steps as shown in figure 1:
step 1, collecting magnetocardiogram data of a subject on a pumping type Rb atom magnetometer, carrying out a series of operations on the magnetocardiogram data to generate a current density vector diagram, then sending all samples to experts for labeling, and finally obtaining 5 types (0 level, 1 level, 2 level, 3 level and 4 level) of labeled data.
And 2, expanding each type of the samples with the labels in the step 1 by 5 times by using a plurality of data enhancement methods to improve the generalization capability of the model, namely changing the chroma, the contrast, the brightness, sharpening and adding random noise of the picture.
And 3, mixing the expanded data set with the original data set, and performing data conversion according to the ratio of 6:2:2, dividing the training set, the verification set and the test set.
And 4, loading a ResNet18 network pre-trained on the ImageNet data set, removing the softmax output layer with the final node of 1000, and adding an output layer with 5 nodes according to the magnetocardiogram classification task, wherein the network structure is shown in FIG. 2. And then inputting the divided samples into a network for training, wherein the bottom layer characteristics of the picture extracted by the convolutional neural network are different, so that the first layers of ResNet are frozen during training, and the ResNet is not allowed to participate in the back propagation of errors.
The input image size was set to 224 x 224, the learning rate of the network was set to 0.0001, the batch _ size was 8, the cross entropy loss function (CrossEntropyLoss) was defined as the loss function of the network, and 200 epochs were trained on the training set and the validation set using the Adam optimizer until the model converged, see fig. 3 and 4.
And 5, randomly extracting pictures in the test set by using the trained network model, inputting the pictures into the network model, outputting the pictures and the probabilities corresponding to the prediction labels by using the network model, and finally outputting evaluation indexes to evaluate the quality of the network model.
The evaluation indexes in step 5 include accuracy (accuracycacy) of the model, precision (precision) of each type, specificity (specificity), recall (call) and confusion matrix.
The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as would occur to those skilled in the art upon consideration of the present inventive concept.
Claims (3)
1. A magnetocardiogram grading method based on deep learning is characterized by comprising the following steps:
step 1, collecting magnetocardiogram data of a subject, generating a current density vector diagram after processing, then sending all samples to experts for labeling, and finally obtaining 5 types (0 level, 1 level, 2 level, 3 level and 4 level) of labeled data;
step 2, expanding the samples marked in the step 1, wherein the expansion operation comprises adjusting the brightness, the chroma and the contrast of the picture, sharpening the picture and adding random noise;
step 3, mixing the expanded data set with the original data, and dividing a training set, a verification set and a test set according to the ratio of 6:2: 2;
step 4, loading a pre-trained ResNet18 network on the ImageNet data set, removing the softmax output layer with the final node of 1000, and adding an output layer with 5 nodes according to the magnetocardiogram classification task; inputting the divided training set and verification set into a network for training to obtain a better network model and storing network weight for subsequent identification and classification; because the bottom layer characteristics of the picture extracted by the convolutional neural network are different, the first layers of ResNet are frozen during training, and the picture is not subjected to back propagation of errors;
and 5, randomly extracting pictures in the test set by using the trained network model, inputting the pictures into the network model, outputting the prediction labels and the corresponding probabilities corresponding to the pictures by using the network model, and outputting the evaluation indexes of the evaluation model.
2. The magnetocardiogram classification method based on deep learning according to claim 1, wherein: the step 4 specifically comprises the following steps: the input image size is set to 224 x 224, the learning rate of the network is set to 0.0001, the batch _ size is 8, a Cross Entropy Loss function (Cross Entropy Loss) is defined as the Loss function of the network, and 200 epochs are trained on a training set and a verification set by using an Adam optimizer until the model converges.
3. The method of claim 1, wherein the evaluation indexes in step 5 include model accuracy (accuracy), precision (precision) of each class, specificity (specificity), recall (recall), and confusion matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110653261.3A CN113361612A (en) | 2021-06-11 | 2021-06-11 | Magnetocardiogram classification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110653261.3A CN113361612A (en) | 2021-06-11 | 2021-06-11 | Magnetocardiogram classification method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113361612A true CN113361612A (en) | 2021-09-07 |
Family
ID=77533906
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110653261.3A Pending CN113361612A (en) | 2021-06-11 | 2021-06-11 | Magnetocardiogram classification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113361612A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117557840A (en) * | 2023-11-10 | 2024-02-13 | 中国矿业大学 | A method for grading fundus lesions based on small sample learning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112508063A (en) * | 2020-11-23 | 2021-03-16 | 刘勇志 | Medical image classification method based on incremental learning |
-
2021
- 2021-06-11 CN CN202110653261.3A patent/CN113361612A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112508063A (en) * | 2020-11-23 | 2021-03-16 | 刘勇志 | Medical image classification method based on incremental learning |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117557840A (en) * | 2023-11-10 | 2024-02-13 | 中国矿业大学 | A method for grading fundus lesions based on small sample learning |
| CN117557840B (en) * | 2023-11-10 | 2024-05-24 | 中国矿业大学 | Fundus lesion grading method based on small sample learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109584254B (en) | Heart left ventricle segmentation method based on deep full convolution neural network | |
| CN107506761B (en) | Brain image segmentation method and system based on saliency learning convolutional neural network | |
| CN115050477B (en) | Bethes-optimized RF and LightGBM disease prediction method | |
| CN111539467A (en) | GAN network architecture and method for data augmentation of medical image datasets based on generative adversarial networks | |
| CN112101451A (en) | Breast cancer histopathology type classification method based on generation of confrontation network screening image blocks | |
| CN104850860A (en) | Cell image recognition method and cell image recognition device | |
| CN114581701A (en) | Method for generating dynamic enhanced image features by weighting image features through T2 | |
| CN114708637B (en) | A meta-learning-based facial action unit detection method | |
| CN115222688A (en) | Medical image classification method based on graph network time sequence | |
| Li et al. | A deep learning method for material performance recognition in laser additive manufacturing | |
| CN114391826A (en) | A method and device for predicting human representation based on edge-driven graph neural network | |
| CN117122303A (en) | Brain network prediction method, system, equipment and storage medium | |
| CN112529063B (en) | A deep domain adaptive classification method for Parkinson's speech datasets | |
| CN110738662A (en) | Image grading method of pituitary tumor texture based on fine-grained medical image segmentation and ground truth discovery data augmentation | |
| CN116759067A (en) | A liver disease diagnosis method based on reconstruction and Tabular data | |
| CN113361612A (en) | Magnetocardiogram classification method based on deep learning | |
| CN105957092A (en) | Mammary gland molybdenum target image feature self-learning extraction method for computer-aided diagnosis | |
| Deng et al. | A Strategy of MR Brain Tissue Images' Suggestive Annotation Based on Modified U-Net | |
| CN114529519A (en) | Image compressed sensing reconstruction method and system based on multi-scale depth cavity residual error network | |
| CN116052872B (en) | Facial expression-based intelligent data evaluation method and system for Parkinson disease | |
| CN118053063A (en) | SCAE-DensNet-based lung cancer disease diagnosis and classification method | |
| CN115409843B (en) | Feature extraction method for brain neuroimaging based on scale-balanced coupled convolutional architecture | |
| Pan et al. | X-ray mammary image segmentation based on convolutional neural network | |
| CN116468763A (en) | A Method of Electron Microscope Image Registration Based on Cost Volume | |
| Li et al. | Distance assisted recursive testing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210907 |