CN114581425A - Myocardial segment defect image processing method based on deep neural network - Google Patents
Myocardial segment defect image processing method based on deep neural network Download PDFInfo
- Publication number
- CN114581425A CN114581425A CN202210238690.9A CN202210238690A CN114581425A CN 114581425 A CN114581425 A CN 114581425A CN 202210238690 A CN202210238690 A CN 202210238690A CN 114581425 A CN114581425 A CN 114581425A
- Authority
- CN
- China
- Prior art keywords
- neural network
- deep neural
- myocardial
- image data
- network 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.)
- Granted
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
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30048—Heart; Cardiac
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a myocardial segment defect image processing method based on a deep neural network, which comprises the steps of obtaining myocardial perfusion imaging image data, and carrying out data preprocessing on an ROI (region of interest) area and a score label in the image data to obtain a training set and a test set; constructing a deep neural network model, acquiring scoring label characteristics, and performing characteristic fusion exchange on the scoring labels; inputting myocardial perfusion imaging image data preprocessing data into the loaded deep neural network model, outputting myocardial segment defect scoring results, updating a network weight through a back propagation algorithm, and performing multiple rounds of iterative training of parameters of the deep neural network model; and inputting the training set into the trained deep neural network model, and outputting the myocardial segment score according to the myocardial segment classification index. The myocardial perfusion imaging image data are processed by constructing the deep neural network model, so that the accuracy and consistency of the myocardial perfusion imaging image data for judging myocardial perfusion imaging are improved, and the judgment result is more objective.
Description
Technical Field
The invention relates to the technical field of medical image analysis, in particular to a myocardial segment defect image processing method based on a deep neural network.
Background
The prevalence rate of cardiovascular diseases in China is in a continuously rising stage, according to the calculation of general survey, the number of the existing cardiovascular diseases in China is 3.3 hundred million, the cardiovascular diseases are the first diseases affecting the health of the people in China, and the death rate is even higher than that of tumors and other diseases. Myocardial Perfusion Imaging (MPI) has very important values in diagnosis, risk stratification, prognosis judgment and treatment scheme formulation of cardiovascular diseases, especially Coronary Heart Disease (CHD), and has been widely applied to clinical application. It can effectively display the blood flow in the myocardial area supplied by the stenotic artery and the blood flow in the myocardial area covered by the non-stenotic blood vessel by detecting the real-time condition of the radionuclide labeled medicine along with the blood flow in the heart, and detect the disease area corresponding to the myocardial ischemia area by comparing. Notably, analysis of myocardial perfusion imaging results and the corresponding routine diagnosis of myocardial ischemia is tedious, time consuming and subject to subjective influences of the reading physician.
Disclosure of Invention
The invention aims to solve the technical problems that the prior art has great influence on myocardial perfusion development image processing and has low judgment accuracy and consistency, and aims to provide a myocardial segment defect image processing method based on a deep neural network.
The invention is realized by the following technical scheme:
a myocardial segment defect image processing method based on a deep neural network comprises the following steps:
s1, acquiring myocardial perfusion imaging image data, and performing data preprocessing on an ROI (region of interest) region and a scoring label in the image data to obtain a training set and a test set;
s2, constructing a deep neural network model, acquiring scoring label characteristics, and performing characteristic fusion exchange on the scoring labels;
s3, inputting myocardial perfusion imaging image data preprocessing data into the loaded deep neural network model, outputting myocardial segment defect scoring results, updating the network weight through a back propagation algorithm, and performing multiple rounds of iterative training of deep neural network model parameters;
and S4, inputting the training set into the trained deep neural network model, outputting the myocardial segment score according to the myocardial segment classification index, and evaluating the myocardial segment score output result through the deep neural network model.
The method comprises the steps of obtaining myocardial perfusion imaging image data, and carrying out data preprocessing on an ROI (region of interest) region and a scoring label in the image data to obtain a training set and a testing set; constructing a deep neural network model, acquiring label characteristics, and performing characteristic fusion exchange on the labels; inputting myocardial perfusion imaging image data preprocessing data into the loaded deep neural network model, outputting myocardial segment defect scoring results, updating a network weight through a back propagation algorithm, and performing multiple rounds of iterative training of parameters of the deep neural network model; after the model training is completed, the training set is input into the trained deep neural network model, the myocardial segment score is output according to the myocardial segment classification index, the myocardial perfusion imaging image data is processed by constructing the deep neural network model, the accuracy and consistency of the myocardial perfusion imaging image data for judging myocardial perfusion imaging are improved, and the judgment result is more objective.
As a further limitation of the invention, the data preprocessing comprises:
extracting an ROI area and a scoring tag in the image data;
dividing the preprocessed image data into a training set and a test set;
and carrying out data augmentation on the test set.
As a further limitation of the present invention, the method for extracting the ROI region in the image data is to use a myocardial perfusion imaging image data slice as a network input;
the method for extracting the scoring label in the image data comprises the step of extracting the myocardial segment scoring label through a regular expression, wherein the scoring label is text information contained in diagnosis information of a myocardial perfusion imaging report.
As a further limitation of the present invention, the dividing the extracted image data into a training set and a test set includes dividing the image data into the training set and the test set at a 4:1 ratio.
As a further limitation of the present invention, said constructing the deep neural network model comprises removing the fully-connected layer and the pooling layer of Resnet50 as a feature extraction skeleton of the deep neural network model.
As a further limitation of the present invention, the obtaining of the scoring tag feature comprises performing a multitasking and a multi-output on the divided tags, the multitasking and the multi-output comprising:
the first task output is used for enabling the characteristics to have task specificity before the characteristics are subjected to characteristic fusion exchange;
the second task output is used for outputting based on the result of the specific characteristic and the fused characteristic.
As a further limitation of the present invention, the feature fusion exchange of the scoring tags specifically includes:
fusing scoring label features with task specificity according to the regional affiliation of the myocardial segments;
obtaining global characteristic information and local characteristic information through the characteristic fusion of the grading labels;
and distributing the global characteristic information and the local characteristic information according to the myocardial segment classification indexes.
As a further limitation of the present invention, the parameters of the deep neural network model specifically include forward calculation, weight update, and loss function definition, and the parameters of the deep neural network model are used to update network weights and find a suitable set of weights, so that the network has the minimum objective function.
As a further limitation of the present invention, the forward calculation specifically includes:
continuously carrying out forward calculation from an input layer to an output layer, wherein the process is as follows:
the activation values for the network output layer neurons are:
ai L＝f(L-1)(W(L-1)·f(L-2)(W(L-2)…ai 0))
wherein, for a feedforward neural network of L layer, the training sample set is set as X ∈ Rm×nWhere m is the dimension of a single sample, n represents the number of training samples, XiFor the (i) -th sample, the sample is,
connecting weight value W from jth neuron of ith layer to kth neuron of l +1 layer(l)Is a connection weight matrix from layer l to layer l +1, f (-) is an activation function of neurons on layer l, ai lThe activation value for the ith sample for layer I neurons.
As a further limitation of the present invention, the loss function is defined as:
where M is the number of myocardial segments, and M is 17. L issRepresenting a loss function of the second output, L'sAnd representing loss functions of the first output, wherein the loss functions are cross entropy loss functions, lambda is a hyperparameter used for balancing proportion, and lambda is 0.5.
Compared with the prior art, the invention has the following advantages and beneficial effects:
after the deep neural network model is trained, a myocardial perfusion imaging diagnosis result can be obtained quickly, the working efficiency is effectively improved, and the processing result is more visual.
1. The deep neural network model is constructed to process the myocardial perfusion imaging image data, so that the accuracy and consistency of the myocardial perfusion imaging image data for judging myocardial perfusion imaging are improved, and the judgment result is more objective.
2. After the deep neural network model is trained, 17 myocardial segment scores can be output only by inputting a myocardial perfusion imaging picture, and the deep neural network model realizes automatic myocardial segment defect scoring and improves the working efficiency.
3. The deep neural network model realizes feature hierarchical fusion exchange through feature fusion exchange, so that each task can be classified based on local semantic information of the current task and global semantic information of fusion features of all tasks.
4. The deep neural network model has two task outputs, so that the deep neural network model has task specificity when score tag feature fusion exchange is carried out, and the performance of the deep neural network model is integrally improved.
Drawings
In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and that for those skilled in the art, other related drawings can be obtained from these drawings without inventive effort. In the drawings:
FIG. 1 is a flow chart of a determination method in an embodiment of the present invention;
FIG. 2 is a diagram of a deep neural network model according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating feature fusion exchange in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to examples and accompanying drawings, and the exemplary embodiments and descriptions thereof are only used for explaining the present invention and are not meant to limit the present invention.
The 17-segment heart model is one of the most commonly used diagnostic methods for myocardial ischemia: it divides the myocardium into 17 myocardial segments, scores the defect condition of each segment, and judges whether the myocardium is ischemic based on all the scoring results. Based on the 17-segment heart model, the heart is subdivided into three regions: left anterior descending LAD comprising myocardial segments 1, 2, 7, 8, 13, 14, 17, left circumflex LCX comprising myocardial segments 3, 4, 9, 10, 15, and right coronary RCA comprising myocardial segments 5, 6, 11, 12, 16. Scoring of myocardial segment defect degree refers to the method of fifths: 0 represents normal, 1 represents mild defect, 2 represents moderate defect, 3 represents severe defect, and 4 represents no visualization. In the daily diagnosis, a physician needs to score the defects of 17 myocardial segments, and when the total score is greater than 3, the region including the myocardial segment is diagnosed as myocardial ischemia, and the case is also diagnosed as myocardial ischemia.
In nuclide myocardial ischemia diagnosis, firstly, 17 myocardial segments are divided, so that the accurate judgment of a final result is greatly influenced and can be accurately finished by a nuclear medicine doctor/technician with abundant experience; secondly, the myocardial perfusion imaging is used as metabolic imaging, the resolution is not as clear as anatomical imaging such as CT and MR, and attenuation artifacts and motion artifacts appear in the image due to the heart pulsation; finally, the myocardial perfusion image can be accurately diagnosed only by a professional nuclear medicine physician with rich reading experience according to the opinion of a clinician; these factors lead to problems of long diagnosis time of myocardial ischemia, high experience dependence, low consistency of conclusions drawn by different physicians, and the like, and are particularly prominent in the primary medical unit.
Example 1
As shown in fig. 1, the present embodiment provides a method for processing a defective image of a myocardial segment based on a deep neural network, including:
s1, acquiring myocardial perfusion imaging image data, and performing data preprocessing on an ROI (region of interest) region and a scoring label in the image data to obtain a training set and a test set;
s2, constructing a deep neural network model, acquiring scoring label characteristics, and performing characteristic fusion exchange on the scoring labels;
s3, inputting myocardial perfusion imaging image data preprocessing data into the loaded deep neural network model, outputting myocardial segment defect scoring results, updating the network weight through a back propagation algorithm, and performing multiple rounds of iterative training of deep neural network model parameters;
and S4, inputting the training set into the trained deep neural network model, outputting the myocardial segment score according to the myocardial segment classification index, and evaluating the myocardial segment score output result through the deep neural network model.
The method comprises the steps of obtaining myocardial perfusion imaging image data, and carrying out data preprocessing on an ROI (region of interest) region and a scoring label in the image data to obtain a training set and a testing set; constructing a deep neural network model, acquiring label characteristics, and performing characteristic fusion exchange on the labels; inputting myocardial perfusion imaging image data preprocessing data into the loaded deep neural network model, outputting myocardial segment defect scoring results, updating a network weight through a back propagation algorithm, and performing multiple rounds of iterative training of parameters of the deep neural network model; after the model training is completed, the training set is input into the trained deep neural network model, the myocardial segment score is output according to the myocardial segment classification index, the myocardial perfusion imaging image data is processed by constructing the deep neural network model, the accuracy and consistency of the myocardial perfusion imaging image data for judging myocardial perfusion imaging are improved, and the judgment result is more objective.
In some possible embodiments, the data pre-processing includes extracting ROI regions and scoring labels in the image data, dividing the pre-processed image data into a training set and a test set, and performing data augmentation on the test set. The method for extracting the ROI area in the image data is to use a data section of the myocardial perfusion imaging image as network input; the method for extracting the scoring label in the image data comprises the step of extracting the myocardial segment scoring label through a regular expression, wherein the scoring label is text information contained in diagnosis information of a myocardial perfusion imaging report. The dividing of the extracted image data into a training set and a test set includes dividing the image data into the training set and the test set in a ratio of 4: 1. The building of the deep neural network model comprises removing a full connection layer and a pooling layer of Resnet50 as a feature extraction skeleton of the deep neural network model.
In some possible embodiments, the obtaining the scoring tag feature comprises performing a multitasking multi-output on the divided tags, the multitasking multi-output comprising: the first task output is used for enabling the characteristics to have task specificity before the characteristics are subjected to characteristic fusion exchange; the second task output is used for outputting based on the result of the specific characteristic and the fused characteristic. The deep neural network model has two task outputs, so that the deep neural network model has task specificity when score tag feature fusion exchange is carried out, and the performance of the deep neural network model is integrally improved.
In some possible embodiments, the performing feature fusion exchange on the scoring tags specifically includes: fusing scoring label features with task specificity according to the regional affiliation of the myocardial segments; obtaining global characteristic information and local characteristic information through the characteristic fusion of the grading labels; and distributing the global characteristic information and the local characteristic information according to the myocardial segment classification indexes. The deep neural network model realizes feature hierarchical fusion exchange through feature fusion exchange, so that each task can be classified based on local semantic information of the current task and global semantic information of fusion features of all tasks.
Example 2
As shown in fig. 2, a deep neural network model is constructed, scoring tag features are obtained, feature fusion exchange is performed on the scoring tags, input information is processed into a feature map by the deep neural network model and is transmitted between network layers, and a prediction result is finally obtained. Compared with a shallow neural network, the deep neural network has more network layers, stronger learning ability and can fit a task with higher difficulty.
Network framework: constructing the deep neural network model includes removing the full-link layer and the pooling layer of Resnet50 as a feature extraction skeleton of the deep neural network model
Multitask and multiple output: the features extracted by Resnet50 were the same for all tasks, so in order to make the features of the scoring labels task specific, two outputs were made in sequence for the 17 classification tasks. The first task output enables the scoring label features to have task specificity before entering feature exchange, and the second task output enables more accurate output based on results of the specific features and the fusion features.
The feature fusion exchange module: since the division of 17 heart segments is only for the convenience of quantitative diagnosis, and in practice, the myocardial defects are usually of many heart segments, there is a certain internal correlation between different heart segment defects. In addition, there is a stronger intrinsic correlation between myocardial segments belonging to a region. Based on the hierarchical relation of myocardial division, a feature fusion exchange module is provided, the feature graph of each myocardial segment scoring task is input into a myocardial segment selection module, and fusion is carried out according to respective region attribution: features from myocardial segments 1, 2, 7, 8, 13, 14, 17 are fused into LAD features, features from myocardial segments 3, 4, 9, 10, 15 are fused into LCX features, and features from myocardial segments 5, 6, 11, 12, 16 are fused into RCA features. The region-level features LAD, LCX, and RCA are then fused into case-level features. Then according to the fusion sequence, the fused case characteristic diagram is fused with the area level characteristics LAD, LCX and RCA respectively to obtain new characteristics LAD, LCX and RCA including global characteristic information and local characteristic information, and then the characteristics are distributed through a myocardial segment distribution module based on the myocardial segment affiliation relationship: the features LAD are assigned to the myocardial segments 1, 2, 7, 8, 13, 14, 17, the features LCX to the myocardial segments 3, 4, 9, 10, 15, and the features RCA to the myocardial segments 5, 6, 11, 12, 16. After the deep neural network model is trained, 17 myocardial segment scores can be output only by inputting a myocardial perfusion imaging graph, the deep neural network model realizes automatic myocardial segment defect scoring, and after the deep neural network model is trained, a myocardial perfusion imaging diagnosis result can be quickly obtained, so that the working efficiency is effectively improved.
Example 3
As shown in fig. 3, after the deep neural network model is constructed, the model may be trained. The goal of the training is to find a suitable set of weights such that the network minimizes the objective function. The specific steps include forward calculation, weight updating and loss function definition:
forward calculation: in general, for a L-layer feedforward neural network, the training sample set is set to X ∈ Rm×nWhere m is the dimension of a single sample and n represents the number of training samples, then the ith sample may be represented as Xi. Let the j-th neuron of the l-th layer to the k-th neuron of the l +1 layer be connected as the weight
Then, the l-th to l + 1-th connection weight matrix W(l). Setting the activation function of the neuron on the l-th layer as f (-) and continuously carrying out forward calculation from the input layer to the output layer, wherein the process is as follows:
wherein, ai lRepresenting the activation values of layer i neurons for the i sample. Then, the activation values for the network output layer neurons are:
ai L＝f(L-1)(W(L-1)·f(L-2)(W(L-2)…ai 0))
updating the weight value: neural networks typically employ cross entropy as an objective function for classification/segmentation tasks, which is defined as follows:
wherein
And diRespectively representing the output and label of the last layer of the network. The network can continuously reduce the value of the target function by solving the gradient of the target function J to the weight and iterating and adopting a gradient descent algorithm, thereby finding a group of proper weights. The gradient descent algorithm is as follows:
where α represents the learning rate constant.
The loss function defines: the invention has a plurality of classification tasks and two outputs before and after, so the loss function is defined as follows:
where M is the number of myocardial segments, which in the present invention is 17. L issRepresenting a loss function of the second output, L'sRepresenting the loss functions of the first output, which are cross entropy loss functions. λ is a hyper-parameter used to balance the ratio, and in the present invention λ is 0.5.
And (3) testing a model: after the deep neural network model training is completed, the recognition effect of the model on the test set needs to be evaluated quantitatively. Commonly used evaluation indices include accuracy, sensitivity, specificity, F-1score, precision, which are defined as follows:
wherein TP, FP, FN and TN represent true positive number, false negative number and true negative number respectively.
The above embodiments are provided to further explain the objects, technical solutions and advantages of the present invention in detail, it should be understood that the above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the scope of the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A myocardial segment defect image processing method based on a deep neural network is characterized by comprising the following steps:
s1, acquiring myocardial perfusion imaging image data, and performing data preprocessing on an ROI (region of interest) region and a scoring label in the image data to obtain a training set and a test set;
s2, constructing a deep neural network model, acquiring scoring label characteristics, and performing characteristic fusion exchange on the scoring labels;
s3, inputting myocardial perfusion imaging image data preprocessing data into the loaded deep neural network model, outputting myocardial segment defect scoring results, updating the network weight through a back propagation algorithm, and performing multiple rounds of iterative training of deep neural network model parameters;
and S4, inputting the training set into the trained deep neural network model, outputting the myocardial segment score according to the myocardial segment classification index, and evaluating the myocardial segment score output result through the deep neural network model.
2. The method for processing the myocardial segment defect image based on the deep neural network as claimed in claim 1, wherein the data preprocessing comprises:
extracting an ROI area and a scoring tag in the image data;
dividing the preprocessed image data into a training set and a test set;
and performing data augmentation on the test set.
3. The method for processing the myocardial segment defect image based on the deep neural network as claimed in claim 2, wherein the method for extracting the ROI area in the image data is to use a myocardial perfusion imaging image data section as a network input;
the method for extracting the scoring label in the image data comprises the step of extracting the myocardial segment scoring label through a regular expression, wherein the scoring label is text information contained in diagnosis information of a myocardial perfusion imaging report.
4. The deep neural network-based myocardial segment defect image processing method of claim 2, wherein the dividing the extracted image data into a training set and a test set comprises dividing the image data into a training set and a test set in a ratio of 4: 1.
5. The myocardial segment defect image processing method based on the deep neural network as claimed in claim 1, wherein the constructing the deep neural network model comprises removing a full connection layer and a pooling layer of Resnet50 as a feature extraction skeleton of the deep neural network model.
6. The deep neural network-based myocardial segment defect image processing method according to claim 1, wherein the obtaining of the score label features comprises performing multitask and output on the divided score labels, and the multitask and output comprises:
the first task output is used for enabling the scoring label characteristics to have task specificity before the characteristic fusion exchange is carried out;
the second task output is used for outputting results based on the specific characteristics and the fusion characteristics of the scoring labels.
7. The myocardial segment defect image processing method based on the deep neural network as claimed in claim 1, wherein the feature fusion exchanging of the score labels specifically comprises:
fusing scoring label features with task specificity according to the regional affiliation of the myocardial segments;
obtaining global characteristic information and local characteristic information through the characteristic fusion of the grading labels;
and distributing the global characteristic information and the local characteristic information according to the myocardial segment classification indexes.
8. The myocardial segment defect image processing method based on the deep neural network as claimed in claim 1, wherein the parameters of the trained deep neural network model specifically include forward calculation, weight update and loss function definition, and the parameters of the trained deep neural network model are used to update network weights for finding a set of suitable weights, so that the network minimizes the objective function.
9. The method for processing the myocardial segment defect image based on the deep neural network as claimed in claim 8, wherein the forward calculation specifically includes:
continuously carrying out forward calculation from an input layer to an output layer, wherein the process is as follows:
the activation values for the network output layer neurons are:
ai L＝f(L-1)(W(L-1)·f(L-2)(W(L-2)…ai 0))
wherein, for a feedforward neural network of L layer, the training sample set is set as X ∈ Rm×nWhere m is the dimension of a single sample, n represents the number of training samples, XiFor the (i) th sample,
connecting weight value W from jth neuron of ith layer to kth neuron of l +1 layer(l)A connection weight matrix from layer l to layer l +1, f (-) is an activation function of neurons on layer l, ai lThe activation value for the ith sample for layer I neurons.
10. The method of claim 8, wherein the loss function is defined as:
where M is the number of myocardial segments, and M is 17. L issRepresenting a loss function of the second output, L'sAnd representing loss functions of the first output, wherein the loss functions are cross entropy loss functions, lambda is a hyperparameter used for balancing proportion, and lambda is 0.5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210238690.9A CN114581425B (en) | 2022-03-10 | 2022-03-10 | Myocardial segment defect image processing method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210238690.9A CN114581425B (en) | 2022-03-10 | 2022-03-10 | Myocardial segment defect image processing method based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114581425A true CN114581425A (en) | 2022-06-03 |
| CN114581425B CN114581425B (en) | 2022-11-01 |
Family
ID=81775615
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210238690.9A Active CN114581425B (en) | 2022-03-10 | 2022-03-10 | Myocardial segment defect image processing method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114581425B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115082409A (en) * | 2022-06-30 | 2022-09-20 | 心禾互联网医院(成都)有限公司 | System for automatically identifying nuclide image and diagnosing myocardial ischemia |
Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680192A (en) * | 2015-02-05 | 2015-06-03 | 国家电网公司 | Electric power image classification method based on deep learning |
| CN109192305A (en) * | 2018-09-12 | 2019-01-11 | 中国人民解放军陆军军医大学第二附属医院 | A kind of heart function automatic analysis method based on deep-cycle neural network |
| CN109614991A (en) * | 2018-11-19 | 2019-04-12 | 成都信息工程大学 | A kind of segmentation and classification method of the multiple dimensioned dilatancy cardiac muscle based on Attention |
| CN109886346A (en) * | 2019-02-26 | 2019-06-14 | 四川大学华西医院 | A kind of cardiac muscle MRI image categorizing system |
| CN109919059A (en) * | 2019-02-26 | 2019-06-21 | 四川大学 | Conspicuousness object detecting method based on depth network layerization and multitask training |
| CN109948526A (en) * | 2019-03-18 | 2019-06-28 | 北京市商汤科技开发有限公司 | Image processing method and device, detection device and storage medium |
| CN110197713A (en) * | 2019-05-10 | 2019-09-03 | 上海依智医疗技术有限公司 | A kind of processing method of medical imaging, device, equipment and medium |
| CN110472817A (en) * | 2019-07-03 | 2019-11-19 | 西北大学 | A kind of XGBoost of combination deep neural network integrates credit evaluation system and its method |
| CN110766691A (en) * | 2019-12-06 | 2020-02-07 | 北京安德医智科技有限公司 | Method and device for cardiac magnetic resonance image analysis and cardiomyopathy prediction |
| CN111680701A (en) * | 2020-05-07 | 2020-09-18 | 北京三快在线科技有限公司 | Training method and device of image recognition model and image recognition method and device |
| CN111723600A (en) * | 2019-03-18 | 2020-09-29 | 四川大学 | Pedestrian re-recognition feature descriptor based on multi-task learning |
| CN112069961A (en) * | 2020-08-28 | 2020-12-11 | 电子科技大学 | Few-sample document layout analysis method based on metric learning |
| CN112669320A (en) * | 2021-03-22 | 2021-04-16 | 四川大学 | SPECT thyroid imaging intelligent identification method based on deep neural network |
| CN113469122A (en) * | 2021-07-21 | 2021-10-01 | 北京师范大学 | Deep learning based crop space-time generalization classification method and system |
| CN113673535A (en) * | 2021-05-24 | 2021-11-19 | 重庆师范大学 | Image description generation method of multi-modal feature fusion network |
| CN113887675A (en) * | 2021-12-06 | 2022-01-04 | 四川大学 | Gesture recognition method based on feature fusion of heterogeneous sensors |
| CN114155474A (en) * | 2021-12-12 | 2022-03-08 | 菅政 | Damage identification technology based on video semantic segmentation algorithm |
-
2022
- 2022-03-10 CN CN202210238690.9A patent/CN114581425B/en active Active
Patent Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104680192A (en) * | 2015-02-05 | 2015-06-03 | 国家电网公司 | Electric power image classification method based on deep learning |
| CN109192305A (en) * | 2018-09-12 | 2019-01-11 | 中国人民解放军陆军军医大学第二附属医院 | A kind of heart function automatic analysis method based on deep-cycle neural network |
| CN109614991A (en) * | 2018-11-19 | 2019-04-12 | 成都信息工程大学 | A kind of segmentation and classification method of the multiple dimensioned dilatancy cardiac muscle based on Attention |
| CN109886346A (en) * | 2019-02-26 | 2019-06-14 | 四川大学华西医院 | A kind of cardiac muscle MRI image categorizing system |
| CN109919059A (en) * | 2019-02-26 | 2019-06-21 | 四川大学 | Conspicuousness object detecting method based on depth network layerization and multitask training |
| CN111723600A (en) * | 2019-03-18 | 2020-09-29 | 四川大学 | Pedestrian re-recognition feature descriptor based on multi-task learning |
| CN109948526A (en) * | 2019-03-18 | 2019-06-28 | 北京市商汤科技开发有限公司 | Image processing method and device, detection device and storage medium |
| CN110197713A (en) * | 2019-05-10 | 2019-09-03 | 上海依智医疗技术有限公司 | A kind of processing method of medical imaging, device, equipment and medium |
| CN110472817A (en) * | 2019-07-03 | 2019-11-19 | 西北大学 | A kind of XGBoost of combination deep neural network integrates credit evaluation system and its method |
| CN110766691A (en) * | 2019-12-06 | 2020-02-07 | 北京安德医智科技有限公司 | Method and device for cardiac magnetic resonance image analysis and cardiomyopathy prediction |
| CN111680701A (en) * | 2020-05-07 | 2020-09-18 | 北京三快在线科技有限公司 | Training method and device of image recognition model and image recognition method and device |
| CN112069961A (en) * | 2020-08-28 | 2020-12-11 | 电子科技大学 | Few-sample document layout analysis method based on metric learning |
| CN112669320A (en) * | 2021-03-22 | 2021-04-16 | 四川大学 | SPECT thyroid imaging intelligent identification method based on deep neural network |
| CN113673535A (en) * | 2021-05-24 | 2021-11-19 | 重庆师范大学 | Image description generation method of multi-modal feature fusion network |
| CN113469122A (en) * | 2021-07-21 | 2021-10-01 | 北京师范大学 | Deep learning based crop space-time generalization classification method and system |
| CN113887675A (en) * | 2021-12-06 | 2022-01-04 | 四川大学 | Gesture recognition method based on feature fusion of heterogeneous sensors |
| CN114155474A (en) * | 2021-12-12 | 2022-03-08 | 菅政 | Damage identification technology based on video semantic segmentation algorithm |
Non-Patent Citations (6)
| Title |
|---|
| LISE WEI 等: ""Fundamentals of radiomics in nuclear medicine and hybrid imaging"", 《BASIC SCIENCES OF NUCLEAR MEDICINE》 * |
| MOLOUD ABDAR 等: ""A new machine learning technique for an accurate diagnosis of coronary artery disease"", 《COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE》 * |
| 余浪: ""基于BP神经网络模型的缺血性心脏病住院医疗费用研究"", 《中国优秀博硕士学位论文全文数据库医药卫生科技辑》 * |
| 刘科伶 等: """一站式"CT心肌灌注成像在评估冠状动脉重度狭窄患者中的应用价值"", 《四川大学学报》 * |
| 刘静静: ""基于MR图像和三维超声心动图的左心室运动跟踪研究"", 《中国优秀博硕士学位论文全文数据库医药卫生科技辑》 * |
| 周轩 等: ""冠状动脉慢性闭塞病变介入治疗的效果影响因素及门控核素心肌灌注显像技术的评估价值"", 《昆明医科大学学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115082409A (en) * | 2022-06-30 | 2022-09-20 | 心禾互联网医院(成都)有限公司 | System for automatically identifying nuclide image and diagnosing myocardial ischemia |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114581425B (en) | 2022-11-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Deep Rank-Based Average Pooling Network for Covid-19 Recognition. | |
| Warren et al. | Functional magnetic resonance imaging, deep learning, and Alzheimer's disease: A systematic review | |
| CN109192305B (en) | Heart function automatic analysis method based on deep circulation neural network | |
| Ligabue et al. | Evaluation of the classification accuracy of the kidney biopsy direct immunofluorescence through convolutional neural networks | |
| Kollias et al. | Ai-enabled analysis of 3-d ct scans for diagnosis of covid-19 & its severity | |
| Sridhar et al. | A Torn ACL Mapping in Knee MRI Images Using Deep Convolution Neural Network with Inception‐v3 | |
| Brandt et al. | Ischemia and outcome prediction by cardiac CT based machine learning | |
| CN114581425B (en) | Myocardial segment defect image processing method based on deep neural network | |
| Seetharam et al. | Clinical inference from cardiovascular imaging: paradigm shift towards machine-based intelligent platform | |
| Badano et al. | Artificial intelligence and cardiovascular imaging: A win-win combination. | |
| Li et al. | Research about tongue image of traditional chinese medicine (TCM) based on artificial intelligence technology | |
| Nagadeepa et al. | Artificial intelligence based cervical cancer risk prediction using m1 algorithms | |
| Lin et al. | Echocardiography-based AI detection of regional wall motion abnormalities and quantification of cardiac function in myocardial infarction | |
| Gulati et al. | Comparative analysis of deep learning approaches for the diagnosis of diabetic retinopathy | |
| Reddy Enumula et al. | Alzheimer's disease prediction and classification using CT images through machine learning. | |
| El Rafaie et al. | On the use of SPECT imaging datasets for automated classification of ventricular heart disease | |
| Cano-Espinosa et al. | On the relevance of the loss function in the Agatston score regression from non-ECG gated CT scans | |
| Xiao et al. | A deep learning model of automatic detection of pulmonary nodules based on convolution neural networks (CNNs) | |
| Kupas et al. | Solving the problem of imbalanced dataset with synthetic image generation for cell classification using deep learning | |
| Sukumar et al. | Deep Learning based Pulmonary Embolism Detection using Convolutional Feature Maps of CT Pulmonary Angiography Images | |
| Xiao et al. | Learning motion based auxiliary task for cardiomyopathy recognition with cardiac magnetic resonance images | |
| Cakir et al. | Automated Aortic Valve Calcific Area Segmentation in Echocardiography Images Using Fully Convolutional Neural Networks | |
| Jin et al. | Lesion Classification of Coronary Artery CTA Images Based on CBAM and Transfer Learning | |
| Rangraz Jeddi et al. | Machine Learning Approaches for Detecting Coronary Artery Disease Using Angiography Imaging: A Scoping Review | |
| Promit et al. | Detection of coronary artery blockage at an early stage using effective deep learning technique |
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 |