CN108492286B - Medical image segmentation method based on dual-channel U-shaped convolutional neural network - Google Patents

Medical image segmentation method based on dual-channel U-shaped convolutional neural network Download PDF

Info

Publication number
CN108492286B
CN108492286B CN201810203917.XA CN201810203917A CN108492286B CN 108492286 B CN108492286 B CN 108492286B CN 201810203917 A CN201810203917 A CN 201810203917A CN 108492286 B CN108492286 B CN 108492286B
Authority
CN
China
Prior art keywords
data
channel
neural network
dual
training set
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.)
Expired - Fee Related
Application number
CN201810203917.XA
Other languages
Chinese (zh)
Other versions
CN108492286A (en
Inventor
于曦
胡科
刘昶
何煜
朱泓超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chengdu University
Original Assignee
Chengdu University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chengdu University filed Critical Chengdu University
Priority to CN201810203917.XA priority Critical patent/CN108492286B/en
Publication of CN108492286A publication Critical patent/CN108492286A/en
Application granted granted Critical
Publication of CN108492286B publication Critical patent/CN108492286B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (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 medical image segmentation method based on a dual-channel U-shaped convolutional neural network, which well fuses a dual-channel module and the U-shaped network together, reuses original features through stacking operation, and simultaneously adds the features to generate new features, thereby improving the training precision of the network; meanwhile, the U-shaped convolutional neural network architecture is utilized to perform transverse feature stacking between blocks, so that the convergence rate of network training is greatly improved, and an ideal training model is obtained more quickly. The invention can meet the requirements of larger data volume to be processed and higher requirements on processing efficiency and precision of medical images, and realizes very good performance in biomedical detection application.

Description

Medical image segmentation method based on dual-channel U-shaped convolutional neural network
Technical Field
The invention belongs to the technical field of medical image segmentation, and particularly relates to a design of a medical image segmentation method based on a dual-channel U-shaped convolutional neural network.
Background
The medical image segmentation comprises manual segmentation, semi-automatic segmentation and full-automatic segmentation, the difficulty of the manual segmentation is high, the process is complicated, although the precision is high, the method depends on the experience and knowledge of an operator to a great extent, and the segmentation result is difficult to reproduce; semi-automatic segmentation is an interactive way that combines manual and computer processing, allowing manual interactive operations to provide some useful information, which is then processed by the computer for segmentation. The semi-automatic segmentation also comprises methods such as Gragh Cut, CRF (conditional random field) and level set, and all the methods need manual input, and one point is designated as a seed point to guide subsequent segmentation. The full-automatic segmentation method overcomes the defects of semi-automatic segmentation and manual segmentation, and obtains better segmentation effect.
The traditional image segmentation technology needs manual input no matter manual segmentation or semi-automatic segmentation, the process is complicated, the process depends on experience and knowledge of an operator to a great extent, the segmentation result is difficult to reproduce, and the efficiency and the precision of full-automatic segmentation cannot be achieved completely. The existing full-automatic segmentation method can be limited on a 2-dimensional level, and the existing three-dimensional network structure is still to be optimized. Thus, an objective criterion for accurately evaluating the success or failure of segmentation does not exist so far.
Disclosure of Invention
The invention aims to provide a medical image segmentation method based on a dual-channel U-shaped convolutional neural network, which improves the accuracy of image segmentation to a pixel level and improves the accuracy of subsequent medical diagnosis.
The technical scheme of the invention is as follows: a medical image segmentation method based on a dual-channel U-shaped convolution neural network comprises the following steps:
and S1, preprocessing the functional magnetic resonance image data to be segmented to obtain training set data and test set data.
Step S1 includes the following substeps:
and S11, performing format conversion on the functional magnetic resonance image data to be segmented.
And S12, normalizing the image after format conversion to a [0,1] interval.
And S13, proportionally dividing the normalized image into training set data and test set data.
And S14, performing data enhancement on the training set data.
The method for enhancing the data comprises the following steps:
and (3) scale transformation: and (3) amplifying or reducing the image according to a specified scale factor, or constructing a scale space by filtering the image by using the specified scale factor according to the SIFT feature extraction idea, and changing the size or the fuzzy degree of the image content.
Translation transformation: the image is translated in the image plane.
Rotation transformation: the image is flipped in either the horizontal or vertical direction.
Scaling transformation: the image is enlarged or reduced according to a set scale.
And S2, constructing a dual-channel U-shaped convolutional neural network.
Step S2 includes the following substeps:
and S21, constructing a dual-channel BO module.
The dual-channel BO module comprises a Sum channel and a density channel, wherein the Sum channel is used for realizing the addition of characteristic values of D1 data and C1 data, and the density channel is used for realizing the stacking of D2 data and C2 data;
the acquisition method of the D1 data and the D2 data comprises the following steps: after the original characteristic diagram data is subjected to BN, activation function and convolution operation, the original characteristic diagram data is divided into two parts according to channels, namely D1 data and D2 data;
the method for acquiring the C1 data and the C2 data comprises the following steps: synthesizing D1 data and D2 data into new feature map data, processing the new feature map data by BN, an activation function, convolution and dropout, and dividing the new feature map data into two parts according to channels, namely C1 data and C2 data;
dropout processing refers to the operation of randomly dropping neural network elements from the network temporarily at a set probability to prevent overfitting.
And S22, constructing a dual-channel U-shaped convolutional neural network according to the dual-channel BO module.
The dual-channel U-shaped convolutional neural network comprises a left side unit and a right side unit;
the left unit comprises N double-channel BO modules, N is more than or equal to 2, and pooling operation is performed between every two adjacent double-channel BO modules in the left unit;
the right side unit comprises N double-channel BO modules, and in the right side unit, up-sampling operation is carried out between every two adjacent double-channel BO modules;
the dual-channel BO modules in the left unit and the dual-channel BO modules in the right unit are in one-to-one correspondence and perform stacking reuse characteristic operation with each other.
And S3, inputting the training set data into a dual-channel U-shaped convolutional neural network for training to obtain a learned convolutional neural network model.
Step S3 includes the following substeps:
s31, dividing the training set data into m batches, and initializing the derivative of the convolution kernel weight and the bias value to the loss function as 0, namely:
ΔW(l)＝0 (1)
Δb(l)＝0 (2)
wherein Δ W(l)Representing the derivative of the convolution kernel weights in the l-th convolutional/deconvolution layer on the loss function, Δ b(l)The derivative of the convolution kernel bias value in the l-th convolutional layer/deconvolution layer to the loss function is represented.
And S32, randomly selecting a batch of untrained training set data to be input into the dual-channel U-shaped convolutional neural network.
And S33, calculating the training set data and each node parameter of the subsequent network unit in the dual-channel U-shaped convolutional neural network, realizing the forward propagation of network training, and outputting a prediction probability map.
S34, calculating the error between the prediction probability graph and the positive sample in the training set data, wherein the calculation formula is as follows:
Figure BDA0001595325790000031
wherein L represents the error between the prediction probability graph and the positive sample in the training set data, M represents the number of the positive samples in the training set data, and the positive sample is an artificial marking area obtained by extracting the characteristics of the training set data by taking a block as a unit; x represents the number of network elements in the neural network, y represents positive sample data in the training set data, a represents the predictive probability map data, and
a＝W(l)*x+b(l)(4)
s35, respectively calculating the partial derivatives of the convolution kernel weight and the partial value to the error function L by adopting a gradient descent method
Figure BDA0001595325790000032
And
Figure BDA0001595325790000033
and adds it to Δ W(l)And Δ b(l)For Δ W(l)And Δ b(l)Updating, wherein the updating formula is as follows:
Figure BDA0001595325790000034
Figure BDA0001595325790000035
wherein Δ W(l)′And Δ b(l)′Representing the convolution kernel weights and the derivatives of the bias values to the loss function before updating, respectively.
And S36, judging whether all training set data are input into a dual-channel U-shaped convolutional neural network for training, if so, completing one iteration, entering the step S37, and otherwise, returning to the step S32.
S37, according to Δ W(l)And Δ b(l)Using batch gradient descent algorithm to weight W of convolution kernel(l)Sum bias value b(l)Updating, wherein the updating formula is as follows:
Figure BDA0001595325790000036
Figure BDA0001595325790000037
wherein W(l)′And b(l)′Respectively representing the weights and bias values of convolution kernels before updating, m representing the batch number of the training set data, α representing the learning rate, and λ representing the kinetic energy.
And S38, judging whether the current iteration number reaches a preset iteration number threshold, if so, entering the step S39, and if not, returning to the step S32 to carry out the next iteration.
S39, the learned convolutional neural network model is output, and the process proceeds to step S4.
And S4, inputting the test set data into the learned convolutional neural network model, and outputting an image segmentation result.
The invention has the beneficial effects that: according to the invention, the dual-channel module and the U-shaped network are well fused together, original features are reused through stacking operation, and new features are generated by adding the features, so that the training precision of the network is improved; meanwhile, the U-shaped convolutional neural network architecture is utilized to perform transverse feature stacking between blocks, so that the convergence rate of network training is greatly improved, and an ideal training model is obtained more quickly; the invention achieves very good performance in biomedical detection applications.
Drawings
Fig. 1 is a flowchart of a medical image segmentation method based on a dual-channel U-shaped convolutional neural network according to an embodiment of the present invention.
Fig. 2 is a flowchart illustrating the step S1 according to an embodiment of the present invention.
Fig. 3 is a schematic structural diagram of a dual-channel BO module according to an embodiment of the present invention.
Fig. 4 is a schematic structural diagram of a dual-channel U-shaped convolutional neural network provided in an embodiment of the present invention.
Fig. 5 is a flowchart illustrating the step S3 according to an embodiment of the present invention.
Fig. 6 is a schematic diagram illustrating a prediction probability map and a segmentation result according to an embodiment of the present invention.
Fig. 7 is a schematic diagram illustrating the accuracy and loss of the training set data and the test set data according to the embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It is to be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, not to limit the scope of the invention.
The embodiment of the invention provides a medical image segmentation method based on a dual-channel U-shaped convolutional neural network, which comprises the following steps of S1-S4 as shown in FIG. 1:
and S1, preprocessing the functional Magnetic Resonance Image (MRI) data to be segmented to obtain training set data and test set data.
As shown in FIG. 2, step S1 includes the following substeps S11-S14:
and S11, performing format conversion on the functional magnetic resonance image data to be segmented.
And S12, normalizing the image after format conversion to a [0,1] interval.
And S13, proportionally dividing the normalized image into training set data and test set data.
And S14, performing data enhancement on the training set data.
Deep learning usually requires a large amount of data for support, and after training set data and test set data are obtained, data enhancement is required for increasing the data amount, avoiding overfitting and improving segmentation precision. In the embodiment of the invention, the amount of input data can be increased by geometric transformation of the image by using one or more of the following combined data enhancement transformations:
and (3) scale transformation: and (3) amplifying or reducing the image according to a specified scale factor, or constructing a scale space by filtering the image by using the specified scale factor according to the SIFT feature extraction idea, and changing the size or the fuzzy degree of the image content.
Translation transformation: the image is translated in the image plane.
Rotation transformation: the image is flipped in either the horizontal or vertical direction.
Scaling transformation: the image is enlarged or reduced according to a set scale.
And S2, constructing a dual-channel U-shaped convolutional neural network.
Step S2 includes the following substeps S21-S22:
and S21, constructing a dual-channel BO module.
As shown in fig. 3, the dual channel BO module includes a Sum channel for realizing the eigenvalue addition of the D1 data and the C1 data, and a density channel for realizing the stacking (size addition) of the D2 data and the C2 data.
The acquisition method of the D1 data and the D2 data comprises the following steps: the original feature map data is divided into two parts, namely D1 data and D2 data, according to channels after being processed by BN (batch normalization), ReLU (activation function) and convolution operation.
The method for acquiring the C1 data and the C2 data comprises the following steps: and synthesizing the D1 data and the D2 data into new feature map data, and dividing the feature map data into two parts according to channels after BN, activation function, convolution and dropout processing, namely C1 data and C2 data.
dropout processing refers to the operation of randomly dropping neural network elements from the network temporarily at a set probability to prevent overfitting.
And S22, constructing a dual-channel U-shaped convolutional neural network according to the dual-channel BO module.
As shown in fig. 4, the dual-path U-shaped convolutional neural network includes a left side unit and a right side unit.
The left side unit comprises N double-channel BO modules, N is larger than or equal to 2, and in the left side unit, pooling operation is performed between every two adjacent double-channel BO modules.
The right side unit comprises N double-channel BO modules, and in the right side unit, up-sampling operation is performed between every two adjacent double-channel BO modules.
The dual-channel BO modules in the left unit and the dual-channel BO modules in the right unit are in one-to-one correspondence and perform stacking reuse characteristic operation with each other.
And S3, inputting the training set data into a dual-channel U-shaped convolutional neural network for training to obtain a learned convolutional neural network model.
As shown in FIG. 5, step S3 includes the following substeps S31-S39:
s31, dividing the training set data into m batches, and initializing the derivative of the convolution kernel weight and the bias value to the loss function as 0, namely:
ΔW(l)＝0 (1)
Δb(l)＝0 (2)
wherein Δ W(l)Representing the derivative of the convolution kernel weights in the l-th convolutional/deconvolution layer on the loss function, Δ b(l)The derivative of the convolution kernel bias value in the l-th convolutional layer/deconvolution layer to the loss function is represented.
And S32, randomly selecting a batch of untrained training set data to be input into the dual-channel U-shaped convolutional neural network.
And S33, calculating the training set data and each node parameter of the subsequent network unit in the dual-channel U-shaped convolutional neural network, realizing the forward propagation of network training, and outputting a prediction probability map.
And S34, calculating the error between the prediction probability graph and the positive sample in the training set data.
In the calculation of the neural network, the difference between the score P1 calculated according to the feature map of the forward propagation output of the neural network and the score P2 calculated according to the real label needs to be calculated, and the cross entropy Loss function Loss is calculated, so that the backward propagation can be applied. In the embodiment of the invention, the error between the prediction probability graph and the positive sample in the training set data is calculated by using a cross entropy loss function, and the calculation formula is as follows:
Figure BDA0001595325790000061
in the embodiment of the invention, after the features of the training set data are extracted by taking a block as a unit, an artificial mark area is used as a positive sample, and a background area is used as a negative sample. x represents the number of network elements in the neural network, y represents the expected output data, which is the positive sample data in the training set data in the embodiment of the invention, a represents the actual output data of the neural network elements, which is the prediction probability map data in the embodiment of the invention
a＝W(l)*x+b(l)(4)
S35, respectively calculating the partial derivatives of the convolution kernel weight and the partial value to the error function L by adopting a gradient descent method
Figure BDA0001595325790000062
And
Figure BDA0001595325790000063
and adds it to Δ W(l)And Δ b(l)For Δ W(l)And Δ b(l)Updating, wherein the updating formula is as follows:
Figure BDA0001595325790000064
Figure BDA0001595325790000065
wherein Δ W(l)′And Δ b(l)′Representing the convolution kernel weights and the derivatives of the bias values to the loss function before updating, respectively.
And S36, judging whether all training set data are input into a dual-channel U-shaped convolutional neural network for training, if so, completing one iteration, entering the step S37, and otherwise, returning to the step S32.
S37, according to Δ W(l)And Δ b(l)Using batch gradient descent algorithm to weight W of convolution kernel(l)Sum bias value b(l)Updating, wherein the updating formula is as follows:
Figure BDA0001595325790000066
Figure BDA0001595325790000071
wherein W(l)′And b(l)′Respectively representing the weight and the bias value of the convolution kernel before updating, wherein m represents the batch number of the training set data, α represents the learning rate, and lambda represents the kinetic energy, so that the influence of the last iteration parameter in the parameter updating process is determined.
And S38, judging whether the current iteration number reaches a preset iteration number threshold, if so, entering the step S39, and if not, returning to the step S32 to carry out the next iteration.
S39, the learned convolutional neural network model is output, and the process proceeds to step S4.
And S4, inputting the test set data into the learned convolutional neural network model, and outputting an image segmentation result.
As shown in fig. 6, the test set data is input into the learned convolutional neural network model, and the resulting image segmentation results are shown in the first column of fig. 6. As shown in fig. 7, the precision and loss variation trend of the training set data and the test set data are reflected, and as the number of iterations increases, the precision of the training set data and the precision of the test set data are continuously improved to finally reach 0.98 and 0.99 respectively; the loss of the training set data and the loss of the test set data are continuously reduced and respectively reach 0.04 and 0.03, so that the preset iteration threshold is necessary to carry out a certain number of iterations.
In the embodiment of the present invention, two indexes, namely, a dice value and an assd value, are respectively used to perform image segmentation prediction evaluation on four groups of test data, as shown in table 1, data enhancement and pre-and post-data enhancement are compared in table 1, and it can be seen that a prediction result with data enhancement is obviously superior to a prediction result without data enhancement, so step S14 in the present invention is necessary.
TABLE 1
Figure BDA0001595325790000072
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.

Claims (4)

1. A medical image segmentation method based on a dual-channel U-shaped convolution neural network is characterized by comprising the following steps:
s1, preprocessing functional magnetic resonance image data to be segmented to obtain training set data and test set data;
s2, constructing a double-channel U-shaped convolution neural network;
s3, inputting training set data into a dual-channel U-shaped convolutional neural network for training to obtain a learned convolutional neural network model;
s4, inputting the test set data into the learned convolutional neural network model, and outputting an image segmentation result;
the step S2 includes the following sub-steps:
s21, constructing a dual-channel BO module;
s22, constructing a dual-channel U-shaped convolutional neural network according to the dual-channel BO module;
the dual-channel U-shaped convolutional neural network comprises a left side unit and a right side unit;
the left side unit comprises N double-channel BO modules, N is more than or equal to 2, and pooling operation is performed between every two adjacent double-channel BO modules in the left side unit;
the right side unit comprises N double-channel BO modules, and up-sampling operation is performed between every two adjacent double-channel BO modules in the right side unit;
the dual-channel BO modules in the left unit and the dual-channel BO modules in the right unit are in one-to-one correspondence and perform stacking reuse characteristic operation with each other;
the dual-channel BO module comprises a Sum channel for enabling eigenvalue addition of D1 data and C1 data and a Dense channel for enabling stacking of D2 data and C2 data;
the method for acquiring the D1 data and the D2 data comprises the following steps: after the original characteristic diagram data is subjected to BN, activation function and convolution operation, the original characteristic diagram data is divided into two parts according to channels, namely D1 data and D2 data;
the method for acquiring the C1 data and the C2 data comprises the following steps: synthesizing D1 data and D2 data into new feature map data, processing the new feature map data by BN, an activation function, convolution and dropout, and dividing the new feature map data into two parts according to channels, namely C1 data and C2 data;
the dropout process represents an operation of temporarily dropping neural network elements from the network at random with a set probability to prevent overfitting.
2. A medical image segmentation method as claimed in claim 1, characterized in that said step S1 comprises the sub-steps of:
s11, converting the format of the functional magnetic resonance image data to be segmented;
s12, normalizing the image after format conversion to a [0,1] interval;
s13, dividing the normalized image into training set data and test set data according to the proportion;
and S14, performing data enhancement on the training set data.
3. The medical image segmentation method according to claim 2, wherein the data enhancement in step S14 includes:
and (3) scale transformation: amplifying or reducing the image according to a specified scale factor, or constructing a scale space by filtering the image by using the specified scale factor according to the SIFT feature extraction idea, and changing the size or the fuzzy degree of the image content;
translation transformation: translating the image on the image plane;
rotation transformation: flipping the image in either the horizontal or vertical direction;
scaling transformation: the image is enlarged or reduced according to a set scale.
4. A medical image segmentation method as claimed in claim 1, characterized in that said step S3 comprises the sub-steps of:
s31, dividing the training set data into m batches, and initializing the derivative of the convolution kernel weight and the bias value to the loss function as 0, namely:
ΔW(l)＝0 (1)
Δb(l)＝0 (2)
wherein Δ W(l)Representing the derivative of the convolution kernel weights in the l-th convolutional/deconvolution layer on the loss function, Δ b(l)Representing the derivative of the convolution kernel offset value in the first layer of convolution layer/deconvolution layer to the loss function;
s32, randomly selecting a batch of untrained training set data to input into a dual-channel U-shaped convolutional neural network;
s33, calculating the training set data and each node parameter of the subsequent network unit in the dual-channel U-shaped convolutional neural network, realizing the forward propagation of network training, and outputting a prediction probability graph;
s34, calculating the error between the prediction probability graph and the positive sample in the training set data, wherein the calculation formula is as follows:
Figure FDA0002385523620000021
wherein L represents the error between the prediction probability graph and the positive sample in the training set data, M represents the number of the positive samples in the training set data, and the positive sample is an artificial marking area obtained by extracting the characteristics of the training set data by taking a block as a unit; x represents the number of network elements in the neural network, y represents positive sample data in the training set data, a represents the predictive probability map data, and
a＝W(l)*x+b(l)(4)
s35, respectively calculating the partial derivatives of the convolution kernel weight and the partial value to the error function L by adopting a gradient descent method
Figure FDA0002385523620000022
And
Figure FDA0002385523620000023
and adds it to Δ W(l)And Δ b(l)For Δ W(l)And Δ b(l)Updating, wherein the updating formula is as follows:
Figure FDA0002385523620000024
Figure FDA0002385523620000031
wherein Δ W(l)′And Δ b(l)′Respectively representing the convolution kernel weight and the derivative of the bias value to the loss function before updating;
s36, judging whether all training set data are input into a dual-channel U-shaped convolutional neural network for training, if so, completing one iteration, entering the step S37, otherwise, returning to the step S32;
s37, according to Δ W(l)And Δ b(l)Using batch gradient descent algorithm to weight W of convolution kernel(l)Sum bias value b(l)Updating, wherein the updating formula is as follows:
Figure FDA0002385523620000032
Figure FDA0002385523620000033
wherein W(l)′And b(l)′Respectively representing the weight and the bias value of the convolution kernel before updating, wherein m represents the batch number of the training set data, α represents the learning rate, and lambda represents the kinetic energy;
s38, judging whether the current iteration number reaches a preset iteration number threshold, if so, entering the step S39, otherwise, returning to the step S32 to carry out the next iteration;
s39, the learned convolutional neural network model is output, and the process proceeds to step S4.
CN201810203917.XA 2018-03-13 2018-03-13 Medical image segmentation method based on dual-channel U-shaped convolutional neural network Expired - Fee Related CN108492286B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810203917.XA CN108492286B (en) 2018-03-13 2018-03-13 Medical image segmentation method based on dual-channel U-shaped convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810203917.XA CN108492286B (en) 2018-03-13 2018-03-13 Medical image segmentation method based on dual-channel U-shaped convolutional neural network

Publications (2)

Publication Number Publication Date
CN108492286A CN108492286A (en) 2018-09-04
CN108492286B true CN108492286B (en) 2020-05-05

Family

ID=63338633

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810203917.XA Expired - Fee Related CN108492286B (en) 2018-03-13 2018-03-13 Medical image segmentation method based on dual-channel U-shaped convolutional neural network

Country Status (1)

Country Link
CN (1) CN108492286B (en)

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109389584A (en) * 2018-09-17 2019-02-26 成都信息工程大学 Multiple dimensioned rhinopharyngeal neoplasm dividing method based on CNN
CN109377501A (en) * 2018-09-30 2019-02-22 上海鹰觉科技有限公司 Remote sensing images naval vessel dividing method and system based on transfer learning
CN109523560A (en) * 2018-11-09 2019-03-26 成都大学 A kind of three-dimensional image segmentation method based on deep learning
CN109754403A (en) * 2018-11-29 2019-05-14 中国科学院深圳先进技术研究院 Tumour automatic division method and system in a kind of CT image
CN109636812A (en) * 2018-12-13 2019-04-16 银河水滴科技（北京）有限公司 A kind of Rail Surface and contact net surface image dividing method and device
CN109785344A (en) * 2019-01-22 2019-05-21 成都大学 The remote sensing image segmentation method of binary channel residual error network based on feature recalibration
CN109919954B (en) * 2019-03-08 2021-06-15 广州视源电子科技股份有限公司 Target object identification method and device
CN109919098B (en) * 2019-03-08 2021-06-15 广州视源电子科技股份有限公司 Target object identification method and device
CN109949323B (en) * 2019-03-19 2022-12-20 广东省农业科学院农业生物基因研究中心 Crop seed cleanliness judgment method based on deep learning convolutional neural network
CN110070175B (en) * 2019-04-12 2021-07-02 北京市商汤科技开发有限公司 Image processing method, model training method and device and electronic equipment
CN110335276B (en) * 2019-07-10 2021-02-26 四川大学 Medical image segmentation model, method, storage medium and electronic device
CN110555512B (en) * 2019-07-30 2021-12-03 北京航空航天大学 Data reuse method and device for binary convolution neural network
CN110473243B (en) * 2019-08-09 2021-11-30 重庆邮电大学 Tooth segmentation method and device based on depth contour perception and computer equipment
CN112446888A (en) * 2019-09-02 2021-03-05 华为技术有限公司 Processing method and processing device for image segmentation model
CN111062252B (en) * 2019-11-15 2023-11-10 浙江大华技术股份有限公司 Real-time dangerous goods semantic segmentation method, device and storage device
CN111127504B (en) * 2019-12-28 2024-02-09 中国科学院深圳先进技术研究院 Method and system for segmenting heart medical image of patient with atrial septal occlusion
WO2021137756A1 (en) 2019-12-30 2021-07-08 Medo Dx Pte. Ltd Apparatus and method for image segmentation using a deep convolutional neural network with a nested u-structure
CN111325658B (en) * 2020-02-19 2022-08-23 成都大学 Color image self-adaptive decolorizing method
CN111325750B (en) * 2020-02-25 2022-08-16 西安交通大学 Medical image segmentation method based on multi-scale fusion U-shaped chain neural network
CN111402268B (en) * 2020-03-16 2023-05-23 苏州科技大学 Liver in medical image and focus segmentation method thereof
CN112748382A (en) * 2020-12-15 2021-05-04 杭州电子科技大学 SPEED magnetic resonance imaging method based on CUNet artifact positioning
CN113177981B (en) * 2021-04-29 2022-10-14 中国科学院自动化研究所 Double-channel craniopharyngioma invasiveness classification and focus region segmentation system thereof

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107220980A (en) * 2017-05-25 2017-09-29 重庆理工大学 A kind of MRI image brain tumor automatic division method based on full convolutional network
CN107633486A (en) * 2017-08-14 2018-01-26 成都大学 Structure Magnetic Resonance Image Denoising based on three-dimensional full convolutional neural networks

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107220980A (en) * 2017-05-25 2017-09-29 重庆理工大学 A kind of MRI image brain tumor automatic division method based on full convolutional network
CN107633486A (en) * 2017-08-14 2018-01-26 成都大学 Structure Magnetic Resonance Image Denoising based on three-dimensional full convolutional neural networks

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于卷积神经网络的图像处理与识别算法研究;满凤环;《中国优秀硕士学位论文全文数据库信息科技辑》;20180215(第2期);第20页 *

Also Published As

Publication number Publication date
CN108492286A (en) 2018-09-04

Similar Documents

Publication Publication Date Title
CN108492286B (en) Medical image segmentation method based on dual-channel U-shaped convolutional neural network
CN109584161A (en) The Remote sensed image super-resolution reconstruction method of convolutional neural networks based on channel attention
CN111583285B (en) Liver image semantic segmentation method based on edge attention strategy
CN109005398B (en) Stereo image parallax matching method based on convolutional neural network
CN112233129B (en) Deep learning-based parallel multi-scale attention mechanism semantic segmentation method and device
CN109102498B (en) Method for segmenting cluster type cell nucleus in cervical smear image
CN110349170B (en) Full-connection CRF cascade FCN and K mean brain tumor segmentation algorithm
CN110517272B (en) Deep learning-based blood cell segmentation method
CN111696046A (en) Watermark removing method and device based on generating type countermeasure network
CN111325750A (en) Medical image segmentation method based on multi-scale fusion U-shaped chain neural network
CN115393293A (en) Electron microscope red blood cell segmentation and positioning method based on UNet network and watershed algorithm
CN114972759A (en) Remote sensing image semantic segmentation method based on hierarchical contour cost function
CN114639102B (en) Cell segmentation method and device based on key point and size regression
CN115409846A (en) Colorectal cancer focus region lightweight segmentation method based on deep learning
CN115170801A (en) FDA-deep Lab semantic segmentation algorithm based on double-attention mechanism fusion
CN115953784A (en) Laser coding character segmentation method based on residual error and feature blocking attention
CN114998756A (en) Yolov 5-based remote sensing image detection method and device and storage medium
CN117935289A (en) Diffusion model graphic symbol anomaly identification and correction method based on classifier
CN113436115A (en) Image shadow detection method based on depth unsupervised learning
CN117437423A (en) Weak supervision medical image segmentation method and device based on SAM collaborative learning and cross-layer feature aggregation enhancement
CN116524495A (en) Traditional Chinese medicine microscopic identification method and system based on multidimensional channel attention mechanism
CN116071331A (en) Workpiece surface defect detection method based on improved SSD algorithm
CN110890143B (en) 2D convolution method introducing spatial information
CN114972155A (en) Polyp image segmentation method based on context information and reverse attention
CN114419078A (en) Surface defect region segmentation method and device based on convolutional neural network

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
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200505

Termination date: 20210313