CN107220980B - A kind of MRI image brain tumor automatic division method based on full convolutional network - Google Patents

A kind of MRI image brain tumor automatic division method based on full convolutional network Download PDF

Info

Publication number
CN107220980B
CN107220980B CN201710379095.6A CN201710379095A CN107220980B CN 107220980 B CN107220980 B CN 107220980B CN 201710379095 A CN201710379095 A CN 201710379095A CN 107220980 B CN107220980 B CN 107220980B
Authority
CN
China
Prior art keywords
layer
brain tumor
segmentation
mri image
convolutional
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.)
Active
Application number
CN201710379095.6A
Other languages
Chinese (zh)
Other versions
CN107220980A (en
Inventor
崔少国
毛雷
熊舒羽
刘畅
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.)
Chongqing Normal University
Original Assignee
Chongqing Normal 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 Chongqing Normal University filed Critical Chongqing Normal University
Priority to CN201710379095.6A priority Critical patent/CN107220980B/en
Publication of CN107220980A publication Critical patent/CN107220980A/en
Application granted granted Critical
Publication of CN107220980B publication Critical patent/CN107220980B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06T7/11—Region-based segmentation
    • 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
    • G06T7/10—Segmentation; Edge detection
    • G06T7/174—Segmentation; Edge detection involving the use of two or more images
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30016—Brain
    • 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
    • G06T2207/30096—Tumor; Lesion

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

The present invention provides a kind of MRI image brain tumor automatic division method based on full convolutional network, including the multi-modal MRI image pretreatment of brain tumor, full convolutional network Construction of A Model, network training and arameter optimization and brain tumor Image Automatic Segmentation, the segmentation of MRI image brain tumor is specifically converted into Pixel-level semantic tagger problem, the different information that MRI different modalities are emphasized, by FLAIR, T1, the two-dimentional whole picture slice synthesis four-way input picture of tetra- mode of T1c and T2, using the convolutional layer of trained convolutional neural networks and pond layer as basic characteristic layer, middle layer is constituted in the convolutional layer that rear addition three is equal with full articulamentum, so that middle layer exports coarse segmentation figure corresponding with semantic segmentation categorical measure, deconvolution network is added behind middle layer, for to coarse Segmentation figure carries out interpolation and obtains and the equirotal fine segmentation figure of original image.The present invention does not need manual intervention, effectively increases segmentation precision and efficiency, shortens the training time.

Description

A kind of MRI image brain tumor automatic division method based on full convolutional network
Technical field
The present invention relates to medical image analysis technical fields, and in particular to a kind of MRI image brain based on full convolutional network Tumour automatic division method.
Background technique
Glioma is a kind of Common Brain Tumor for seriously jeopardizing patient vitals, and the most common therapeutic scheme of brain tumor is just It is operation excision.Magnetic resonance imaging (Magnetic resonance imaging, MRI) is shown inside brain with image mode Information is the powerful that medical worker analyzes intracranial tumors.MRI image brain tumor is divided for early diagnosing, treating meter It draws and treatment assessment plays a crucial role.But early stage manual dividing mark mode intricate operation and subjectivity is strong, brain The obscure boundary of glioma and normal tissue is clear, and MRI image itself is by noise, offset field-effect and partial volume effect Deng influence.Therefore design automatic segmentation algorithm is to solve the insufficient best mode of manual markings, this is the development of cutting techniques One of direction.
Brain tumor conventional segmentation methods based on MRI image can substantially be divided into three classes, i.e., based on boundary, based on region with And the method that the two combines.Wherein, the method based on boundary is exactly to identify the boundary of target, the side based on region in the picture Method is exactly the region that mark target is occupied in the picture, and the method that the two combines is exactly will be based on boundary and based on region Method combines, the advantage both taken and the method for avoiding its disadvantage.But the present inventor has found after study, It in these conventional methods, goes to understand image by the subjective consciousness of people, so that specific characteristic information is extracted, as gray scale is believed Breath, texture information and symmetric information etc. realize the segmentation of brain tumor, as a result can only have relatively good segmentation for specific image As a result, thus segmentation result it is excessively coarse and segmentation efficiency it is lower.
With the appearance of artificial intelligence especially deep learning, new direction is provided for the automatic segmentation of brain tumor, tradition Brain tumor dividing method gradually replaced by the mode based on machine learning.Wherein, convolutional neural networks are as supervised learning Representative, can directly learn character representation from data, by layer-by-layer feature extraction, by image from simple edge, angle point Equal low-level image features, successively combination forms more abstract high-level feature, achieves remarkable result in field of image recognition, by It is widely used in Medical Image Processing.The dividing method for being currently based on machine learning is to use image block classification, but this Method needs carry out prediction classification to each pixel in a manner of sliding window, cause to divide low efficiency and may not apply to reality In the clinical medicine of border.Therefore the present inventor provides a kind of more efficient MRI image accurately based on full convolutional network Brain tumor automatic division method.
Summary of the invention
It to go to understand image by the subjective consciousness of people for existing conventional segmentation methods, to extract specific feature letter Breath, such as grayscale information, texture information and symmetric information realize the segmentation of brain tumor, as a result can only have for specific image and compare Good segmentation result, thus segmentation result is excessively coarse and divides the lower technical problem of efficiency, the present invention provides one kind and is based on The MRI image brain tumor automatic division method of full convolutional network, the dividing method can be widely applied for medical image segmentation neck Domain, especially brain tumor segmentation aspect.
In order to solve the above-mentioned technical problem, present invention employs the following technical solutions:
A kind of MRI image brain tumor automatic division method based on full convolutional network, comprising the following steps:
The multi-modal MRI image pretreatment of S1, brain tumor comprising:
S11, field offset correct operation is carried out to two mode MRI images of T1 and T1c；
S12, the MRI image slice for extracting tetra- mode of FLAIR, T1, T1c and T2, in each MRI image slice, It will be greater than the highest gray value that 1% highest gray scale is set as 0.99 times, 0.99 times minimum will be set as less than 1% minimum gray scale Gray value；
S13, data normalization and linear normalization operation are carried out to the gray value of each MRI image slice；
S14, normalized after MRI image same layer tumor biopsy, according to the sequence of FLAIR, T1, T1c and T2 It is combined into four-way image, wherein as training set, verifying integrates and the ratio of test set is 10:1:1；
S2, full convolutional network Construction of A Model comprising:
S21, using transfer learning method, obtain the intermediate features layer conduct of trained convolutional neural networks model The foundation characteristic layer of full convolutional network model, the foundation characteristic layer include four convolutional layer groups and pond layer；
S22, after the foundation characteristic layer, addition three be equal with full articulamentum convolutional layer composition middle layers, make It obtains middle layer and exports coarse segmentation figure corresponding with semantic segmentation categorical measure；
S23, add warp lamination again after middle layer and carry out interpolation and obtain segmentation result 1, segmentation result 1 again with basis The output characteristic pattern of penultimate pond layer is merged in characteristic layer, and deconvolution layer interpolation is carried out after fusion and obtains segmentation knot Fruit 2, and so on, then carry out with third last in foundation characteristic layer and the 4th pond layer merging respectively and warp lamination is inserted Value, finally obtains segmentation result 3 and segmentation result 4, and the segmentation result 4 is corresponding with semantic segmentation categorical measure, and with The equirotal prediction score matrix of original image, the class that each pixel on segmentation result 4 is divided are them in prediction score matrix Index corresponding to middle maximum value forms final segmentation result figure；
S3, network training and arameter optimization: being converted to Pixel-level semantic tagger problem for the segmentation of MRI image brain tumor, Weight by three convolutional layers added in Gaussian random variable initialization step S22, utilizes pretreated four-way figure Picture is as training sample and is entered into the full convolutional network model of construction, then using stochastic gradient descent method as optimization Method minimizes loss function with having supervision, to carry out tuning to the weight for all convolutional layers that full convolutional network model includes Training；
S4, brain tumor Image Automatic Segmentation: the brain tumor four-way image to be split after normalization, which is input to, to be had In the full convolutional network model for optimizing network weight, prediction score matrix corresponding with semantic segmentation categorical measure, root are obtained According to these matrix values, it is partitioned into brain tumor and image of internal structure.
Further, in the step S11, biased field correct operation is carried out using N4ITK method.
Further, in the step S13, data standard is carried out using following formula to the gray value of each MRI image slice Change operation:
Wherein, the gray value of the i-th row j column of the corresponding slice X of x (i, j),And XsIt is the mean value and variance for being sliced X respectively, X ' (i, j) is the standardized gray scale of x (i, j)；
The gray value of each slice of linear transfor carries out linear normalization operation in the range of [0,1], and using following formula:
Wherein, x ' (i, j) is the gray scale after standardization, X'maxAnd X'minRespectively be sliced X standardization after maximum value and Minimum value,It is the gray scale after linear normalization.
Further, in the step S21, the foundation characteristic layer of full convolutional network model is constructed, is to obtain to have trained 16 layers of VGG model first four convolutional layer group and pond layer, it is described using transfer learning method be by trained power It is worth the initial weight of the foundation characteristic layer as full convolutional network model.
Further, in the step S21, in the foundation characteristic layer of four convolution groups building, the number of every group of convolutional layer Respectively 2,2,3,3, the convolution kernel number of every group of convolutional layer are respectively 64,128,256,512, and each convolutional layer is using 3 × 3 Small convolution filter carries out convolution, and each convolutional layer group is 2 × 2 followed by sampling window, the pond layer that step-length is 2.
Further, output characteristic pattern O corresponding to one of them described convolution kernelkIt is calculated using following formula:
Wherein, bkIt is bias term corresponding to k-th of convolution kernel, C is the port number of input data, WkiIt is k-th of convolution I-th channel weight matrix of core,It is convolution operation, X is input data.
Further, the foundation characteristic layer further includes rectification linear unit, and the rectification linear unit is used for convolution kernel Corresponding output characteristic pattern OkIn each value carry out non-linear transfer, the rectification linear unit is defined as follows:
F (x)=max (0, x)
Wherein, f (x) indicates to rectify linear unit function, and x is an input value.
Further, in the step S22, three convolutional layer convolution kernel numbers of addition are respectively 4096,4096,5, convolution Core size is all 1 × 1, and step-length is also all 1.
Further, in the step S23, warp lamination interpolation uses bilinear interpolation method.
Further, in the step S3, optimization method uses cross entropy Classification Loss function, is defined as follows:
Wherein, N is the size of a batch, and l' is truthful data, and l is the predicted vector of softmax function output.
Compared with existing conventional segmentation methods, the MRI image brain tumor provided by the invention based on full convolutional network is automatic Dividing method, including the multi-modal MRI image pretreatment of brain tumor, full convolutional network Construction of A Model, network training and arameter optimization With brain tumor Image Automatic Segmentation step, the segmentation of MRI image brain tumor is specially converted into Pixel-level semantic tagger problem, To the different information that MRI different modalities are emphasized, by the two-dimentional whole picture slice synthesis four-way of tetra- mode of FLAIR, T1, T1c and T2 Road input picture, using the convolutional layer group of trained convolutional neural networks and pond layer as the basis of full convolutional network Characteristic layer so that middle layer exports coarse segmentation figure corresponding with semantic segmentation categorical measure, and adds behind middle layer Add deconvolution network, is obtained and the equirotal segmentation figure of original image for carrying out interpolation to coarse segmentation figure.Thus, the present invention Having the advantages that 1, compared with the dividing method based on image block classification, the present invention does not need artificial intervention not only, but also It is not necessary to consider the setting of tile size, it is an automatic, simple brain tumor dividing method, not only increases segmentation Precision also substantially increases the efficiency of segmentation；2, the present invention can more consider classified pixels using entire slice as input picture Appearance and Space Consistency obtain the higher accuracy of separation；3, present invention utilizes transfer learning methods, obtain trained Initial weight of the intermediate features layer weight of good convolutional neural networks as full convolutional network foundation characteristic layer, shortens training Time has saved data mark cost, has still there is preferable learning performance under Small Sample Size.
Detailed description of the invention
Fig. 1 is that the MRI image brain tumor automatic division method basic procedure provided by the invention based on full convolutional network shows It is intended to.
Fig. 2 is full convolutional network training provided by the invention and test method flow diagram.
Fig. 3 is full convolutional network model schematic provided by the invention.
Fig. 4 is bilinear interpolation method schematic diagram provided by the invention.
Specific embodiment
In order to be easy to understand the technical means, the creative features, the aims and the efficiencies achieved by the present invention, tie below Conjunction is specifically illustrating, and the present invention is further explained.
It please refers to shown in Fig. 1, the present invention provides a kind of MRI image brain tumor based on the full convolutional network side of segmentation automatically Method, comprising the following steps:
The multi-modal MRI image pretreatment of S1, brain tumor comprising:
S11, field offset correct operation is carried out to two mode MRI images of T1 and T1c, can specifically uses N4ITK method Carry out biased field correct operation；
S12, the MRI image slice for extracting tetra- mode of FLAIR, T1, T1c and T2, in each MRI image slice, It will be greater than the highest gray value that 1% highest gray scale is set as 0.99 times, 0.99 times minimum will be set as less than 1% minimum gray scale Gray value (explanation: the highest gray value that the value of 1% highest gray scale is 0.99 times is shown in Baidupedia: gray scale color mode)；
S13, data normalization and linear normalization operation are carried out to the gray value of each MRI image slice, in which: right The gray value of each MRI image slice carries out data normalization operation using following formula:
Wherein, the gray value of the i-th row j column of the corresponding slice X of x (i, j),And XsIt is the mean value and variance for being sliced X respectively, X ' (i, j) is the standardized gray scale of x (i, j)；
The gray value of each slice of linear transfor carries out linear normalization operation in the range of [0,1], and using following formula:
Wherein, x ' (i, j) is the gray scale after standardization, X'maxAnd X'minRespectively be sliced X standardization after maximum value and Minimum value,It is the gray scale after linear normalization.
S14, normalized after MRI image same layer tumor biopsy, according to the sequence of FLAIR, T1, T1c and T2 It is combined into four-way image, thus completes the multi-modal MRI image pretreatment of brain tumor；Wherein as training set, verifying collection and survey The ratio for trying to integrate is 10:1:1, and specific present inventor obtains the four-way image of 12000 patients, wherein training Collection, verifying collect, the particular number of test set is 10000,1000,1000 respectively.
S2, full convolutional network Construction of A Model comprising:
S21, using transfer learning method, obtain the intermediate features layer conduct of trained convolutional neural networks model The foundation characteristic layer of full convolutional network model.
As specific embodiment, in the step S21, the foundation characteristic layer of full convolutional network model is constructed, is to obtain The first four convolutional layer group and pond layer of trained 16 layers of good VGG model, the use transfer learning method is will to have trained Initial weight of the good weight as the foundation characteristic layer of full convolutional network model.
As specific embodiment, please refer to shown in Fig. 3, the present invention is to provide with 4 convolution groups network model, In the step S21, foundation characteristic layer includes the 1st~4 convolutional layer group, i.e. foundation characteristic layer includes convolutional layer group 1, convolution Layer organizes 2, convolutional layer group 3 and convolutional layer group 4, there is a pond layer, the number difference of every group of convolutional layer after each convolutional layer group Be 2,2,3,3, the number of convolution kernel is respectively 64,128,256,512, each convolutional layer using 3 × 3 small convolution filter into Row convolution, and the sampling window of all pond layers is 2 × 2, step-length 2.Wherein, the signal of first convolutional layer forward-propagating is The signal of the bottom-up informations such as the simple edges of original image and angle point, subsequent convolutional layer and pond layer forward-propagating be it is some compared with Abstract combined information, such as the local grain information of image.Since the convolution kernel of all convolutional layers all uses 3 × 3 small nut, It is possible thereby to deeper network is constructed, and connection weight will not abruptly increase.Wherein, output feature corresponding to a convolution kernel Scheme OkIt is calculated using following formula:
Wherein, bkIt is bias term corresponding to k-th of convolution kernel, C is the port number of input data, WkiIt is k-th of convolution I-th channel weight matrix of core,It is convolution operation, X is input data.
As specific embodiment, in order to improve the non-linear expression ability of network, the foundation characteristic layer further includes rectification Linear unit (Rectifier linear units, ReLU), the linear unit R eLU of rectification the swashing as full convolutional network Function living, for by output characteristic pattern O corresponding to convolution kernelkIn each value carry out non-linear transfer, the rectification is linear Unit R eLU is defined as follows:
F (x)=max (0, x)
Wherein, f (x) indicates to rectify linear unit function, and x is an input value；It follows that when input value is bigger than 0, Then functional value is exactly input value itself, it is on the contrary then be 0.
Simultaneously as the output characteristic pattern after convolutional layer convolution may include a large amount of redundancy, therefore, in convolution Possible redundancy feature can be eliminated using maximum value pond (maximum pond layer, MaxPool) operation after output layer, to make Characteristic pattern size must be exported to become smaller and translate the very little of image and deform with invariance.
S22, after the foundation characteristic layer of acquisition, three convolutional layers for being equal with full articulamentum of addition constitute centres Layer, three convolutional layers are convolutional layer 5, convolutional layer 6 shown in Fig. 3 and convolutional layer 7, the last one convolutional layer is among guaranteeing Layer exports coarse segmentation figure corresponding with semantic segmentation categorical measure, and the size of the characteristic pattern exported by middle layer is than original Beginning image is small very much, and the convolution kernel number of the convolutional layer 5, convolutional layer 6 and convolutional layer 7 is respectively 4096,4096,5, convolution Core size is all 1 × 1, and step-length is all 1.
S23, it please refers to shown in Fig. 3, adds warp lamination progress interpolation again after two convolutional layers of addition and divided Result 1 is cut, the probably accurate position of target object can be marked out in segmentation result 1 substantially；Segmentation result 1 again with foundation characteristic The local message of penultimate pond layer (pond layer 3) is merged in layer, and deconvolution layer interpolation is carried out after fusion and is divided Cut result 2；Segmentation result 2 is merged with the local message of third last pond layer (pond layer 2) in foundation characteristic layer again, Deconvolution layer interpolation is carried out after fusion obtains segmentation result 3；Segmentation result 3 again with fourth from the last pond layer in foundation characteristic layer The local message of (pond layer 1) is merged, and deconvolution layer interpolation is carried out after fusion and obtains segmentation result 4；Segmentation knot at this time Fruit 4, segmentation result 3, segmentation result 2 and the segmentation result 1 of opposite front, segmentation result is finer, more acurrate, and and original image The size of picture is identical.
Wherein, the segmentation result 4 has merged reciprocal the of global information that middle layer finally exports and foundation characteristic layer Two, the local message of third and the 4th pond layer, the local message is that the output characteristic pattern of pond layer (exports square Battle array), before merging with local message, need to obtain by similar last convolutional layer 7 effect opposite with semantic segmentation categorical measure The local message answered, the fusion are will to export the corresponding addition of characteristic pattern matrix, and the last output result of full convolutional network is Segmentation result 4 is corresponding with semantic segmentation categorical measure, and with the equirotal prediction score matrix (score of original image Matrixs), wherein the class that each pixel is divided is its index corresponding to maximum value in prediction matrix, to finally obtain Finer and more accurate segmentation result figure.
As specific embodiment, in the step S23, warp lamination interpolation uses bilinear interpolation method, please refers to Fig. 4 It is shown, method includes the following steps:
Step 1: in the linear interpolation of X-direction, in Q12、Q22Middle insertion point R2, in Q11、Q21Middle insertion point R1, calculate Formula are as follows:
Step 2: linear interpolation in the Y direction, passes through the calculated R of the first step1And R2Interpolation calculation goes out P in the Y direction Point, its calculation formula is:
S3, network training and arameter optimization: please referring to shown in Fig. 2, and the segmentation of MRI image brain tumor is converted to Pixel-level Semantic tagger problem passes through three convolutional layers (convolutional layer 5, convolutional layer 6 added in Gaussian random variable initialization step S22 With convolutional layer 7) weight, using pretreated four-way image as training sample and be transported to the full convolution of construction In network model, then had using stochastic gradient descent method as optimization method and minimize loss function with supervising, thus to full volume The weight for all convolutional layers that product network model includes carries out tuning training.As one embodiment, the optimization method uses Cross entropy Classification Loss function, is defined as follows:
Wherein, N is the size of a batch, and l' is truthful data, and l is the predicted vector of softmax function output.
S4, brain tumor Image Automatic Segmentation: please referring to shown in Fig. 2, selects the preferable model of segmentation performance index, will be pre- Treated, and brain tumor four-way image to be split is input to in the full convolutional network model for having optimized network weight, is obtained Prediction score matrix corresponding with semantic segmentation categorical measure is partitioned into brain tumor image and inside according to these matrix values Structural images.
The MRI image brain tumor automatic division method based on full convolutional network provided for a better understanding of the present invention, Technical solution of the present invention is illustrated below with reference to specific embodiment.
It please refers to shown in Fig. 2 and Fig. 3, by taking the MRI brain tumor image of tetra- mode of FLAIR, T1, T1c and T2 as an example, specifically Include the following steps:
1, in the input layer of full convolutional network, using pretreated multi-modal MRI image whole slices as 240 × 240 × 4 input pictures (wherein 4 be input multi-modality images type: FLAIR, T1, T1c and T2, according to actual needs 4 can be into The corresponding adjustment of row)；
2, four-way image is input to the 1st~4 convolutional layer group of full convolutional network, every group of convolutional layer number is respectively 2,2,3,3, convolution kernel number is respectively 64,128,256,512, and each convolutional layer is rolled up using 3 × 3 small convolution filter Product, and other than the pad parameter of first convolutional layer of first convolutional layer group is set as 100, all convolutional layer pad parameters It is both configured to 1 with step-length, keeps image size constant.After each convolutional layer group followed by, sampling window is 2 × 2, step-length 2 Pond layer maximum value pond is carried out to the feature output block of each convolutional layer group, respectively obtain 438 × 438 × 64,219 × 219 × 128,110 × 110 × 256,55 × 55 × 512 characteristic pattern；
3, after the output characteristic image block of the 4th pond layer being input to convolutional layer 5, convolutional layer 6 and convolutional layer 7, output Coarse segmentation figure corresponding with semantic segmentation categorical measure, the coarse segmentation figure are small much relative to original image；
4, coarse segmentation figure is input to warp lamination and carries out interpolation, and exported characteristic pattern with third pond layer 3 and carry out Fusion, obtained segmentation figure carry out warp lamination interpolation again；Then the segmentation figure after interpolation is exported with second pond layer 2 again Characteristic pattern is merged, and obtained segmentation figure carries out warp lamination interpolation again；Finally again by after interpolation segmentation figure with first Pond layer 1 exports characteristic pattern and is merged, and obtained segmentation figure carries out interpolation in warp lamination, obtain final 240 × 240 × 5 segmentation figures；For each pixel, class corresponding to output characteristic pattern maximum value where taking is as final classification results, i.e., The class that each pixel is divided is that it has index (since 0) corresponding to maximum value, such shape in 5 prediction score matrix At final segmentation result figure, the most numerical value of the segmentation result is 0, indicates that background or normal tissue, non-zero are then considered It is tumor region；
5, true value label image is combined using SoftmaxWithLoss layers, 240 × 240 prediction to convolutional layer output Score matrix carries out penalty values calculating, updates network weight with having supervision by backpropagation and stochastic gradient descent method, minimum Change loss function；
6, in the test model stage, execute the process of abovementioned steps 1~4, obtain lesion segmentation as a result, by segmentation result with True value label image is compared, to calculate segmentation performance index；Final segmentation result includes oedema structure, enhancing tumour The result of tumor structure totally 4 classifications of structure, non-reinforcing tumor structure and necrosis；Described image segmentation performance index includes Dice similarity factor (Dice Similarity Coefficient, DSC), positive predictive value (Positive Predictive Value, PPV), the indexs such as sensitivity (Sensitivity)；
7, in the brain tumor Image Automatic Segmentation stage, pretreated brain tumor four-way image to be split is input to tool Have in the full convolutional network model for having optimized network weight, obtains brain tumor segmentation result, in order to distinguish brain tumor internal structure, Downright bad structure is indicated with red, indicates oedema structure with green, indicates non-reinforcing tumor structure with yellow, blue indicates that enhancing is swollen Tumor structure.
Compared with existing conventional segmentation methods, the MRI image brain tumor provided by the invention based on full convolutional network is automatic Dividing method, including the multi-modal MRI image pretreatment of brain tumor, full convolutional network Construction of A Model, network training and arameter optimization With brain tumor Image Automatic Segmentation step, the segmentation of MRI image brain tumor is specially converted into Pixel-level semantic tagger problem, To the different information that MRI different modalities are emphasized, by the two-dimentional whole picture slice synthesis four-way of tetra- mode of FLAIR, T1, T1c and T2 Road input picture, using the convolutional layer group of trained convolutional neural networks and pond layer as the basis of full convolutional network Characteristic layer so that middle layer exports coarse segmentation figure corresponding with semantic segmentation categorical measure, and adds behind middle layer Add deconvolution network, is obtained and the equirotal segmentation figure of original image for carrying out interpolation to coarse segmentation figure.Thus, the present invention Having the advantages that 1, compared with the dividing method based on image block classification, the present invention does not need artificial intervention not only, but also It is not necessary to consider the setting of tile size, it is an automatic, simple brain tumor dividing method, not only increases segmentation Precision also substantially increases the efficiency of segmentation；2, the present invention can more consider classified pixels using entire slice as input picture Appearance and Space Consistency obtain the higher accuracy of separation；3, present invention utilizes transfer learning methods, obtain trained Initial weight of the intermediate features layer weight of good convolutional neural networks as full convolutional network foundation characteristic layer, shortens training Time has saved data mark cost, has still there is preferable learning performance under Small Sample Size.
Finally, it is stated that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although referring to compared with Good embodiment describes the invention in detail, those skilled in the art should understand that, it can be to skill of the invention Art scheme is modified or replaced equivalently, and without departing from the objective and range of technical solution of the present invention, should all be covered at this In the scope of the claims of invention.

Claims (10)

1. a kind of MRI image brain tumor automatic division method based on full convolutional network, which comprises the following steps:
The multi-modal MRI image pretreatment of S1, brain tumor comprising:
S11, field offset correct operation is carried out to two mode MRI images of T1 and T1c；
S12, the MRI image slice for extracting tetra- mode of FLAIR, T1, T1c and T2 will be big in each MRI image slice It is set as 0.99 times of highest gray value in 1% highest gray scale, 0.99 times of minimum gray scale will be set as less than 1% minimum gray scale Value；
S13, data normalization and linear normalization operation are carried out to the gray value of each MRI image slice；
S14, normalized after MRI image same layer tumor biopsy, according to FLAIR, T1, T1c and T2 sequence combine At four-way image, wherein as training set, verifying integrates and the ratio of test set is 10:1:1；
S2, full convolutional network Construction of A Model comprising:
S21, using transfer learning method, obtain the intermediate features layer of trained convolutional neural networks model as full volume The foundation characteristic layer of product network model, the foundation characteristic layer include four convolutional layer groups and pond layer；
S22, after the foundation characteristic layer, addition three be equal with full articulamentum convolutional layer composition middle layers so that in Interbed exports coarse segmentation figure corresponding with semantic segmentation categorical measure；
S23, add warp lamination again after middle layer and carry out interpolation and obtain segmentation result 1, segmentation result 1 again with foundation characteristic The output characteristic pattern of penultimate pond layer is merged in layer, and deconvolution layer interpolation is carried out after fusion and obtains segmentation result 2, And so on, then carry out with third last in foundation characteristic layer and the 4th pond layer merging respectively and warp lamination interpolation, Finally obtain segmentation result 3 and segmentation result 4, the segmentation result 4 is, and and original image corresponding with semantic segmentation categorical measure As equirotal prediction score matrix, the class that each pixel on segmentation result 4 is divided is each pixel in prediction score square Index corresponding to maximum value, forms final segmentation result figure in battle array；
S3, network training and arameter optimization: the segmentation of MRI image brain tumor is converted into Pixel-level semantic tagger problem, is passed through The weight of three convolutional layers added in Gaussian random variable initialization step S22 is made using pretreated four-way image It for training sample and is entered into the full convolutional network model of construction, then using stochastic gradient descent method as optimization method Loss function is minimized with having supervision, to carry out tuning instruction to the weight for all convolutional layers that full convolutional network model includes Practice；
S4, brain tumor Image Automatic Segmentation: by the brain tumor four-way image to be split after normalization be input to have optimized In the full convolutional network model of network weight, prediction score matrix corresponding with semantic segmentation categorical measure is obtained, according to this A little matrix values are partitioned into brain tumor and image of internal structure.
2. the MRI image brain tumor automatic division method according to claim 1 based on full convolutional network, feature exist In in the step S11, using N4ITK method progress biased field correct operation.
3. the MRI image brain tumor automatic division method according to claim 1 based on full convolutional network, feature exist In, in the step S13, to each MRI image slice gray value using following formula carry out data normalization operation:
Wherein, the gray value of the i-th row j column of the corresponding slice X of x (i, j),And XsRespectively be sliced X mean value and variance, x ' (i, It j) is the standardized gray scale of x (i, j)；
The gray value of each slice of linear transfor carries out linear normalization operation in the range of [0,1], and using following formula:
Wherein, x ' (i, j) is the gray scale after standardization, X'maxAnd X'minIt is the maximum value and minimum being sliced after X standardization respectively Value,It is the gray scale after linear normalization.
4. the MRI image brain tumor automatic division method according to claim 1 based on full convolutional network, feature exist In in the step S21, constructing the foundation characteristic layer of full convolutional network model, be to obtain trained 16 layers of VGG model First four convolutional layer group and pond layer, it is described using transfer learning method be using trained weight as full convolution net The initial weight of the foundation characteristic layer of network model.
5. the MRI image brain tumor automatic division method according to claim 1 based on full convolutional network, feature exist In, in the step S21, in the foundation characteristic layer of four convolution groups building, the number of every group of convolutional layer is respectively 2,2,3, 3, the convolution kernel number of every group of convolutional layer is respectively 64,128,256,512, and each convolutional layer uses 3 × 3 small convolution filter Convolution is carried out, and each convolutional layer group is 2 × 2 followed by sampling window, the pond layer that step-length is 2.
6. the MRI image brain tumor automatic division method according to claim 5 based on full convolutional network, feature exist In output characteristic pattern O corresponding to one of them described convolution kernelkIt is calculated using following formula:
Wherein, bkIt is bias term corresponding to k-th of convolution kernel, C is the port number of input data, WkiIt is k-th of convolution kernel I-th channel weight matrix,It is convolution operation, X is input data.
7. the MRI image brain tumor automatic division method according to claim 6 based on full convolutional network, feature exist In the foundation characteristic layer further includes rectification linear unit, and the rectification linear unit is used for output corresponding to convolution kernel Characteristic pattern OkIn each value carry out non-linear transfer, the rectification linear unit is defined as follows:
F (x)=max (0, x)
Wherein, f (x) indicates to rectify linear unit function, and x is an input value.
8. the MRI image brain tumor automatic division method according to claim 1 based on full convolutional network, feature exist In in the step S22, three convolutional layer convolution kernel numbers of addition are respectively 4096,4096,5, and convolution kernel size is all 1 × 1, step-length is also all 1.
9. the MRI image brain tumor automatic division method according to claim 1 based on full convolutional network, feature exist In in the step S23, warp lamination interpolation uses bilinear interpolation method.
10. the MRI image brain tumor automatic division method according to claim 1 based on full convolutional network, feature exist In, in the step S3, optimization method uses cross entropy Classification Loss function, it is defined as follows:
Wherein, N is the size of a batch, and l' is truthful data, and l is the predicted vector of softmax function output.
CN201710379095.6A 2017-05-25 2017-05-25 A kind of MRI image brain tumor automatic division method based on full convolutional network Active CN107220980B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710379095.6A CN107220980B (en) 2017-05-25 2017-05-25 A kind of MRI image brain tumor automatic division method based on full convolutional network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710379095.6A CN107220980B (en) 2017-05-25 2017-05-25 A kind of MRI image brain tumor automatic division method based on full convolutional network

Publications (2)

Publication Number Publication Date
CN107220980A CN107220980A (en) 2017-09-29
CN107220980B true CN107220980B (en) 2019-12-03

Family

ID=59944489

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710379095.6A Active CN107220980B (en) 2017-05-25 2017-05-25 A kind of MRI image brain tumor automatic division method based on full convolutional network

Country Status (1)

Country Link
CN (1) CN107220980B (en)

Families Citing this family (74)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107909577A (en) * 2017-10-18 2018-04-13 天津大学 Fuzzy C-mean algorithm continuous type max-flow min-cut brain tumor image partition method
CN107767378B (en) * 2017-11-13 2020-08-04 浙江中医药大学 GBM multi-mode magnetic resonance image segmentation method based on deep neural network
CN108171232B (en) * 2017-11-15 2021-12-28 中山大学 Deep learning algorithm-based bacterial and viral pneumonia classification method for children
CN108038860A (en) * 2017-11-30 2018-05-15 杭州电子科技大学 Spine segmentation method based on the full convolutional neural networks of 3D
CN109886992A (en) * 2017-12-06 2019-06-14 深圳博脑医疗科技有限公司 For dividing the full convolutional network model training method in abnormal signal area in MRI image
CN108492297B (en) * 2017-12-25 2021-11-19 重庆师范大学 MRI brain tumor positioning and intratumoral segmentation method based on deep cascade convolution network
CN108053401A (en) * 2017-12-27 2018-05-18 广州市申迪计算机系统有限公司 A kind of B ultrasound image processing method and device
CN108062753B (en) * 2017-12-29 2020-04-17 重庆理工大学 Unsupervised domain self-adaptive brain tumor semantic segmentation method based on deep counterstudy
CN108109152A (en) * 2018-01-03 2018-06-01 深圳北航新兴产业技术研究院 Medical Images Classification and dividing method and device
US11288576B2 (en) * 2018-01-05 2022-03-29 Illumina, Inc. Predicting quality of sequencing results using deep neural networks
CN108268870B (en) * 2018-01-29 2020-10-09 重庆师范大学 Multi-scale feature fusion ultrasonic image semantic segmentation method based on counterstudy
CN108427951B (en) * 2018-02-08 2023-08-04 腾讯科技（深圳）有限公司 Image processing method, device, storage medium and computer equipment
WO2019159324A1 (en) * 2018-02-16 2019-08-22 パナソニックＩｐマネジメント株式会社 Processing method and processing device using this
CN108447049A (en) * 2018-02-27 2018-08-24 中国海洋大学 A kind of digitlization physiology organism dividing method fighting network based on production
CN108492286B (en) * 2018-03-13 2020-05-05 成都大学 Medical image segmentation method based on dual-channel U-shaped convolutional neural network
JPWO2019176806A1 (en) * 2018-03-16 2021-04-08 富士フイルム株式会社 Machine learning equipment and methods
CN110297483B (en) * 2018-03-21 2020-12-18 广州极飞科技有限公司 Method and device for obtaining boundary of area to be operated and operation route planning method
CN108564582B (en) * 2018-04-23 2020-08-11 重庆大学 MRI brain tumor image automatic optimization method based on deep neural network
CN108648182B (en) * 2018-04-27 2022-02-11 南京信息工程大学 Breast cancer nuclear magnetic resonance image tumor region segmentation method based on molecular subtype
CN108776805A (en) * 2018-05-03 2018-11-09 北斗导航位置服务（北京）有限公司 It is a kind of establish image classification model, characteristics of image classification method and device
CN108665461B (en) * 2018-05-09 2019-03-12 电子科技大学 A kind of breast ultrasound image partition method corrected based on FCN and iteration sound shadow
CN108898608A (en) * 2018-05-28 2018-11-27 广东技术师范学院 A kind of prostate ultrasonic image division method and equipment
WO2019237342A1 (en) * 2018-06-15 2019-12-19 富士通株式会社 Training method and apparatus for classification neural network for semantic segmentation, and electronic device
CN108961274B (en) * 2018-07-05 2021-03-02 四川大学 Automatic head and neck tumor segmentation method in MRI (magnetic resonance imaging) image
CN109035356B (en) * 2018-07-05 2020-07-10 四川大学 System and method based on PET (positron emission tomography) graphic imaging
CN108986115B (en) * 2018-07-12 2020-12-18 佛山生物图腾科技有限公司 Medical image segmentation method and device and intelligent terminal
CN108986891A (en) * 2018-07-24 2018-12-11 北京市商汤科技开发有限公司 Medical imaging processing method and processing device, electronic equipment and storage medium
CN108932720A (en) * 2018-07-24 2018-12-04 深圳先进技术研究院 A kind of the encephalic patch and carotid plaques intelligent scissor method of magnetic resonance image
CN109118504B (en) * 2018-07-26 2021-03-30 深圳辰视智能科技有限公司 Image edge detection method, device and equipment based on neural network
CN109087318A (en) * 2018-07-26 2018-12-25 东北大学 A kind of MRI brain tumor image partition method based on optimization U-net network model
CN109063710B (en) * 2018-08-09 2022-08-16 成都信息工程大学 3D CNN nasopharyngeal carcinoma segmentation method based on multi-scale feature pyramid
CN109190682B (en) * 2018-08-13 2020-12-18 北京安德医智科技有限公司 Method and equipment for classifying brain abnormalities based on 3D nuclear magnetic resonance image
CN109215041B (en) * 2018-08-17 2022-06-17 上海交通大学医学院附属第九人民医院 Full-automatic pelvic tumor segmentation method and system, storage medium and terminal
CN109242860B (en) * 2018-08-21 2021-10-08 电子科技大学 Brain tumor image segmentation method based on deep learning and weight space integration
CN109345538B (en) * 2018-08-30 2021-08-10 华南理工大学 Retinal vessel segmentation method based on convolutional neural network
CN109360633B (en) * 2018-09-04 2022-08-30 北京市商汤科技开发有限公司 Medical image processing method and device, processing equipment and storage medium
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
CN109325954B (en) * 2018-09-18 2021-08-10 北京旷视科技有限公司 Image segmentation method and device and electronic equipment
CN109377500B (en) * 2018-09-18 2023-07-25 平安科技（深圳）有限公司 Image segmentation method based on neural network and terminal equipment
CN109377480A (en) * 2018-09-27 2019-02-22 中国电子科技集团公司第五十四研究所 Arable land use change detection method based on deep learning
CN109492668B (en) * 2018-10-10 2021-08-03 华中科技大学 MRI (magnetic resonance imaging) different-phase multimode image characterization method based on multi-channel convolutional neural network
CN109685819B (en) * 2018-12-11 2021-02-26 厦门大学 Three-dimensional medical image segmentation method based on feature enhancement
CN109886922B (en) * 2019-01-17 2023-08-18 丽水市中心医院 Automatic grading method for hepatocellular carcinoma based on SE-DenseNet deep learning framework and enhanced MR image
CN111598891B (en) * 2019-02-20 2023-08-08 深圳先进技术研究院 Plaque stability identification method, plaque stability identification device, plaque stability identification equipment and storage medium
CN109948607A (en) * 2019-02-21 2019-06-28 电子科技大学 Candidate frame based on deep learning deconvolution network generates and object detection method
CN109949275A (en) * 2019-02-26 2019-06-28 中山大学肿瘤防治中心 A kind of diagnostic method and device of superior gastrointestinal endoscope image
CN109949318B (en) * 2019-03-07 2023-11-14 西安电子科技大学 Full convolution neural network epileptic focus segmentation method based on multi-modal image
EP3938968A4 (en) * 2019-03-15 2022-11-16 The Trustees of Columbia University in the City of New York System, method and computer-accessible medium for image reconstruction of non-cartesian magnetic resonance imaging information using deep learning
CN109977955B (en) * 2019-04-03 2021-11-30 南昌航空大学 Cervical carcinoma pre-lesion identification method based on deep learning
CN110220902B (en) * 2019-05-09 2022-10-04 杭州迪英加科技有限公司 Digital pathological section analysis method and device
CN110458842B (en) * 2019-06-26 2022-09-20 南京邮电大学 Brain tumor segmentation method based on two-channel three-dimensional dense connection network
CN110444277B (en) * 2019-07-19 2023-03-28 重庆邮电大学 Multi-mode brain MRI image bidirectional conversion method based on multi-generation and multi-confrontation
CN110211127B (en) * 2019-08-01 2019-11-26 成都考拉悠然科技有限公司 Image partition method based on bicoherence network
US10937158B1 (en) 2019-08-13 2021-03-02 Hong Kong Applied Science and Technology Research Institute Company Limited Medical image segmentation based on mixed context CNN model
CN110782427B (en) * 2019-08-19 2023-06-20 大连大学 Magnetic resonance brain tumor automatic segmentation method based on separable cavity convolution
CN110533676B (en) * 2019-09-06 2022-08-16 青岛海信医疗设备股份有限公司 Tumor image segmentation method and device and terminal equipment
CN110751651B (en) * 2019-09-27 2022-03-04 西安电子科技大学 MRI pancreas image segmentation method based on multi-scale migration learning
CN110853048A (en) * 2019-10-14 2020-02-28 北京缙铖医疗科技有限公司 MRI image segmentation method, device and storage medium based on rough training and fine training
CN110827291A (en) * 2019-10-18 2020-02-21 北京量健智能科技有限公司 Method and device for automatic brain MRI quantitative analysis
CN110910351B (en) * 2019-10-31 2022-07-29 上海交通大学 Ultrasound image modality migration and classification method and terminal based on generation countermeasure network
CN110992320B (en) * 2019-11-22 2023-03-21 电子科技大学 Medical image segmentation network based on double interleaving
CN111145147B (en) * 2019-12-14 2023-08-22 中国科学院深圳先进技术研究院 Multi-mode medical image segmentation method and terminal equipment
CN111080658A (en) * 2019-12-16 2020-04-28 中南民族大学 Cervical MRI image segmentation method based on deformable registration and DCNN
CN113408325A (en) * 2020-03-17 2021-09-17 北京百度网讯科技有限公司 Method and device for identifying surrounding environment of vehicle and related equipment
KR102534762B1 (en) * 2020-03-30 2023-05-19 주식회사 뷰웍스 Method and appararus for generating synthetic 2d image
CN111754520B (en) * 2020-06-09 2023-09-15 江苏师范大学 Deep learning-based cerebral hematoma segmentation method and system
CN112001301B (en) * 2020-08-21 2021-07-20 江苏三意楼宇科技股份有限公司 Building monitoring method and device based on global cross entropy weighting and electronic equipment
CN112215795B (en) * 2020-09-02 2024-04-09 苏州超集信息科技有限公司 Intelligent detection method for server component based on deep learning
CN112200811A (en) * 2020-10-27 2021-01-08 西北工业大学深圳研究院 Method for generating improved glioma segmentation by adopting cross-sequence nuclear magnetic resonance image
CN112561868B (en) * 2020-12-09 2021-12-07 深圳大学 Cerebrovascular segmentation method based on multi-view cascade deep learning network
CN113139581B (en) * 2021-03-23 2023-09-01 广东省科学院智能制造研究所 Image classification method and system based on multi-image fusion
CN113012168B (en) * 2021-03-24 2022-11-04 哈尔滨理工大学 Brain glioma MRI image segmentation method based on convolutional neural network
CN113450256B (en) * 2021-06-30 2023-03-03 重庆理工大学 Automatic aorta segmentation method and system based on CT image of level set
CN115601550B (en) * 2022-12-13 2023-04-07 深圳思谋信息科技有限公司 Model determination method, model determination device, computer equipment and computer readable storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104834943A (en) * 2015-05-25 2015-08-12 电子科技大学 Brain tumor classification method based on deep learning
CN104851101A (en) * 2015-05-25 2015-08-19 电子科技大学 Brain tumor automatic segmentation method based on deep learning
CN106296699A (en) * 2016-08-16 2017-01-04 电子科技大学 Cerebral tumor dividing method based on deep neural network and multi-modal MRI image
CN106600571A (en) * 2016-11-07 2017-04-26 中国科学院自动化研究所 Brain tumor automatic segmentation method through fusion of full convolutional neural network and conditional random field

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8908948B2 (en) * 2011-12-21 2014-12-09 Institute Of Automation, Chinese Academy Of Sciences Method for brain tumor segmentation in multi-parametric image based on statistical information and multi-scale structure information

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104834943A (en) * 2015-05-25 2015-08-12 电子科技大学 Brain tumor classification method based on deep learning
CN104851101A (en) * 2015-05-25 2015-08-19 电子科技大学 Brain tumor automatic segmentation method based on deep learning
CN106296699A (en) * 2016-08-16 2017-01-04 电子科技大学 Cerebral tumor dividing method based on deep neural network and multi-modal MRI image
CN106600571A (en) * 2016-11-07 2017-04-26 中国科学院自动化研究所 Brain tumor automatic segmentation method through fusion of full convolutional neural network and conditional random field

Also Published As

Publication number Publication date
CN107220980A (en) 2017-09-29

Similar Documents

Publication Publication Date Title
CN107220980B (en) A kind of MRI image brain tumor automatic division method based on full convolutional network
CN110889852B (en) Liver segmentation method based on residual error-attention deep neural network
CN108765408A (en) Build the method in cancer pathology image virtual case library and the multiple dimensioned cancer detection system based on convolutional neural networks
CN112070772A (en) Blood leukocyte image segmentation method based on UNet + + and ResNet
CN109886986A (en) A kind of skin lens image dividing method based on multiple-limb convolutional neural networks
CN108596248A (en) A kind of classification of remote-sensing images model based on improvement depth convolutional neural networks
CN110675411B (en) Cervical squamous intraepithelial lesion recognition algorithm based on deep learning
CN113256641B (en) Skin lesion image segmentation method based on deep learning
CN106296699A (en) Cerebral tumor dividing method based on deep neural network and multi-modal MRI image
CN105469100A (en) Deep learning-based skin biopsy image pathological characteristic recognition method
CN110889853A (en) Tumor segmentation method based on residual error-attention deep neural network
CN106296695A (en) Adaptive threshold natural target image based on significance segmentation extraction algorithm
CN111400536B (en) Low-cost tomato leaf disease identification method based on lightweight deep neural network
CN108416353A (en) Crop field spike of rice fast partition method based on the full convolutional neural networks of depth
DE112015000964T5 (en) Image processing apparatus, image processing method and image processing program
CN102496023A (en) Region of interest extraction method of pixel level
CN106611423B (en) SAR image segmentation method based on ridge ripple filter and deconvolution structural model
CN108053398A (en) A kind of melanoma automatic testing method of semi-supervised feature learning
CN109671094A (en) A kind of eye fundus image blood vessel segmentation method based on frequency domain classification
CN106845551A (en) A kind of histopathology image-recognizing method
CN104361574B (en) No-reference color image quality assessment method on basis of sparse representation
CN114283158A (en) Retinal blood vessel image segmentation method and device and computer equipment
CN109766805A (en) A kind of double-deck license plate character recognition method based on deep learning
CN111161287A (en) Retinal vessel segmentation method based on symmetric bidirectional cascade network deep learning
CN110751636A (en) Fundus image retinal arteriosclerosis detection method based on improved coding and decoding 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
TA01 Transfer of patent application right

Effective date of registration: 20191024

Address after: 400000 No. 12 Chen Cheng Road, Shapingba District, Chongqing

Applicant after: Chongqing Normal University

Address before: No. 69 lijiatuo Chongqing District of Banan City Road 400054 red

Applicant before: chongqing university of technology

TA01 Transfer of patent application right
GR01 Patent grant
GR01 Patent grant