CN111798464A - Lymphoma pathological image intelligent identification method based on deep learning - Google Patents
Lymphoma pathological image intelligent identification method based on deep learning Download PDFInfo
- Publication number
- CN111798464A CN111798464A CN202010622536.2A CN202010622536A CN111798464A CN 111798464 A CN111798464 A CN 111798464A CN 202010622536 A CN202010622536 A CN 202010622536A CN 111798464 A CN111798464 A CN 111798464A
- Authority
- CN
- China
- Prior art keywords
- layer
- lymphoma
- image
- neural network
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000001575 pathological effect Effects 0.000 title claims abstract description 82
- 206010025323 Lymphomas Diseases 0.000 title claims abstract description 81
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 238000013528 artificial neural network Methods 0.000 claims abstract description 56
- 238000012549 training Methods 0.000 claims abstract description 40
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 238000012360 testing method Methods 0.000 claims abstract description 9
- 210000003563 lymphoid tissue Anatomy 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000013145 classification model Methods 0.000 claims abstract description 4
- 238000011176 pooling Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 19
- 238000005070 sampling Methods 0.000 claims description 19
- 230000004913 activation Effects 0.000 claims description 15
- 210000002569 neuron Anatomy 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 13
- 230000007170 pathology Effects 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 9
- 238000004043 dyeing Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 7
- 238000009826 distribution Methods 0.000 claims description 6
- 238000000265 homogenisation Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 210000004204 blood vessel Anatomy 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 2
- 230000008859 change Effects 0.000 claims description 2
- 238000009795 derivation Methods 0.000 claims description 2
- 239000000284 extract Substances 0.000 claims description 2
- 230000004927 fusion Effects 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 238000010200 validation analysis Methods 0.000 claims description 2
- 230000000903 blocking effect Effects 0.000 claims 1
- 238000003745 diagnosis Methods 0.000 abstract description 7
- 210000002751 lymph Anatomy 0.000 description 6
- 208000015914 Non-Hodgkin lymphomas Diseases 0.000 description 4
- 208000031671 Large B-Cell Diffuse Lymphoma Diseases 0.000 description 3
- 206010012818 diffuse large B-cell lymphoma Diseases 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 206010028980 Neoplasm Diseases 0.000 description 2
- 206010042971 T-cell lymphoma Diseases 0.000 description 2
- 208000027585 T-cell non-Hodgkin lymphoma Diseases 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000010827 pathological analysis Methods 0.000 description 2
- 238000004393 prognosis Methods 0.000 description 2
- 238000010186 staining Methods 0.000 description 2
- 230000004083 survival effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004195 computer-aided diagnosis Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 206010020718 hyperplasia Diseases 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000003703 image analysis method Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000010837 poor prognosis Methods 0.000 description 1
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
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—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
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Public Health (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Pathology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A lymphoma pathological image intelligent identification method based on deep learning comprises the following steps: preprocessing lymphoma pathological section image data; constructing a full convolution neural network for segmenting the lymphatic tissue region, wherein the full convolution neural network comprises an encoder sub-network and a decoder sub-network; constructing a lymphoma three-classification convolutional neural network under high-power resolution, which comprises 6 convolutional layers and 3 full-connection layers which are sequentially connected; training the full convolution neural network and the lymphoma three-classification convolutional neural network to finally obtain a lymphoma pathological section image classification model, and sequentially passing through the full convolution neural network and the lymphoma three-classification convolutional neural network during testing to finally obtain a lymphoma classification result. The method provides reliable intermediate data for a pathologist to judge the lymphoma subtype category, and provides auxiliary diagnosis reference for the pathologist to classify the lymphoma subtype by analyzing the digitally scanned lymphoma pathological image, so that the pathologist is helped to realize the rapid judgment of the lymphoma suffering condition of a patient.
Description
Technical Field
The invention relates to a lymphoma auxiliary diagnosis system. In particular to a lymphoma pathological image intelligent identification method based on deep learning.
Background
Lymphoma is one of common malignant tumors in China, and is a tumor which is difficult to diagnose clinically and pathologically at present because the lymphoma has complex and various pathological types and no specific pathological histological manifestation, and clinical pathological diagnosis is easily confused with other tumors and is easy to cause misdiagnosis. The pathological misdiagnosis rate of the lymphoma is 10 to 33.33 percent, and misdiagnosed patients cannot be treated in time, often miss the best treatment opportunity, and seriously affect the treatment and prognosis of the patients. Diffuse large B-cell lymphoma (DLBCL) is the most common non-hodgkin lymphoma (NHL), accounts for about 30% -40% of all NHLs, is a group of invasive lymphomas with significant heterogeneity, has rapid clinical progression and poor prognosis, and has an overall survival rate of only 46% in 5 years for DLBCL patients. T cell lymphoma accounts for 10-15% of non-Hodgkin lymphoma, and the total 5-year survival rate of subtype patients is only 10-30%.
In recent years, deep learning has been successful in images, and the proportion of medical image data in medical data is increasing, so that a basis is provided for improving the performance of pathological recognition and diagnosis. The multi-layer framework of the neural network in deep learning enables the neural network to extract high-level abstract features hidden in original data layer by layer, and therefore the neural network can be trained in the direct face of the original data. When the data volume is continuously increased, the performance of the neural network can be continuously improved, and the continuously increased data volume of the current medical industry provides favorable conditions for improving the performance of the neural network model.
The correct pathological diagnosis of lymphoma is not clear from the characteristic factors of imaging. Therefore, the computer-aided analysis is carried out by utilizing the digital medical pathological images, the efficiency of pathological image analysis by a pathological doctor can be improved, the misdiagnosis rate of lymphoma pathology is reduced, and great convenience is brought to the treatment and prognosis of patients. However, the complexity of the pathohistological manifestations of lymphomas and the high resolution nature of their pathological images make computer-aided diagnosis difficult.
Disclosure of Invention
The invention aims to solve the technical problem of providing a lymphoma pathological image intelligent identification method based on deep learning, which combines a pathological doctor film evaluation method to realize the extraction of features with different scales from coarse to fine and provides auxiliary diagnosis reference for the pathological doctor to classify the subtype of the lymphoma.
The technical scheme adopted by the invention is as follows: a lymphoma pathological image intelligent identification method based on deep learning comprises the following steps:
1) preprocessing lymphoma pathological section image data, comprising:
(1) obtaining an original digital scan lymphoma pathology section image dataset, each pathology section image X in the dataset being processed by a medical professionaliCarrying out manual marking, wherein the manual marking comprises marking of a focus, a blood vessel and a fat area, and generating each pathological section image X according to the result of the manual markingiThe corresponding mask image Yi;
(2) performing dyeing homogenization treatment on the pathological section image data set to adapt to the problem of different color shades in the sections caused by different dyeing conditions;
(3) for each pathological section image X in the data setiCutting and dicing are carried out, and the resolution of each image block is 512 multiplied by 512;
2) constructing a full convolution neural network for segmenting a lymphatic tissue region, wherein the full convolution neural network comprises an encoder sub-network and a decoder sub-network, wherein the encoder sub-network consists of a first convolution layer, a second convolution layer, a third convolution layer, a first down-sampling layer, a fourth convolution layer, a fifth convolution layer, a sixth convolution layer and a second down-sampling layer which are sequentially connected; the decoder sub-network consists of a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a first up-sampling layer, a tenth convolutional layer, an eleventh convolutional layer, a twelfth convolutional layer and a second up-sampling layer which are connected in sequence; the output of the second downsampling layer is the output of the seventh convolutional layerAnd the output of the second upper sampling layer forms the output of a full convolution neural network, the output of the full convolution neural network is a probability graph with the same resolution as the image block input by the full convolution neural network, and each pixel point in the probability graph corresponds to the pathological section image XiThe probability of each pixel point belonging to the lymphatic tissue area is small;
3) constructing a lymphoma three-classification convolutional neural network under high-power resolution, which comprises 6 convolutional layers and 3 full-connection layers which are sequentially connected;
4) training the full convolution neural network in the step 2) and the lymphoma three-classification convolution neural network in the step 3) to finally obtain a lymphoma pathological section image classification model, and sequentially passing through the full convolution neural network and the lymphoma three-classification convolution neural network during testing to finally obtain a classification result of the lymphoma.
The intelligent lymphoma pathological image identification method based on deep learning is combined with pathological imaging and judgment habits of professional pathologists to construct a full convolution neural network model, partition out a lymphatic tissue area, reduce noise influence and construct a classification convolution neural network, and realize high-accuracy classification of pathological images of different lymphoma subtype. The method can provide reliable intermediate data for a pathologist to judge the lymphoma subtype category, and provides auxiliary diagnosis reference for the pathologist to classify the lymphoma subtype by analyzing the digitally scanned lymphoma pathological image, so that the pathologist is helped to realize the rapid judgment of the lymphoma suffering condition of a patient, and the misdiagnosis rate of the pathologist is reduced.
Drawings
FIG. 1 is a flow chart of training in the method of the present invention;
FIG. 2 is a flow chart of a test in the method of the present invention.
Detailed Description
The method for intelligently identifying the lymphoma pathological image based on deep learning of the invention is described in detail below with reference to the embodiments and the accompanying drawings.
The invention discloses a lymphoma pathological image intelligent identification method based on deep learning, relates to the field of deep learning and pathological images, solves the problems of lymphoma subtype identification accuracy rate and misdiagnosis rate reduction under high-resolution pathological images, and adopts the technical scheme that through providing a deep learning model and combining a pathological doctor evaluation method, coarse-to-fine different-scale feature extraction is realized, and auxiliary reference is provided for a pathological doctor to classify the lymphoma subtypes.
The invention discloses a lymphoma pathological image intelligent identification method based on deep learning, which comprises the following steps:
1) preprocessing lymphoma pathological section image data, comprising:
(1) obtaining an original digital scan lymphoma pathology section image dataset, each pathology section image X in the dataset being processed by a medical professionaliCarrying out manual marking, wherein the manual marking comprises marking of a focus, a blood vessel and a fat area, and generating each pathological section image X according to the result of the manual markingiThe corresponding mask image Yi;
(2) and (3) carrying out dyeing homogenization treatment on the pathological section image data set so as to adapt to the problem of different color shades in the sections caused by different dyeing conditions.
Such as: selecting a target image H from the data set by a medical professionalsourceRemoving H from the data setsourceOther pathological section images are selected as HtargetWith HsourceThe color base of (A) is a standard replacement HtargetColor base while retaining HtargetRelative staining density, staining uniformity matrix VsourceExpressed as:
wherein P is99Represents 99 quantiles, WtargetRepresenting the separated color basis matrix for relative optical density.
The source light density for each channel and each pixel is then dye normalized to the target image using an inverse beer-lambert transform and converted back to pixel intensity space, the formula being:
c represents three channels of the image, x represents the pixel position, target represents the selected target image, ic,x,sourceRepresenting each pixel point value, i, after conversion0Representing the maximum intensity value, e is a natural constant.
(3) For each pathological section image X in the data setiCutting and dicing are carried out, and the resolution of each image block is 512 multiplied by 512;
cutting each pathological section image in the data set into blocks under 4 x image layers, wherein the resolution of each image block is 512 x 512, and obtaining a training set LT under 4 x image layers1Corresponding to a mask tag set LTy1. Simultaneously, each pathological section image in the data set is cut into blocks in a non-overlapping mode under a 40 x image layer, the resolution of each image block is 512 x 512, and a training set HT under the 40 x image layer is obtained1And corresponding class label HTy1。
In training set and validation set under 4 × image layer and 40 × image layer, each image block
Has a resolution of 512X 512, wherein i represents the ith pathological section image in the data set, and z represents 4X image layer or 40X image layer, so as to complete the pathological section image XiThe lower left corner point is a coordinate zero point, x and Y respectively represent coordinate points of the upper left corner point of the image block on the abscissa axis and the ordinate axis, and simultaneously correspond to the mask image YiThe cutting and dicing are performed under the same coordinates.
2) Constructing a full convolution neural network for segmenting a lymphatic tissue region, wherein the full convolution neural network comprises an encoder sub-network and a decoder sub-network, wherein the encoder sub-network consists of a first convolution layer, a second convolution layer, a third convolution layer, a first down-sampling layer, a fourth convolution layer, a fifth convolution layer, a sixth convolution layer and a second down-sampling layer which are sequentially connected; the decoder sub-network comprises a seventh convolutional layer, an eighth convolutional layer and a ninth convolutional layer which are connected in sequenceA first up-sampling layer, a tenth convolution layer, an eleventh convolution layer, a twelfth convolution layer and a second up-sampling layer; the output of the second down-sampling layer is the input of a seventh convolution layer, the output of the second up-sampling layer forms the output of a full convolution neural network, the output of the full convolution neural network is a probability map with the same resolution as the image block input by the full convolution neural network, and each pixel point in the probability map corresponds to a pathological section image XiThe probability of each pixel point belonging to the lymphatic tissue area is small; wherein,
in the encoder subnetwork, the convolution kernel size of each convolution layer is 3 x 3, the step length is 1, a batch normalization layer and ReLU nonlinear activation are arranged behind each convolution layer, the downsampling layer is selected from Max scaling, the window size is 2 x 2, and the step length is 2;
in the decoder subnetwork, the convolution kernel size of each convolution layer is 3 multiplied by 3, the step length is 1, a batch normalization layer and ReLU nonlinear activation are arranged behind each convolution layer, bilinear interpolation is selected for an upper sampling layer, the window size is 2 multiplied by 2, and the step length is 2;
each convolutional layer extracts a feature map according to the following formula:
wherein,
to round down, xinIndicating input feature size, xoutRepresenting the size of an output characteristic diagram, padding representing the size of pixel points of a filling characteristic diagram, s representing a step length, and f representing the size of a convolution kernel;
in the present invention, for example, the input of the first convolutional layer is 512 × 512, the convolutional kernel size is 3 × 3, the padding size is 1, and the step size is 1, so the output signature size of the first convolutional layer is:
the ReLU activation function introduces a nonlinear factor to the neuron, and the formula of the ReLU activation function is as follows:
f(x)＝max(0,x)
batch normalization is to forcibly pull back the distribution of the input value of any neuron in each layer of the full convolution neural network to a standard normal distribution with the mean value of 0 and the variance of 1 by a certain normalization means, wherein the normalization function is in the form as follows:
wherein a isiFor a certain value of the original activation of a neuron,
is a normalized value after normalization operation. The whole normalization process can be decomposed into two steps, wherein the first step is to regulate the activation value to be within a normal distribution range with the average value of 0 and the variance of 1. Where μ is an average value obtained from activation values of m neurons included in the neuron set S, that is:
the standard deviation of the activation values is calculated according to the mean value and the respective activation values of the neurons in the set S:
among these are small constant data added to increase training stability.
The second step is mainly aimed at enabling each neuron to learn two corresponding adjusting factors in the training process, and carrying out fine adjustment on the value normalized to the mean value 0 and the variance 1. Because normalization, after the first step of operation, may reduce the non-linear expressive power of the neural network, the expressive power of the neural network after the normalization operation is compensated in this way.
3) Constructing a lymphoma three-classification convolutional neural network under high-power resolution, which comprises 6 convolutional layers and 3 full-connection layers which are sequentially connected;
in the lymphoma three-classification convolutional neural network, the first 3 convolutional layers of 6 convolutional layers all comprise 64 3 × 3 filters, the 4 th convolutional layer and the 5 th convolutional layer all comprise 128 3 × 3 filters, the last convolutional layer all comprises 256 3 × 3 filters, the 1 st, 2 nd, 4 th and 6 th convolutional layers sequentially execute 2 × 2 maximum pooling operation, the step length is 2, except the last convolutional layer, a batch normalization layer and ReLU nonlinear activation are arranged behind the rest convolutional layers, the three fully-connected layers correspondingly comprise 512 neurons, 512 neurons and 3 neurons in front-back order, and droout operation is carried out with the probability of 0.5; wherein
(1) Pooling layer for maximum pooling operation
After the maximum pooling operation is used for convolution operation, the function is feature fusion and dimensionality reduction, and the formula for extracting the feature map is as follows:
wherein,
to round down, xinIndicating input feature size, xoutRepresenting the size of the output characteristic diagram, padding representing the size of pixel points of the filling characteristic diagram, s representing the step length, and n representing the size of the pooling window;
in the pooling operation, we can see that it is a parameter without participating in the forward calculation, so the calculation of the gradient of the pooling operation in the network only needs to be implemented by using the following formula:
wherein, loss is expressed as a loss function value, Powing _ in is expressed as a Pooling operation input, Powing _ out is expressed as a Pooling operation output, and Powing.backward is expressed as a backward propagation of a Pooling layer;
compared with a full convolution neural network, the lymphoma three-classification convolution neural network is added with a full connection layer at last, namely, a characteristic diagram (matrix) obtained by convolution of the last layer is unfolded into a one-dimensional vector, and an input is provided for a classifier.
(2) dropout operation
The dropout operation reserves part of neurons with probability p in the forward propagation of each training batch, the complexity of a neural network can be simplified, and the overfitting risk is reduced, wherein the change of a previous full-connection layer after the dropout operation is as follows:
Zl+1＝W·Al+b
Al+1＝R(Zl+1)
rl～Bernoulli(q)
Al+1＝R(Zl+1)
wherein l denotes the l-th layer of the network, AlRepresenting the input of the fully-connected layer, Bernoulli represents the Bernoulli function,
representing a part of the neuron remaining with a probability q, Zl+1Represents the output of the fully connected layer, W represents the weight of the l-th layer, b represents the bias of the l-th layer, and R represents the nonlinear activation function.
The full convolution neural network in the step 2) and the lymphoma three-classification convolution neural network in the step 3) are both optimized by using an Adam optimizer, and the parameter optimization updating rule is as follows:
(1) calculating the exponential moving average of the t-time gradient
First, the gradient is calculated
Wherein, gtRepresenting the parameter theta at t time stepstThe gradient of the gradient to be obtained is determined,
representing a gradient operator, J (θ)t-1) Representing a differentiable random scalar function;
calculating the exponential moving average m of the t time gradient according to the gradienttThe formula is as follows:
mt＝β1mt-1+(1-β1)gt
the first order moment vector m is initialized to 0, coefficient beta1An exponential decay rate, usually taking a value close to 1, defaulted to 0.9, for controlling the weight distribution (momentum and current gradient);
(2) calculating the exponential moving average v of the square of the gradienttThe formula is as follows:
initializing the second moment vector v to 0 by the coefficient beta2The default is 0.999 for controlling the influence of the square of the previous gradient;
(3) since the first order moment vector m is initialized to 0, it results in an exponential moving average m of the t-time gradienttBiased towards 0, especially in the early stages of training. Exponential moving average m to be applied to t time gradienttAnd (3) correcting deviation to reduce the influence of the deviation on the initial training stage:
wherein,
the gradient mean value after deviation correction is carried out;
(4) since the initialization of the second moment vector v to 0 results in the training of the exponentially moving average v of the gradient squared in the initial phasetBiased toward 0, the exponentially moving average of the squared gradient is corrected as follows:
wherein,
estimating a second-order original moment after deviation correction;
(5) updating a parameter θ of a networktThe formula is as follows:
where α is a learning rate, the default learning rate α has an initial value of 0.001, and θt-1For the parameters of the network before updating, the constant is 10-8Avoid the divisor changing to 0;
it can be seen from the expression that the updated step size calculation can be adaptively adjusted from two angles of gradient mean and gradient square, rather than being directly determined by the current gradient.
When the network is trained, the learning rate α gradually attenuates with the increase of the training times, and the attenuation formula is:
where decay is expressed as decay rate, epoch _ num is expressed as number of training sessions, α0Expressed as the initial learning rate.
The loss function used by the full convolution neural network and the lymphoma three-classification convolution neural network is cross entropy loss, is used for measuring the difference between two probability distributions, and is divided into two-classification and multi-classification conditions: wherein,
(1) case of two classes
In the case of bisection, the final result to be predicted by the network has only two cases, and the prediction probabilities obtained by prediction for each category are p and 1-p, where the expression is:
wherein Loss is the Loss value of the cross quotient, yiA label representing the ith pathological section image in the data set, wherein the positive class is 1, and the negative class is 0; p is a radical ofiRepresenting the probability that the ith pathological section image in the data set is predicted to be positive; n represents the number of pathological section images in the data set;
(2) case of multiple classifications
The multi-classification case is an extension to two classes:
where M denotes the number of categories, yicOne-hot unique coding, p, representing the ith pathological section image category cicRepresenting the probability of predicting that the ith pathological section image belongs to the class c;
at the same time, since cross entropy involves calculating the probability of each class, the last layer of the network uses the softmax function, and the probability S taken for each class is in the functional form:
wherein e is a natural constant, j and k both represent category indexes, the total number of categories is C, and V is the input of the softmax layer;
since the softmax function maps values between 0-1, and the sum is 1, then there are:
and (3) carrying out derivation on the cross entropy Loss value Loss to obtain:
only need to find hjSubtracting the result by 1 is the gradient of the inverse update.
4) Training the said complete convolution neural network and the said lymphoma three-classification convolution neural network, 4 times training set LT under the image layer1And tag LTy1As training input for the full convolutional neural network, a training set HT under 40 × layer1And tag HTy1And (3) as training input of the lymphoma three-classification convolutional neural network, and finally obtaining a lymphoma pathological section image classification model through a training process shown in figure 1. During testing, the test flow is shown in figure 2 after passing through the full convolution neural network and the lymphoma three-classification convolution neural network in sequence, and finally the classification result of the lymphoma is obtained.
In the application test, the output of the full convolution neural network is used as the input of the lymphoma three-classification convolution neural network after threshold value binarization processing, and the binarization threshold value set in the binarization processing is 0.5.
The invention discloses an application of a lymphoma pathological image auxiliary diagnosis method based on deep learning, which is used for obtaining reference data for classifying lymphoma subtypes of patients by professional pathologists through an image analysis method based on deep learning on the basis of acquiring lymphoma digital scanning slice images.
Specific examples are given below:
three types of lymphoma pathology sections were collected from 184 individuals and digitally scanned, where a: pathological section images of reactive hyperplasia of 67 patients, class B: pathological section images of diffuse large B lymphoma of 54 patients, class C: pathological section images of T cell lymphoma of 63 patients. And preprocessing the pathological section image obtained by digital scanning, namely labeling, dyeing homogenization and cutting the image into blocks by a professional pathologist. The following example is used to verify the validity of the method: the lymph tissue area is divided under low resolution, and A, B, C types of the divided lymph tissue area are classified under high resolution, so that the purpose of auxiliary diagnosis of lymphoma diseases is achieved. Thus, it can be seen that:
(a) the method comprises the steps of segmenting lymph tissue regions under low resolution to achieve the purpose of removing image noise, taking 126 pathological section images as a training set, taking 20 pathological section images as a test set, enabling all case sections to be provided with pixel level labels by professional pathologists, constructing a full convolution neural network through dyeing homogenization and cutting blocks, pre-training on a camelyon16 data set, then performing fine tuning training on the constructed low resolution training set, obtaining a trained full convolution network model after 50 rounds, and enabling an IOU value on a verification set to be 0.88.
(b) Classifying lymph tissue regions under high resolution, cutting the training set and the verification set into blocks under high resolution, constructing a convolutional neural network, pre-training the convolutional neural network on a mitos-2012 data set, then performing fine tuning training on the constructed low resolution training set, and obtaining a trained convolutional network model after 50 rounds, wherein the accuracy rate on the verification set is 97%.
(c) In the testing stage, after preprocessing 38 cases of pathological section images, firstly, a probability image of a lymph tissue area is obtained by using a trained full convolution neural network, and after threshold value binarization, the lymph tissue area is cut and cut into blocks under high resolution to be used as input of the trained convolution network model in the step (b), and finally, a classification result of lymphoma subtype three classification of the whole pathological section is obtained.
The above description of the present invention and the embodiments is not limited thereto, and the description of the embodiments is only one of the implementation manners of the present invention, and any structure or embodiment similar to the technical solution without inventive design is within the protection scope of the present invention without departing from the inventive spirit of the present invention.
Claims (8)
1. A lymphoma pathological image intelligent identification method based on deep learning is characterized by comprising the following steps:
1) preprocessing lymphoma pathological section image data, comprising:
(1) obtaining an original digital scan lymphoma pathology section image dataset, each pathology section image X in the dataset being processed by a medical professionaliCarrying out manual marking, wherein the manual marking comprises marking of a focus, a blood vessel and a fat area, and generating each pathological section image X according to the result of the manual markingiThe corresponding mask image Yi;
(2) performing dyeing homogenization treatment on the pathological section image data set to adapt to the problem of different color shades in the sections caused by different dyeing conditions;
(3) for each pathological section image X in the data setiCutting and dicing are carried out, and the resolution of each image block is 512 multiplied by 512;
2) constructing a full convolution neural network for segmenting a lymphatic tissue region, wherein the full convolution neural network comprises an encoder sub-network and a decoder sub-network, wherein the encoder sub-network consists of a first convolution layer, a second convolution layer, a third convolution layer, a first down-sampling layer, a fourth convolution layer, a fifth convolution layer, a sixth convolution layer and a second down-sampling layer which are sequentially connected; the decoder sub-network consists of a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a first up-sampling layer, a tenth convolutional layer, an eleventh convolutional layer, a twelfth convolutional layer and a second up-sampling layer which are connected in sequence; the output of the second down-sampling layer is the input of a seventh convolution layer, the output of the second up-sampling layer forms the output of a full convolution neural network, the output of the full convolution neural network is a probability map with the same resolution as the image block input by the full convolution neural network, and each pixel point in the probability map corresponds to a pathological section image XiThe probability of each pixel point belonging to the lymphatic tissue area is small;
3) constructing a lymphoma three-classification convolutional neural network under high-power resolution, which comprises 6 convolutional layers and 3 full-connection layers which are sequentially connected;
4) training the full convolution neural network in the step 2) and the lymphoma three-classification convolution neural network in the step 3) to finally obtain a lymphoma pathological section image classification model, and sequentially passing through the full convolution neural network and the lymphoma three-classification convolution neural network during testing to finally obtain a classification result of the lymphoma.
2. The method for intelligently identifying pathological images of lymphoma according to claim 1, wherein the step (3) in step 1) is that each pathological section image X in data set isiCutting and slicing each pathological section image in the data set under a 4 x image layer, wherein the resolution of each image block is 512 x 512, all the image blocks are randomly distributed to a training set and a verification set under the 4 x image layer in a ratio of 8:2, and a 4 x image layer lower training set LT is obtained1Corresponding to a mask tag set LTy1Verification set LV under 4 x layer1Corresponding to the mask tag set LVy1(ii) a Simultaneously, each pathological section image in the data set is subjected to non-overlapping cutting and blocking under a 40 x image layer, the resolution of each image block is 512 x 512, all the image blocks are randomly distributed to a training set and a verification set under the 40 x image layer, the ratio is 8:2, and a training set HT under the 40 x image layer is obtained1And class label HTy1Verification set HV1And classification label HVy1；
In training set and validation set under 4 × image layer and 40 × image layer, each image block
Is 512X 512, wherein i represents the ith pathological section image in the data set, z represents 4X image layer or 40X image layer, and the pathological section image X is integratediThe lower left corner point is a coordinate zero point, x and Y respectively represent coordinate points of the upper left corner point of the image block on the abscissa axis and the ordinate axis, and simultaneously correspond to the mask image YiThe cutting and dicing are performed under the same coordinates.
3. The method for intelligently identifying pathological images of lymphoma according to claim 1, wherein in the encoder subnetwork in step 2), the convolution kernel size of each convolution layer is 3 × 3, the step size is 1, and each convolution layer is followed by a batch normalization layer and ReLU nonlinear activation, and the downsampling layer is selected from MaxPooling, the window size is 2 × 2, and the step size is 2; in the decoder subnetwork, the convolution kernel size of each convolution layer is 3 multiplied by 3, the step length is 1, a batch normalization layer and ReLU nonlinear activation are arranged behind each convolution layer, bilinear interpolation is selected for an upper sampling layer, the window size is 2 multiplied by 2, and the step length is 2;
each convolutional layer extracts a feature map according to the following formula:
wherein,
to round down, xinIndicating input feature size, xoutAnd the size of the output characteristic diagram is represented, padding represents the size of pixel points of the filling characteristic diagram, s represents the step length, and f represents the size of a convolution kernel.
4. The method according to claim 1, wherein in the lymphoma tri-classification convolutional neural network in step 3), the first 3 convolutional layers of the 6 convolutional layers each include 64 3 × 3 filters, the 4 th and 5 th convolutional layers each include 128 3 × 3 filters, the last convolutional layer each includes 256 3 × 3 filters, the 1 st, 2 th, 4 th and 6 th convolutional layers sequentially perform 2 × 2 maximal pooling, the step size is 2, except the last convolutional layer, the rest convolutional layers are followed by a batch normalization layer and ReLU nonlinear activation, and the three fully-connected layers respectively include 512, 512 and 3 neurons in a front-to-back order and perform a dropout operation with a probability of 0.5; wherein
(1) Pooling layer for maximum pooling operation
After the maximum pooling operation is used for convolution operation, the function is feature fusion and dimensionality reduction, and the formula for extracting the feature map is as follows:
wherein,
to round down, xinIndicating input feature size, xoutRepresenting the size of the output characteristic diagram, padding representing the size of pixel points of the filling characteristic diagram, s representing the step length, and n representing the size of the pooling window;
in the pooling operation, the calculation of the pooling operation gradient in the network is realized by adopting the following formula:
wherein, loss is expressed as a loss function value, Powing _ in is expressed as a Pooling operation input, Powing _ out is expressed as a Pooling operation output, and Powing.backward is expressed as a backward propagation of a Pooling layer;
(2) dropout operation
dropout operation, in the forward propagation of each training batch, a part of neurons are reserved with a probability q, and the change of a previous fully-connected layer after the dropout operation is as follows:
Zl+1＝W·Al+b
Al+1＝R(Zl+1)
rl～Bernoulli(q)
Al+1＝R(Zl+1)
wherein l denotes the l-th layer of the network, AlRepresenting the input of the fully-connected layer, Bernoulli represents the Bernoulli function,
representing a part of the neuron remaining with a probability q, Zl+1Represents the output of the fully connected layer, W represents the weight of the l-th layer, b represents the bias of the l-th layer, and R represents the nonlinear activation function.
5. The intelligent lymphoma pathology image identification method based on deep learning according to claim 1, wherein the full convolution neural network in the step 2) and the lymphoma three-classification convolution neural network in the step 3) are optimized by using an Adam optimizer, and the parameter optimization updating rule is as follows:
(1) calculating the exponential moving average of the t-time gradient
First, the gradient is calculated
Wherein, gtRepresenting the parameter theta at t time stepstThe gradient of the gradient to be obtained is determined,
representing a gradient operator, J (θ)t-1) Representing a differentiable random scalar function;
calculating the exponential moving average m of the t time gradient according to the gradienttThe formula is as follows:
mt＝β1mt-1+(1-β1)gt
the first order moment vector m is initialized to 0, coefficient beta1Is exponential decay rate, defaults to 0.9, and is used for controllingDistributing the weight;
(2) calculating the exponential moving average v of the square of the gradienttThe formula is as follows:
initializing the second moment vector v to 0 by the coefficient beta2The default is 0.999 for controlling the influence of the square of the previous gradient;
(3) since the first order moment vector m is initialized to 0, it results in an exponential moving average m of the t-time gradienttBiased toward 0, to exponentially move the mean m of the time gradient of ttAnd (3) correcting deviation to reduce the influence of the deviation on the initial training stage:
wherein,
the gradient mean value after deviation correction is carried out;
(4) since the initialization of the second moment vector v to 0 results in the training of the exponentially moving average v of the gradient squared in the initial phasetBiased toward 0, the exponentially moving average of the squared gradient is corrected as follows:
wherein,
estimating a second-order original moment after deviation correction;
(5) updating a parameter θ of a networktThe formula is as follows:
where α is a learning rate, the default learning rate α has an initial value of 0.001, and θt-1For the parameters of the network before updating, the constant is 10-8Avoid the divisor changing to 0;
when the network is trained, the learning rate α gradually attenuates with the increase of the training times, and the attenuation formula is:
where decay is expressed as decay rate, epoch _ num is expressed as number of training sessions, α0Expressed as the initial learning rate.
6. The method for intelligently identifying pathological images of lymphoma according to claim 1, wherein the full convolution neural network in step 2) and the lymphoma three-classification convolution neural network in step 3) use cross-entropy loss as a loss function for measuring the difference between two probability distributions, which is classified into two-classification and multi-classification cases: wherein,
(1) case of two classes
In the case of bisection, the final result to be predicted by the network has only two cases, and the prediction probabilities obtained by prediction for each category are p and 1-p, where the expression is:
wherein Loss is the Loss value of the cross quotient, yiA label representing the ith pathological section image in the data set, wherein the positive class is 1, and the negative class is 0; p is a radical ofiRepresenting the probability that the ith pathological section image in the data set is predicted to be positive; n represents the number of pathological section images in the data set;
(2) case of multiple classifications
The multi-classification case is an extension to two classes:
where M denotes the number of categories, yicOne-hot unique coding, p, representing the ith pathological section image category cicRepresenting the probability of predicting that the ith pathological section image belongs to the class c;
at the same time, since cross entropy involves calculating the probability of each class, the last layer of the network uses the softmax function, and the probability S taken for each class is in the functional form:
wherein e is a natural constant, j and k both represent category indexes, the total number of categories is C, and V is the input of the softmax layer;
since the softmax function maps values between 0-1, and the sum is 1, then there are:
and (3) carrying out derivation on the cross entropy Loss value Loss to obtain:
only need to find hjSubtracting the result by 1 is the gradient of the inverse update.
7. The method for intelligently identifying pathological images of lymphoma based on deep learning according to claim 1, wherein in the training of step 4), the output of the full convolution neural network is subjected to threshold binarization processing and then used as the input of the lymphoma three-classification convolution neural network.
8. The method for intelligently identifying the lymphoma pathology image based on deep learning according to claim 7, wherein the binarization threshold value set in the binarization processing is 0.5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010622536.2A CN111798464A (en) | 2020-06-30 | 2020-06-30 | Lymphoma pathological image intelligent identification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010622536.2A CN111798464A (en) | 2020-06-30 | 2020-06-30 | Lymphoma pathological image intelligent identification method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111798464A true CN111798464A (en) | 2020-10-20 |
Family
ID=72809938
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010622536.2A Pending CN111798464A (en) | 2020-06-30 | 2020-06-30 | Lymphoma pathological image intelligent identification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111798464A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112270666A (en) * | 2020-11-03 | 2021-01-26 | 辽宁工程技术大学 | Non-small cell lung cancer pathological section identification method based on deep convolutional neural network |
| CN112330627A (en) * | 2020-11-03 | 2021-02-05 | 广州信瑞医疗技术有限公司 | Slice image processing method and model training method and system |
| CN112364920A (en) * | 2020-11-12 | 2021-02-12 | 西安电子科技大学 | Thyroid cancer pathological image classification method based on deep learning |
| CN112560957A (en) * | 2020-12-17 | 2021-03-26 | 北京邮电大学 | Neural network training and detecting method, device and equipment |
| CN112581301A (en) * | 2020-12-17 | 2021-03-30 | 塔里木大学 | Detection and early warning method and system for residual film quantity of farmland based on deep learning |
| CN112634204A (en) * | 2020-12-03 | 2021-04-09 | 北京信息科技大学 | Cancer boundary intelligent identification method based on convolutional neural network |
| CN112651927A (en) * | 2020-12-03 | 2021-04-13 | 北京信息科技大学 | Raman spectrum intelligent identification method based on convolutional neural network and support vector machine |
| CN113160175A (en) * | 2021-04-23 | 2021-07-23 | 杭州迪英加科技有限公司 | Tumor lymphatic vessel infiltration detection method based on cascade network |
| CN113239993A (en) * | 2021-04-29 | 2021-08-10 | 中国人民解放军海军军医大学第三附属医院 | Pathological image classification method, pathological image classification system, terminal and computer-readable storage medium |
| CN113421250A (en) * | 2021-07-05 | 2021-09-21 | 北京理工大学 | Intelligent fundus disease diagnosis method based on lesion-free image training |
| CN113689376A (en) * | 2021-06-02 | 2021-11-23 | 北京信息职业技术学院 | Pathological image recognition method based on computer vision |
| CN113707312A (en) * | 2021-09-16 | 2021-11-26 | 人工智能与数字经济广东省实验室(广州) | Blood vessel quantitative identification method and device based on deep learning |
| CN113838558A (en) * | 2021-08-16 | 2021-12-24 | 电子科技大学 | Method and device for analyzing breast cancer pathological image based on convolutional neural network |
| CN114240836A (en) * | 2021-11-12 | 2022-03-25 | 杭州迪英加科技有限公司 | Nasal polyp pathological section analysis method and system and readable storage medium |
| WO2022166800A1 (en) * | 2021-02-02 | 2022-08-11 | 广州柏视医疗科技有限公司 | Deep learning network-based automatic delineation method for mediastinal lymphatic drainage region |
| CN114897922A (en) * | 2022-04-03 | 2022-08-12 | 西北工业大学 | Histopathology image segmentation method based on deep reinforcement learning |
| CN115083574A (en) * | 2022-08-22 | 2022-09-20 | 四川大学华西医院 | Cancer prognosis survival prediction method, system, computer device and storage medium |
| CN115294126A (en) * | 2022-10-08 | 2022-11-04 | 南京诺源医疗器械有限公司 | Intelligent cancer cell identification method for pathological image |
| TWI790572B (en) * | 2021-03-19 | 2023-01-21 | 宏碁智醫股份有限公司 | Detecting method and detecting apparatus related to image |
| CN116188474A (en) * | 2023-05-05 | 2023-05-30 | 四川省肿瘤医院 | Three-level lymphatic structure identification method and system based on image semantic segmentation |
| CN117455906A (en) * | 2023-12-20 | 2024-01-26 | 东南大学 | Digital pathological pancreatic cancer nerve segmentation method based on multi-scale cross fusion and boundary guidance |
| CN117636064A (en) * | 2023-12-21 | 2024-03-01 | 浙江大学 | Intelligent neuroblastoma classification system based on pathological sections of children |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034208A (en) * | 2018-07-03 | 2018-12-18 | 怀光智能科技（武汉）有限公司 | A kind of cervical cell pathological section classification method of high-low resolution combination |
| CN110060245A (en) * | 2019-04-15 | 2019-07-26 | 青岛大学 | A kind of gastric cancer lymph node dyeing pathological image automatic identifying method based on deep neural network |
| CN110660045A (en) * | 2019-08-30 | 2020-01-07 | 杭州电子科技大学 | Lymph node identification semi-supervision method based on convolutional neural network |
| CN110689525A (en) * | 2019-09-09 | 2020-01-14 | 上海中医药大学附属龙华医院 | Method and device for recognizing lymph nodes based on neural network |
-
2020
- 2020-06-30 CN CN202010622536.2A patent/CN111798464A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034208A (en) * | 2018-07-03 | 2018-12-18 | 怀光智能科技（武汉）有限公司 | A kind of cervical cell pathological section classification method of high-low resolution combination |
| CN110060245A (en) * | 2019-04-15 | 2019-07-26 | 青岛大学 | A kind of gastric cancer lymph node dyeing pathological image automatic identifying method based on deep neural network |
| CN110660045A (en) * | 2019-08-30 | 2020-01-07 | 杭州电子科技大学 | Lymph node identification semi-supervision method based on convolutional neural network |
| CN110689525A (en) * | 2019-09-09 | 2020-01-14 | 上海中医药大学附属龙华医院 | Method and device for recognizing lymph nodes based on neural network |
Non-Patent Citations (6)
| Title |
|---|
| GOUTHAM RAMAKRISHNAN ET AL.: "FAST GPU-ENABLED COLOR NORMALIZATION FOR DIGITAL PATHOLOGY", 《ARXIV》 * |
| SACHIN MEHTA ET AL.: "Y-Net: Joint Segmentation and Classification for Diagnosis of Breast Biopsy Images" * |
| SACHIN MEHTA ET AL.: "Y-Net: Joint Segmentation and Classification for Diagnosis of Breast Biopsy Images", 《MICCAI 2018》 * |
| 刘玉良等: "《深度学习》", 31 January 2020 * |
| 尚荣华等: "《人工智能技术入门》", 30 September 2019 * |
| 王玮: "基于Dropout深度信念网络的人脸图像识别研究", 《物联网技术》 * |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112270666A (en) * | 2020-11-03 | 2021-01-26 | 辽宁工程技术大学 | Non-small cell lung cancer pathological section identification method based on deep convolutional neural network |
| CN112330627A (en) * | 2020-11-03 | 2021-02-05 | 广州信瑞医疗技术有限公司 | Slice image processing method and model training method and system |
| CN112364920A (en) * | 2020-11-12 | 2021-02-12 | 西安电子科技大学 | Thyroid cancer pathological image classification method based on deep learning |
| CN112364920B (en) * | 2020-11-12 | 2023-05-23 | 西安电子科技大学 | Thyroid cancer pathological image classification method based on deep learning |
| CN112634204A (en) * | 2020-12-03 | 2021-04-09 | 北京信息科技大学 | Cancer boundary intelligent identification method based on convolutional neural network |
| CN112651927A (en) * | 2020-12-03 | 2021-04-13 | 北京信息科技大学 | Raman spectrum intelligent identification method based on convolutional neural network and support vector machine |
| CN112560957A (en) * | 2020-12-17 | 2021-03-26 | 北京邮电大学 | Neural network training and detecting method, device and equipment |
| CN112581301A (en) * | 2020-12-17 | 2021-03-30 | 塔里木大学 | Detection and early warning method and system for residual film quantity of farmland based on deep learning |
| CN112581301B (en) * | 2020-12-17 | 2023-12-29 | 塔里木大学 | Detection and early warning method and system for residual quantity of farmland residual film based on deep learning |
| CN112560957B (en) * | 2020-12-17 | 2023-04-07 | 北京邮电大学 | Neural network training and detecting method, device and equipment |
| WO2022166800A1 (en) * | 2021-02-02 | 2022-08-11 | 广州柏视医疗科技有限公司 | Deep learning network-based automatic delineation method for mediastinal lymphatic drainage region |
| TWI790572B (en) * | 2021-03-19 | 2023-01-21 | 宏碁智醫股份有限公司 | Detecting method and detecting apparatus related to image |
| CN113160175A (en) * | 2021-04-23 | 2021-07-23 | 杭州迪英加科技有限公司 | Tumor lymphatic vessel infiltration detection method based on cascade network |
| CN113239993A (en) * | 2021-04-29 | 2021-08-10 | 中国人民解放军海军军医大学第三附属医院 | Pathological image classification method, pathological image classification system, terminal and computer-readable storage medium |
| CN113239993B (en) * | 2021-04-29 | 2022-12-16 | 中国人民解放军海军军医大学第三附属医院 | Pathological image classification system, terminal and computer readable storage medium |
| CN113689376A (en) * | 2021-06-02 | 2021-11-23 | 北京信息职业技术学院 | Pathological image recognition method based on computer vision |
| CN113421250A (en) * | 2021-07-05 | 2021-09-21 | 北京理工大学 | Intelligent fundus disease diagnosis method based on lesion-free image training |
| CN113838558A (en) * | 2021-08-16 | 2021-12-24 | 电子科技大学 | Method and device for analyzing breast cancer pathological image based on convolutional neural network |
| CN113838558B (en) * | 2021-08-16 | 2023-04-18 | 电子科技大学 | Method and device for analyzing breast cancer pathological image based on convolutional neural network |
| CN113707312A (en) * | 2021-09-16 | 2021-11-26 | 人工智能与数字经济广东省实验室(广州) | Blood vessel quantitative identification method and device based on deep learning |
| CN114240836A (en) * | 2021-11-12 | 2022-03-25 | 杭州迪英加科技有限公司 | Nasal polyp pathological section analysis method and system and readable storage medium |
| CN114897922A (en) * | 2022-04-03 | 2022-08-12 | 西北工业大学 | Histopathology image segmentation method based on deep reinforcement learning |
| CN114897922B (en) * | 2022-04-03 | 2024-04-26 | 西北工业大学 | Tissue pathology image segmentation method based on deep reinforcement learning |
| CN115083574A (en) * | 2022-08-22 | 2022-09-20 | 四川大学华西医院 | Cancer prognosis survival prediction method, system, computer device and storage medium |
| CN115083574B (en) * | 2022-08-22 | 2022-12-06 | 四川大学华西医院 | Cancer prognosis survival prediction method, system, computer device and storage medium |
| CN115294126A (en) * | 2022-10-08 | 2022-11-04 | 南京诺源医疗器械有限公司 | Intelligent cancer cell identification method for pathological image |
| CN116188474A (en) * | 2023-05-05 | 2023-05-30 | 四川省肿瘤医院 | Three-level lymphatic structure identification method and system based on image semantic segmentation |
| CN117455906A (en) * | 2023-12-20 | 2024-01-26 | 东南大学 | Digital pathological pancreatic cancer nerve segmentation method based on multi-scale cross fusion and boundary guidance |
| CN117455906B (en) * | 2023-12-20 | 2024-03-19 | 东南大学 | Digital pathological pancreatic cancer nerve segmentation method based on multi-scale cross fusion and boundary guidance |
| CN117636064A (en) * | 2023-12-21 | 2024-03-01 | 浙江大学 | Intelligent neuroblastoma classification system based on pathological sections of children |
| CN117636064B (en) * | 2023-12-21 | 2024-05-28 | 浙江大学 | Intelligent neuroblastoma classification system based on pathological sections of children |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111798464A (en) | Lymphoma pathological image intelligent identification method based on deep learning | |
| CN111784671B (en) | Pathological image focus region detection method based on multi-scale deep learning | |
| CN112132817B (en) | Retina blood vessel segmentation method for fundus image based on mixed attention mechanism | |
| CN110097554B (en) | Retina blood vessel segmentation method based on dense convolution and depth separable convolution | |
| CN111488921B (en) | Intelligent analysis system and method for panoramic digital pathological image | |
| CN111462042B (en) | Cancer prognosis analysis method and system | |
| Moradi et al. | New features for automatic classification of human chromosomes: A feasibility study | |
| CN113674253A (en) | Rectal cancer CT image automatic segmentation method based on U-transducer | |
| CN113610859B (en) | Automatic thyroid nodule segmentation method based on ultrasonic image | |
| Popescu et al. | Retinal blood vessel segmentation using pix2pix gan | |
| Pandit et al. | Deep learning neural network for lung cancer classification: enhanced optimization function | |
| CN113450305B (en) | Medical image processing method, system, equipment and readable storage medium | |
| CN113420793A (en) | Improved convolutional neural network ResNeSt 50-based gastric ring cell carcinoma classification method | |
| CN115471701A (en) | Lung adenocarcinoma histology subtype classification method based on deep learning and transfer learning | |
| CN113129310B (en) | Medical image segmentation system based on attention routing | |
| CN115880245A (en) | Self-supervision-based breast cancer disease classification method | |
| CN115661185A (en) | Fundus image blood vessel segmentation method and system | |
| CN111178174B (en) | Urine formed component image identification method based on deep convolutional neural network | |
| CN114359308A (en) | Aortic dissection method based on edge response and nonlinear loss | |
| Harine et al. | Improved Breast Cancer Detection using Wavelet Based Image Enhancement | |
| Kaoungku et al. | Colorectal Cancer Histology Image Classification Using Stacked Ensembles | |
| CN113936165B (en) | CT image processing method, terminal and computer storage medium | |
| Chen et al. | Leukocyte Segmentation Method Based on Adaptive Retinex Correction and U‐Net | |
| Bharali et al. | Identification of the Types of Skin Cancers from Skin Cancer Images and Covid-19 Detection on Chest X-Ray Images using Deep Learning | |
| CN117994628A (en) | Deep learning-based clear cell carcinoma image histology analysis device for kidney |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201020 |