CN111951221B - Glomerular cell image recognition method based on deep neural network - Google Patents
Glomerular cell image recognition method based on deep neural network Download PDFInfo
- Publication number
- CN111951221B CN111951221B CN202010670136.9A CN202010670136A CN111951221B CN 111951221 B CN111951221 B CN 111951221B CN 202010670136 A CN202010670136 A CN 202010670136A CN 111951221 B CN111951221 B CN 111951221B
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- layer
- slice
- segmentation
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 97
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 68
- 210000003904 glomerular cell Anatomy 0.000 title claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 70
- 230000001575 pathological effect Effects 0.000 claims abstract description 56
- 230000011218 segmentation Effects 0.000 claims abstract description 44
- 210000004027 cell Anatomy 0.000 claims abstract description 43
- 238000003062 neural network model Methods 0.000 claims abstract description 32
- 230000001434 glomerular Effects 0.000 claims abstract description 27
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 210000003855 cell nucleus Anatomy 0.000 claims description 29
- 238000010586 diagram Methods 0.000 claims description 28
- 238000005520 cutting process Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 14
- 238000002372 labelling Methods 0.000 claims description 9
- 238000012805 post-processing Methods 0.000 claims description 8
- 210000004940 nucleus Anatomy 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000002790 cross-validation Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000003709 image segmentation Methods 0.000 abstract description 30
- 238000012545 processing Methods 0.000 abstract description 8
- 238000013473 artificial intelligence Methods 0.000 abstract description 5
- 238000013135 deep learning Methods 0.000 abstract description 5
- 238000005516 engineering process Methods 0.000 abstract description 4
- 230000004927 fusion Effects 0.000 abstract 1
- 210000003734 kidney Anatomy 0.000 abstract 1
- 238000000605 extraction Methods 0.000 description 11
- 230000000694 effects Effects 0.000 description 9
- 238000003745 diagnosis Methods 0.000 description 7
- 238000003672 processing method Methods 0.000 description 7
- 230000008602 contraction Effects 0.000 description 6
- 210000001519 tissue Anatomy 0.000 description 6
- 230000002159 abnormal effect Effects 0.000 description 5
- 230000002902 bimodal effect Effects 0.000 description 4
- 238000001574 biopsy Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000002360 preparation method Methods 0.000 description 3
- 238000011282 treatment Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000002059 diagnostic imaging Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 206010020718 hyperplasia Diseases 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000011862 kidney biopsy Methods 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 230000007170 pathology Effects 0.000 description 2
- 210000005084 renal tissue Anatomy 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 208000017169 kidney disease Diseases 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 230000004660 morphological change Effects 0.000 description 1
- 238000011369 optimal treatment Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000010827 pathological analysis Methods 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007790 scraping Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30024—Cell structures in vitro; Tissue sections in vitro
-
- 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
- G06T2207/30084—Kidney; Renal
-
- 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
- G06T2207/30242—Counting objects in image
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The application discloses a glomerular cell image recognition method based on a deep neural network, which is used for acquiring a pathological image to be detected based on artificial intelligence and a deep learning technology; preprocessing the pathological image to obtain a plurality of slice images; inputting each slice image into a preset neural network model for identification and segmentation to obtain a glomerular region map; cell counting was performed on glomerular area map; the method can quickly and accurately divide the sub-image of the glomerulus in the kidney in the pathological image, count the cells in the glomerulus by using the traditional method and the deep learning fusion model, and solve the problems of high workload and low efficiency and high misdiagnosis rate of manually identifying the glomerulus in the pathological image; the application relates to the field of biomedical image processing, in particular to a method for optimizing glomerular sub-image segmentation and glomerular cell counting algorithms in pathological images, improving segmentation and counting accuracy by using more data training algorithms.
Description
Technical Field
The application relates to the technical field of biomedical image processing and deep learning, in particular to a glomerular cell image recognition method based on a deep neural network.
Background
The rapid development of modern medicine has prompted the detection methods and display means to become more accurate, intuitive and complete. Human medical images contain extremely rich human information, which, when acquired by means of medical imaging techniques, need to be analyzed, identified, segmented, calibrated, classified and interpreted, and for clinical applications of medical images and medical problems to be solved, it is determined which parts should be enhanced or which features should be extracted, thereby providing more intuitive data. The method can be used as a basis for reasonably arranging the examination procedures of the patient, so that the objective diagnosis purpose can be achieved by the fastest speed and the most economical means, and the selection, the determination and the implementation of the optimal treatment scheme are further facilitated. Therefore, the processing of acquired medical images by high-tech means is an urgent research work in medical imaging. The method has important significance for reducing the misdiagnosis probability caused by the factors of the person in charge, improving the working efficiency of medical workers and providing more objective diagnosis basis. Clinically, biopsy is generally used for kidney disease examination, and the diagnosis result obtained by biopsy is most accurate. Biopsy, known as biopsy, is performed by performing pathological examination of lesion tissue in a living body of a patient by surgical methods such as partial excision, forceps, needle suction, scraping, and removal, to determine diagnosis. The method has the advantages of fresh tissue and basically keeping the original appearance of the lesions. For clinical work, the examination method is helpful for timely and accurate diagnosis of diseases and treatment effect judgment. In general, a trained physician obtains pathological morphological changes of glomeruli and hyperplasia and distribution of nuclei in the glomeruli by observing tissue slice images obtained by the kidney biopsy, and analyzes the pathological changes and the hyperplasia and distribution of the nuclei in the glomeruli in combination with clinical experience of the physician to obtain a series of conclusions, thereby giving a pathological diagnosis report. To ensure that the existing problems can be found, the doctor needs to observe the obtained kidney biopsy tissue sections as much as possible, and about 800 pictures of each person to be examined clinically, as can be seen from the figure, the workload of identifying the target object by naked eyes is great, the efficiency is low, and the misdiagnosis rate is increased along with the increase of the fatigue degree of the observer. With the rapid development of computer technology, imaging technology and image processing technology, automatic analysis of kidney tissue sections is possible, and after computer intelligent processing is carried out on the graph, traditional visual observation and main pipe judgment can be abandoned in image diagnosis. The deep learning algorithm based on glomerulus segmentation and intraglomerulus cell counting of pathological images is provided, glomerulus extraction and intraglomerulus cell nucleus counting are completed, and qualitative analysis and quantitative calculation of kidney tissue section images are realized. At present, the glomerular segmentation and the cell segmentation and technical accuracy thereof are still to be improved, and cannot reach the practical level, and further optimization is required, so that the glomerular segmentation accuracy is improved and the intraglomerular cell count is realized. It has mainly 3 drawbacks: 1. the traditional method for identifying glomerulus in pathological images by naked eyes by doctors has the defects of great workload and low efficiency, and the misdiagnosis rate can be increased along with the increase of the fatigue degree of observers. 2. The existing glomerular segmentation and glomerular cell counting algorithms have low accuracy, and need to be continuously optimized to improve the accuracy of segmentation and counting. 3. Because the pathological image is large, cells in the glomerulus are dense, the identification rate is low by using the traditional single neural network method, and the cells can be better identified by a specific post-processing method after deep learning segmentation.
Disclosure of Invention
The application aims to provide a glomerular cell image recognition method based on a deep neural network, which aims to solve one or more technical problems in the prior art and at least provides a beneficial selection or creation condition.
The application provides a glomerular cell image recognition method based on a deep neural network, which utilizes a corresponding ROI extraction algorithm to extract a region of interest of a pathological image, filters out a blank region of the pathological image and reduces the workload of subsequent image cutting; the process of observing the slice by a doctor by using a microscope is simulated by adopting a multi-layer cutting or layer-by-layer focusing picture cutting method; generating multi-level slices with different sizes for preparing subsequent characteristic extraction when generating the slices, and enhancing the generalization performance of learning; the method for detecting the glomerular cells by using the image segmentation method based on the threshold value, such as the combination of a histogram bimodal image segmentation method, a fixed threshold value image segmentation method, a half threshold value segmentation image segmentation method, an iterative threshold value image segmentation method, a self-adaptive threshold value image segmentation method, an optimal threshold value image segmentation method and the like with FastFCN, segNet, unet and deep neural network cell detection algorithms such as variants, refinnenet and the like, realizes more accurate counting of the glomerular cells.
The application aims at solving the problems and provides a glomerular cell image identification method based on a deep neural network, which specifically comprises the following steps:
acquiring a pathological image to be detected;
preprocessing the pathological image to obtain a plurality of slice images;
inputting each slice image into an identification segmentation neural network model for identification segmentation to obtain a glomerular region slice image;
the slice diagram of the glomerulus region is input into a preset neural network model to identify and count cells in the glomerulus.
Further, the method for preprocessing the pathological image to obtain a plurality of slice images comprises the following steps:
labeling and extracting a region of interest in a pathological image:
the region of the pathological image to be processed, namely Region of Interest (ROI), is marked manually in the modes of boxes, circles, ellipses, irregular polygons and the like. For pathological full-field slicing (Whole Slide Image, WSI), an edge blank area generated during slice preparation and gaps among tissues exist in an image, and the workload and the working difficulty of subsequent processing can be effectively reduced by utilizing a corresponding ROI extraction algorithm.
Performing multi-layer cutting on the region of interest in the pathological image to obtain a plurality of slice images:
a large feature of WSI pathology images is the very large number of pixels in the image, typically in the scale range of 10≡6 x 10≡6. Because of the physical memory and the computing speed limitations of today's computers, it is impractical to directly input the entire WSI into the computer memory for operation. Therefore, image segmentation, i.e. image slicing, is performed on the characteristics of the WSI image, and the pathological image is subjected to image slicing into a plurality of slice images and then to corresponding training and prediction. The multi-layer cutting or layer-by-layer focused picture cutting method simulates the process of a doctor observing a slice using a microscope. In general, with the field of view of the lens unchanged, the physical field of view is scaled down and the pixel resolution is scaled up as the magnification increases. Using the concept of layer-by-layer focusing, multiple layers of different sized slices are generated in preparation for subsequent feature extraction when generating the slices. The method can ensure that the image characteristics of textures, colors and the like with different scales can be well observed in the deep convolutional neural network, and the large focus and the tiny focus of the whole block can be effectively captured, so that the generalization performance of learning is enhanced.
Further, the method for extracting the region of interest in the pathological image comprises any one of histogram bimodal image segmentation, fixed threshold image segmentation, half threshold segmentation image segmentation, iterative threshold image segmentation, self-adaptive threshold image segmentation and optimal threshold image segmentation.
Further, the method for obtaining the slice map of the glomerulus region by inputting each slice image into the identification and segmentation neural network model for identification and segmentation comprises the following steps:
constructing a FastFCN, segNet, U-net and any one neural network of various variants and refinnenets;
dividing each slice image into a training set and a testing set by any one of a leave-out method, a cross-validation method, a leave-in method and a self-service method;
the preset number of slice images in the training set can be selected by adjusting the size of the slice images according to the learning and training requirements of the neural network; the greater the predetermined number, the greater the accuracy of the identified segmented neural network model trained.
And inputting the slice images in the training set into a neural network for training to obtain a slice image in the training set, inputting the slice images in the neural network for training to obtain an identification segmentation neural network model, and inputting each slice image into a preset neural network model for identification segmentation to obtain a slice image containing the glomerulus region.
Preferably, the neural network is a U-Net, the U-Net is a full convolution neural network, and the neural network is an end-to-end network, namely, the input and the output are images; inputting a pathological image into U-Net, performing downsampling on the pathological image by adopting an activation function after passing through a convolution layer in a contraction path, extracting a feature map of the pathological image, performing upsampling in an expansion path, adding a corresponding feature map obtained in the contraction path during each upsampling, and finally realizing the extraction of a first abnormal probability matrix of each pixel point in the pathological image; the input of the model is three-channel pictures with the size of 256 x 3 (256, 3), single-channel pictures with the size of 256 x 1 (256, 1) are output, and the input size of the pictures can be adjusted according to actual conditions.
Preferably, the U-Net is composed of an encoder and a decoder, the input of the encoder is a slice image, the encoder outputs a 963×32 size feature map output by the fourth layer to the fourth layer of the decoder, and the feature maps output by the first, second and third layers of the encoder are respectively output to the first, second and third layers of the decoder; the encoder is used for extracting cell characteristic information: color, morphology, size, location, texture, etc., each consisting of 5 levels, each level containing several modified ResBlock and using a downsampling operation to generate the next level of input, the encoder input image in the neural network being a 256 x 3 RGB color map, the encoder first level generating a 256 x 67 size first level of feature map by 1 modified ResBlock, downsampling by 1 to obtain a 128 x 67 size 2 level of input, downsampling by 1 to generate a 128 x 195 size second level of feature map by 1 to obtain a 64 x 195 size 3 level of input, the third layer generates a third layer characteristic diagram with the size of 64 x 451 through 1 improved Resblock, the 4 th layer input with the size of 32 x 451 is obtained through 1 time downsampling, the fourth layer generates a fourth layer characteristic diagram with the size of 32 x 963 through 1 time improved Resblock, the 5 th layer input with the size of 16 x 963 is obtained through 1 time downsampling, the fifth layer generates a characteristic diagram with the size of 16 x 1987 through 1 time improved Resblock as a final characteristic diagram of a decoder, and therefore the encoder extracts global information of the whole image with low resolution through level-by-level convolution and downsampling operation; the context information of the cells in the whole image can be provided, and the context information can be understood as the characteristics of the dependency relationship between the cells and the surrounding environment, and the characteristics are helpful for judging the category of the cells.
The decoder consists of 4 layers, and the second layer, the third layer and the fourth layer respectively output the feature images to the upper layer through up sampling;
the fourth level first generates a feature map with the size of 32 x 512 from the feature map with the size of 16 x 963 of the 5 th level of the encoder through up-sampling, then splices the feature maps with the same size as the fourth level of the encoder through jump connection, and finally generates a feature map with the size of 32 x 1987 of the fourth level of the decoder through 1 improved Resblock. Similarly, the third layer, the second layer and the first layer of the decoder perform the same process, and the corresponding produced feature sizes are 64×64×963, 128×128×451, 256×256×195. The jump connection on the same level of the encoder and the decoder can fuse the fine position information of the encoder and the rich expressed semantic information of the decoder, thereby extracting the characteristic with stronger expression capability for the segmentation task.
Furthermore, the application uses the improved U-Net deep neural network to divide glomerulus and intraglomerular cells in the small picture, combines the feature images by referring to the short connection mode of the ResNet model and the concat mode in the DenseNet on the basis of the traditional U-Net, namely the final summation operation of the original Resblock is replaced by the concat operation, namely the improved Resblock, thereby obtaining the basic structure used by the application, and further realizing the multiplexing of the feature images and improving the effect of the model. The network can identify pathological features of glomerulus, can accurately segment glomerulus, and the segmentation accuracy rate can reach 98.1%.
Training the neural network through the training set to obtain a slice image in the training set, and inputting the slice image into the neural network for training to obtain a preset neural network model;
inputting the slice images in the training set into a neural network for training;
because the slice images in the training set are still bigger, the slice images in the training set are directly input into the neural network for training, and because of the limitation of computer hardware, a computer reports memory overflow errors, so that training fails; therefore, before inputting the pictures into the neural network for training, the slice images in the training set are cut into sizes of 256×256×3 (256, 3), then every 8 pictures of 256×256×3 are used as a batch for training in the neural network, the binary label images corresponding to the slice images are also processed in the same way, and the sizes of the slice images and the inputted batch are cut into small images which can be adjusted according to the hardware level corresponding to the computer.
Further, the training process of the preset neural network model comprises the following steps:
marking glomeruli in a pathological image, and performing sub-image overlapping cutting on a region of interest (ROI) in the pathological image according to a certain size (for example, 2048 x 3 size) to obtain sub-image data;
before inputting the sub-image data into the neural network for training, performing image enhancement operation, wherein the image enhancement operation comprises random overturn along the x or y axis of an image matrix, rotation according to a random angle and random contrast adjustment, so that the generalization capability of the neural network is improved;
after the glomerulus prediction model is trained, the prediction effect of the model is further improved by using a specific post-processing method for the model output result, namely, the prediction result is optimized to obtain a final predicted glomerulus label, and the post-processing method comprises the steps of corroding, expanding and screening the prediction result; dividing the glomerulus from the pathological image by using openlide according to the finally predicted glomerular marker, generating a segmentation subgraph containing the glomerulus image and a binary labeling image corresponding to the segmentation subgraph, marking the part which does not contain the glomerulus in the binary labeling image as a background as 0, marking the part which contains the glomerulus as an abnormal value as 1, and removing cells in the region except the glomerulus by using the binary image of the glomerulus in the cell counting process;
further, the glomerular region slice is input into a preset neural network model to identify cells in glomerulus and count as follows:
performing cell nucleus marking on the glomerular region slice, and taking the marked binary image as a label corresponding to the glomerular cell nucleus; after the nuclei in the glomerulus are marked and corresponding binary images are generated, the segmentation subgraphs are used as training data, and the corresponding binary images are used as labels to be input into an improved neural network for training.
After the cell prediction model is trained, the trained neural network is used for predicting the cell nucleus in the segmented subgraph, the predicted cell nucleus is the cell nucleus in the whole segmented subgraph and also comprises the cell nucleus of the area outside the glomerulus, at the moment, the binary image of the glomerulus in the corresponding segmented subgraph is multiplied by the predicted cell nucleus binary image, the area outside the glomerulus is changed into 0, and the numerical value of the area inside the glomerulus is unchanged.
Dividing glomeruli from a pathological image according to the marking of the glomeruli to obtain a division sub-image containing the glomeruli image and generating a binary image corresponding to the division sub-image, marking the part which does not contain the glomeruli in the binary image as a background as 0, and marking the part which contains the glomeruli as 1; inputting the slice images into a neural network for training to obtain a preset neural network model.
And then optimizing the cell nucleus prediction result by using morphological post-processing methods and the like, solving the problems of noise, adhesion and the like existing in the cell nucleus prediction result, and improving the final prediction result.
The number of nuclei in the predicted binary image is then counted to perform cell counting.
The beneficial effects of the application are as follows: the application discloses a glomerular cell image recognition method based on a deep neural network, which effectively combines an artificial intelligence method with a traditional post-treatment method: the identification of glomeruli and cells thereof is a complex process, the traditional image processing method is difficult to realize, and meanwhile, the deep learning model cannot well correct a certain detail, and the traditional image processing method can be better adjusted for the specific detail. Therefore, the algorithm uses an artificial intelligence method to carry out overall prediction and uses a traditional image processing method to carry out detail adjustment, so that a better cell identification effect is achieved. The problems that the workload of the traditional doctor for identifying glomerulus in pathological images by naked eyes is large, the efficiency is low, and the misdiagnosis rate is higher and higher for long-time observation are solved; the glomerular sub-image segmentation and glomerular cell counting algorithm in the pathological image is optimized, and the segmentation and counting accuracy is improved by using more data training algorithms.
Drawings
The above and other features of the present application will become more apparent from the detailed description of the embodiments thereof given in conjunction with the accompanying drawings, in which like reference characters designate like or similar elements, and it is apparent that the drawings in the following description are merely some examples of the present application, and other drawings may be obtained from these drawings without inventive effort to those of ordinary skill in the art, in which:
FIG. 1 is a flow chart of a method for identifying glomerular cells based on deep neural network according to the present application;
FIG. 2 is a graph showing the contrast of images before and after a threshold segmentation method;
fig. 3 is a diagram showing a neural network model used in the present application.
Detailed Description
The conception, specific structure, and technical effects produced by the present application will be clearly and completely described below with reference to the embodiments and the drawings to fully understand the objects, aspects, and effects of the present application. It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other.
A flowchart of a method for identifying glomerular cells based on a deep neural network in accordance with the present application is shown in fig. 1, and a method in accordance with an embodiment of the present application is described below in conjunction with fig. 1.
The application provides a glomerular cell image recognition method based on a deep neural network, which specifically comprises the following steps:
acquiring a pathological image to be detected;
preprocessing the pathological image to obtain a plurality of slice images;
inputting each slice image into an identification segmentation neural network model for identification segmentation to obtain a glomerular region slice image;
the slice diagram of the glomerulus region is input into a preset neural network model to identify and count cells in the glomerulus.
Further, the method for preprocessing the pathological image to obtain a plurality of slice images comprises the following steps:
labeling and extracting a region of interest in a pathological image:
in machine vision, image processing problems, a region to be processed, i.e., a region of interest, region of Interest (ROI), is outlined from the processed image in the form of a box, circle, ellipse, irregular polygon, or the like. For pathological full-view slice/pathological image (Whole Slide Image, WSI), an edge blank area generated during film production and gaps among tissues exist in the image, and the workload and the working difficulty of subsequent processing can be effectively reduced by utilizing a corresponding ROI extraction algorithm.
Performing multi-layer cutting on the region of interest in the pathological image to obtain a plurality of slice images:
a large feature of pathology images (WSIs) is the extremely high number of pixels in the image, typically in the scale range of 10≡6 x 10≡6. Because of the physical memory and the computing speed limitations of today's computers, it is impractical to directly input the entire WSI into the computer memory for operation. Therefore, image segmentation, i.e. image slicing, is performed on the characteristics of the WSI image, and the pathological image is subjected to image slicing into a plurality of slice images and then to corresponding training and prediction. The multi-layer cutting or layer-by-layer focused picture cutting method simulates the process of a doctor observing a slice using a microscope. In general, with the field of view of the lens unchanged, the physical field of view is scaled down and the pixel resolution is scaled up as the magnification increases. Using the concept of layer-by-layer focusing, multiple layers of different sized slices are generated in preparation for subsequent feature extraction when generating the slices. The method can ensure that the image characteristics of textures, colors and the like with different scales can be well observed in the deep convolutional neural network, and the large focus and the tiny focus of the whole block can be effectively captured, so that the generalization performance of learning is enhanced.
Further, the method for extracting the region of interest in the pathological image comprises any one of histogram bimodal image segmentation, fixed threshold image segmentation, half threshold segmentation image segmentation, iterative threshold image segmentation, self-adaptive threshold image segmentation and optimal threshold image segmentation.
As shown in fig. 2, the image contrast map before and after the threshold segmentation method uses a threshold-based image segmentation method, such as histogram bimodal image segmentation, fixed threshold image segmentation, half-threshold segmentation image segmentation, iterative threshold image segmentation, adaptive threshold image segmentation, optimal threshold image segmentation, and the like. The method uses the clustering idea, and divides the gray level number of the image into 2 parts according to gray level levels, so that the gray level value difference between the two parts is maximum. The application applies the processing method to the extraction of the ROI in the WSI image so as to reduce the subsequent calculation amount.
Further, the method for obtaining the slice map of the glomerulus region by inputting each slice image into the identification and segmentation neural network model for identification and segmentation comprises the following steps:
constructing a FastFCN, segNet, U-net and any one neural network of various variants and refinnenets;
dividing each slice image into a training set and a testing set by any one of a leave-out method, a cross-validation method, a leave-in method and a self-service method;
the method for training the neural network through the training set to obtain the slice images in the training set to be input into the neural network for training to obtain the preset neural network model comprises the following steps: and inputting the slice images in the training set into the neural network for training to obtain a preset neural network model, and inputting the slice images in the training set into the neural network for training.
The predetermined number of slice images in the training set can be selected by changing the slice size according to the learning and training requirements of the neural network; the greater the predetermined number, the greater the accuracy of the trained preset neural network model.
Preferably, the neural network is a U-Net, the U-Net is a full convolution neural network, and the neural network is an end-to-end network, namely, the input and the output are images; inputting a pathological image into U-Net, performing downsampling on the pathological image by adopting an activation function after passing through a convolution layer in a contraction path, extracting a feature map of the pathological image, performing upsampling in an expansion path, adding a corresponding feature map obtained in the contraction path during each upsampling, and finally realizing the extraction of a first abnormal probability matrix of each pixel point in the pathological image; the input of the model is three-channel pictures with the size of 256 x 3 (256, 3), single-channel pictures with the size of 256 x 1 (256, 1) are output, and the input size of the pictures can be adjusted according to actual conditions.
As shown in fig. 3, which shows a neural network model structure diagram used in the present application, the neural network model structure U-Net is composed of an encoder and a decoder, the input of the encoder is a slice image, the encoder outputs a 963×32 size feature map output by a fourth layer to the fourth layer of the decoder, and outputs feature maps output by the first, second and third layers to the first, second and third layers of the decoder, respectively; the encoder is used for extracting cell characteristic information: color, morphology, size, location, texture, etc., each consisting of 5 levels, each level containing several modified ResBlock and using a downsampling operation to generate the next level of input, the encoder input image in the neural network being a 256 x 3 RGB color map, the encoder first level generating a 256 x 67 size first level of feature map by 1 modified ResBlock, downsampling by 1 to obtain a 128 x 67 size 2 level of input, downsampling by 1 to generate a 128 x 195 size second level of feature map by 1 to obtain a 64 x 195 size 3 level of input, the third layer generates a third layer characteristic diagram with the size of 64 x 451 through 1 improved Resblock, the 4 th layer input with the size of 32 x 451 is obtained through 1 time downsampling, the fourth layer generates a fourth layer characteristic diagram with the size of 32 x 963 through 1 time improved Resblock, the 5 th layer input with the size of 16 x 963 is obtained through 1 time downsampling, the fifth layer generates a characteristic diagram with the size of 16 x 1987 through 1 time improved Resblock as a final characteristic diagram of a decoder, and therefore the encoder extracts global information of the whole image with low resolution through level-by-level convolution and downsampling operation; the context information of the cells in the whole image can be provided, and the context information can be understood as the characteristics of the dependency relationship between the cells and the surrounding environment, and the characteristics are helpful for judging the category of the cells.
The decoder consists of 4 layers, and the second layer, the third layer and the fourth layer respectively output the feature images to the upper layer through up sampling;
the fourth level first generates a feature map with the size of 32 x 512 from the feature map with the size of 16 x 963 of the 5 th level of the encoder through up-sampling, then splices the feature maps with the same size as the fourth level of the encoder through jump connection, and finally generates a feature map with the size of 32 x 1987 of the fourth level of the decoder through 1 improved Resblock. Similarly, the third layer, the second layer and the first layer of the decoder perform the same process, and the corresponding produced feature sizes are 64×64×963, 128×128×451, 256×256×195. The jump connection on the same level of the encoder and the decoder can fuse the fine position information of the encoder and the rich expressed semantic information of the decoder, thereby extracting the characteristic with stronger expression capability for the segmentation task.
Furthermore, the application uses the improved U-net neural network to divide cells in the small picture, the network can identify pathological characteristics of glomerulus, and the glomerular cells can be accurately divided.
Training the neural network through the training set to obtain a slice image in the training set, and inputting the slice image into the neural network for training to obtain a preset neural network model;
inputting the slice images in the training set into a neural network for training;
because the slice images in the training set are still bigger, the slice images in the training set are directly input into the neural network for training, and because of the limitation of computer hardware, a computer reports memory overflow errors, so that training fails; therefore, before inputting the pictures into the neural network for training, the slice images in the training set are cut into 256 x 3 images, then every 8 pictures with 256 x 3 images are used as a batch for training in the neural network, the binary images corresponding to the slice images are also processed in the same way, and the sizes of the slice images and the inputted batch are cut into small images which can be adjusted according to the hardware level corresponding to the computer.
After the glomerular prediction model is trained, the prediction effect of the model is further improved by using a specific post-processing method (including but not limited to corrosion, expansion and screening of the prediction result) on the model output result, namely, the prediction result is optimized, so that the finally predicted glomerular label is obtained.
Further, the glomerular region slice diagram is input into a preset neural network model to identify cells in glomerulus, and the counting process is as follows:
the glomeruli is cut out from WSI by using openslide according to the glomeruli labeling, the glomeruli is stored as jpg format, and a glomeruli binary labeling picture corresponding to the glomeruli is generated, the background which is the part which does not contain the glomeruli in the binary labeling picture is set as 0, the abnormal value which is the part which contains the glomeruli is set as 1, and the cell nucleus of the region outside the glomeruli is removed by using the binary labeling image of the glomeruli in the cell counting process.
Training a glomerular intracell prediction model:
after marking the cell nucleus in the glomerulus slice and generating a corresponding cell nucleus binary marking graph, the glomerulus slice is used as training data, the corresponding cell nucleus binary marking data is used as a label to be input into an improved U-Net model for training, and the result of the model is the same as that of a glomerulus prediction model, except that the input is changed into 128 x 3, and the output is changed into 128 x 1.
When the picture is input into the model training, considering that the segmentation algorithm has poor prediction effect on the edge of the picture, before the picture is input into the model training, the glomerular slice picture is cut into smaller sizes, so that overlapping cutting and prediction can be performed during subsequent prediction.
After training the neural network, predicting the cell nucleus in the slice image through a preset neural network model, wherein the predicted cell nucleus in the glomerulus slice image also comprises the cell nucleus of the area outside the glomerulus, and at the moment, multiplying the binary label image corresponding to the glomerulus slice image by the predicted cell nucleus binary label image, so that the area outside the glomerulus is changed into 0, and the numerical value of the area inside the glomerulus is unchanged.
And then optimizing the cell nucleus prediction result by using morphological post-processing methods and the like, solving the problems of noise, adhesion and the like existing in the cell nucleus prediction result, and improving the final prediction result. The number of nuclei in the predicted binary image is then counted to perform cell counting.
The method effectively combines the artificial intelligence method with the traditional post-treatment method: the identification of glomeruli and cells thereof is a complex process, the traditional image processing method is difficult to realize, and meanwhile, the deep learning model cannot well correct a certain detail, and the traditional image processing method can be better adjusted for the specific detail. Therefore, the algorithm uses an artificial intelligence method to carry out overall prediction and uses a traditional image processing method to carry out detail adjustment, so that a better cell identification effect is achieved.
Further, in performing predictive diagnosis on a renal pathological WSI map, the algorithm flow can be summarized as follows: WSI interesting region cutting map; predicting glomerular regions by a glomerular prediction model; correcting the prediction result by the traditional method, and cutting out the glomerular region; predicting cells within the glomerular region by a cell prediction model; the traditional method corrects the predicted result and counts the number of cells in the current glomerular region.
U-Net is a full convolution neural network, which is an end-to-end network, namely, the input and the output are images. Inputting a pathological image into a U-Net segmentation model, performing downsampling on the pathological image by adopting an activation function after passing through a convolution layer in a contraction path, extracting a feature map of the pathological image, performing upsampling in an expansion path, adding a corresponding feature map obtained in the contraction path during each upsampling, and finally realizing the extraction of a first abnormal probability matrix of each pixel point in the pathological image.
The input of the model is a three-channel picture (256, 3), the output is a single-channel picture (256, 1), and the input size of the picture can be adjusted according to actual conditions.
In order to ensure the accuracy of cell counting, a deep neural network cell detection algorithm is also used, and can rapidly detect all cells in the glomerulus after a great amount of training. And finally, combining the results obtained by the two methods to obtain the accurate cell number.
Although the description of the present disclosure has been illustrated in considerable detail and with particularity, it is not intended to be limited to any such detail or embodiment or any particular embodiment so as to effectively cover the intended scope of the present disclosure. Furthermore, the foregoing description of the present disclosure has been presented in terms of embodiments foreseen by the inventor for the purpose of providing a enabling description for enabling the enabling description to be available, notwithstanding that insubstantial changes in the disclosure, not presently foreseen, may nonetheless represent equivalents thereto.
Claims (3)
1. A glomerular cell image recognition method based on a deep neural network, which is characterized by comprising the following steps:
acquiring a pathological image to be detected;
preprocessing the pathological image to obtain a plurality of slice images;
inputting each slice image into an identification segmentation neural network model for identification segmentation to obtain a glomerular region slice image;
inputting the slice diagram of the glomerulus region into a preset neural network model to identify and count cells in the glomerulus;
the method for inputting the glomerular region slice diagram into a preset neural network model to identify and count cells in glomerulus comprises the following steps:
marking a cell nucleus region in the glomerular region slice diagram, and taking the binary image marked by the cell nucleus region as a label corresponding to the glomerular cell nucleus;
after labeling a cell nucleus region in the glomerular region slice diagram and generating a corresponding binary image, taking the segmentation subgraph as training data, and inputting the corresponding binary image as a label into a neural network for training;
after the neural network training is completed, predicting the cell nucleus in the segmented subgraph by using the trained neural network, wherein the predicted cell nucleus region is the cell nucleus in the whole segmented subgraph and also comprises the cell nucleus of the region outside the glomerulus, and multiplying the binary image of the glomerulus in the corresponding segmented subgraph by the predicted cell nucleus binary image at the moment to enable the region outside the glomerulus to be 0 and the numerical value of the region inside the glomerulus to be unchanged;
optimizing the predicted result by a post-processing method to obtain a final predicted cell nucleus label, wherein the post-processing method comprises the steps of screening the predicted result by corroding the predicted result and a threshold value; finally, counting the number of cell nuclei in the predicted binary image so as to count the cells;
the neural network is a U-Net, the U-Net is a full convolution neural network, and the neural network is an end-to-end network, namely, the input and the output are images;
the U-Net consists of an encoder and a decoder, wherein the input of the encoder is a slice image, the encoder outputs a characteristic diagram output by a fourth layer to the decoder, and the characteristic diagrams output by a first layer, a second layer and a third layer of the encoder are respectively output to the first layer, the second layer and the third layer of the decoder; the encoder is used for extracting cell characteristic information: color, morphology, size, position and texture, each layer comprises a plurality of improved ResBlock and downsampling operation, the 1 st layer of the encoder in the neural network generates a first layer of characteristic diagram through 1 improved ResBlock, the 2 nd layer of characteristic diagram is obtained through 1 downsampling, the 2 nd layer generates a second layer of characteristic diagram through 1 improved ResBlock, the 3 rd layer of characteristic diagram is obtained through 1 downsampling, the third layer generates a third layer of characteristic diagram through 1 improved ResBlock, the 4 th layer of characteristic diagram is obtained through 1 downsampling, the fourth layer generates a fourth layer of characteristic diagram through 1 improved ResBlock, the 5 th layer of characteristic diagram is obtained through 1 downsampling, and the fifth layer generates a final decoder characteristic diagram through 1 improved ResBlock, so that the encoder extracts global information of a low-resolution whole image through level-by-level convolution and downsampling operation; the decoder consists of 4 layers, wherein the fourth layer firstly generates a characteristic image of the 5 th layer of the encoder through up-sampling, then splices the characteristic images with the same size as the fourth layer of the encoder through jump connection, and finally generates a fourth layer characteristic image of the decoder through 1 improved ResBlock; similarly, the decoder performs the same process on the 3 rd layer, the 2 nd layer and the 1 st layer;
the improved ResBlock is obtained by replacing the last summation operation in the original ResBlock with a concat operation.
2. The method for identifying glomerular cell images based on deep neural network according to claim 1, wherein the method for obtaining glomerular region slice images by inputting each slice image into an identification and segmentation neural network model for identification and segmentation comprises the following steps:
the acquisition process of the identification and segmentation neural network model comprises the following steps:
constructing a FastFCN, segNet, U-net and any one neural network of various variants and refinnenets;
dividing each slice image into a training set and a testing set by any one of a leave-out method, a cross-validation method, a leave-in method and a self-service method;
and inputting the slice images in the training set into the neural network for training to obtain a recognition segmentation neural network model, inputting the slice images in the training set into the neural network for recognition segmentation, and inputting the slice images into the recognition segmentation neural network model for recognition segmentation to obtain a slice image containing the glomerulus region.
3. The method for identifying the glomerular cell image based on the deep neural network according to claim 1, wherein the training process of the preset neural network model comprises the following steps:
marking glomeruli in the pathological image, and performing sub-image overlapping cutting on an interested region in the pathological image to obtain sub-image data;
before inputting the sub-image data into the neural network for training, performing image enhancement operation, wherein the image enhancement operation comprises random overturn along the x or y axis of an image matrix, rotation according to a random angle and random contrast adjustment;
dividing glomeruli from a pathological image according to the marking of the glomeruli to obtain a division sub-image containing the glomeruli image and generating a binary image corresponding to the division sub-image, marking the part which does not contain the glomeruli in the binary image as a background as 0, and marking the part which contains the glomeruli as 1; inputting the slice images into a neural network for training to obtain a preset neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010670136.9A CN111951221B (en) | 2020-07-13 | 2020-07-13 | Glomerular cell image recognition method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010670136.9A CN111951221B (en) | 2020-07-13 | 2020-07-13 | Glomerular cell image recognition method based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111951221A CN111951221A (en) | 2020-11-17 |
| CN111951221B true CN111951221B (en) | 2023-10-31 |
Family
ID=73341298
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010670136.9A Active CN111951221B (en) | 2020-07-13 | 2020-07-13 | Glomerular cell image recognition method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111951221B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112233112B (en) * | 2020-12-07 | 2021-06-01 | 深圳大学 | Prediction method of CK19 of hepatocellular carcinoma based on deep learning analysis |
| CN112562855B (en) * | 2020-12-18 | 2021-11-02 | 深圳大学 | Hepatocellular carcinoma postoperative early recurrence risk prediction method, medium and terminal equipment |
| CN114387201B (en) * | 2021-04-08 | 2023-01-17 | 透彻影像科技(南京)有限公司 | Cytopathic image auxiliary diagnosis system based on deep learning and reinforcement learning |
| CN113192077B (en) * | 2021-04-15 | 2022-08-02 | 华中科技大学 | Automatic classification method and system for pathological graphs of cells and regional levels |
| CN113222944B (en) * | 2021-05-18 | 2022-10-14 | 湖南医药学院 | Cell nucleus segmentation method and cancer auxiliary analysis system and device based on pathological image |
| CN113256577B (en) * | 2021-05-18 | 2022-06-28 | 湖南医药学院 | Cancer auxiliary analysis system and device based on HE staining pathological image |
| CN113362294B (en) * | 2021-05-27 | 2022-09-02 | 同济大学 | Puncture needle identification method, system and equipment containing puncture needle ultrasonic blood vessel image |
| CN113591919B (en) * | 2021-06-29 | 2023-07-21 | 复旦大学附属中山医院 | Analysis method and system for prognosis of early hepatocellular carcinoma postoperative recurrence based on AI |
| CN113409293A (en) * | 2021-06-30 | 2021-09-17 | 上海市第一人民医院 | Pathology image automatic segmentation system based on deep learning |
| CN114219807B (en) * | 2022-02-22 | 2022-07-12 | 成都爱迦飞诗特科技有限公司 | Mammary gland ultrasonic examination image grading method, device, equipment and storage medium |
| CN114943723A (en) * | 2022-06-08 | 2022-08-26 | 北京大学口腔医学院 | Method for segmenting and counting irregular cells and related equipment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2019135939A (en) * | 2018-02-07 | 2019-08-22 | 澁谷工業株式会社 | Cell count method, cell count device and cell culture method |
| CN110151181A (en) * | 2019-04-16 | 2019-08-23 | 杭州电子科技大学 | Rapid magnetic resonance imaging method based on the U-shaped network of recurrence residual error |
| CN110427970A (en) * | 2019-07-05 | 2019-11-08 | 平安科技（深圳）有限公司 | Image classification method, device, computer equipment and storage medium |
| CN110490840A (en) * | 2019-07-11 | 2019-11-22 | 平安科技（深圳）有限公司 | A kind of cell detection method, device and the equipment of glomerulus pathology sectioning image |
| CN111080575A (en) * | 2019-11-22 | 2020-04-28 | 东南大学 | Thalamus segmentation method based on residual error dense U-shaped network model |
| WO2020093042A1 (en) * | 2018-11-02 | 2020-05-07 | Deep Lens, Inc. | Neural networks for biomedical image analysis |
| CN111275191A (en) * | 2020-02-26 | 2020-06-12 | 上海商汤智能科技有限公司 | Method and apparatus for detecting cell, electronic device, and storage medium |
| CN111340711A (en) * | 2020-05-21 | 2020-06-26 | 腾讯科技（深圳）有限公司 | Super-resolution reconstruction method, device, equipment and storage medium |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018091486A1 (en) * | 2016-11-16 | 2018-05-24 | Ventana Medical Systems, Inc. | Convolutional neural networks for locating objects of interest in images of biological samples |
| US10789451B2 (en) * | 2017-11-16 | 2020-09-29 | Global Life Sciences Solutions Usa Llc | System and method for single channel whole cell segmentation |
| CN108734211B (en) * | 2018-05-17 | 2019-12-24 | 腾讯科技（深圳）有限公司 | Image processing method and device |
-
2020
- 2020-07-13 CN CN202010670136.9A patent/CN111951221B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2019135939A (en) * | 2018-02-07 | 2019-08-22 | 澁谷工業株式会社 | Cell count method, cell count device and cell culture method |
| WO2020093042A1 (en) * | 2018-11-02 | 2020-05-07 | Deep Lens, Inc. | Neural networks for biomedical image analysis |
| CN110151181A (en) * | 2019-04-16 | 2019-08-23 | 杭州电子科技大学 | Rapid magnetic resonance imaging method based on the U-shaped network of recurrence residual error |
| CN110427970A (en) * | 2019-07-05 | 2019-11-08 | 平安科技（深圳）有限公司 | Image classification method, device, computer equipment and storage medium |
| CN110490840A (en) * | 2019-07-11 | 2019-11-22 | 平安科技（深圳）有限公司 | A kind of cell detection method, device and the equipment of glomerulus pathology sectioning image |
| CN111080575A (en) * | 2019-11-22 | 2020-04-28 | 东南大学 | Thalamus segmentation method based on residual error dense U-shaped network model |
| CN111275191A (en) * | 2020-02-26 | 2020-06-12 | 上海商汤智能科技有限公司 | Method and apparatus for detecting cell, electronic device, and storage medium |
| CN111340711A (en) * | 2020-05-21 | 2020-06-26 | 腾讯科技（深圳）有限公司 | Super-resolution reconstruction method, device, equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| Glomerulosclerosis identification in whole slide images using semantic segmentation;Gloria Bueno 等;《Computer Methods and Programs in Biomedicine》;第184卷(第105273期);1-10 * |
| 应用化肝脏病灶分割的预测方案;裘静韬 等;《计算机工程与应用》;第56卷(第20期);173-178 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111951221A (en) | 2020-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111951221B (en) | Glomerular cell image recognition method based on deep neural network | |
| CN110599448B (en) | Migratory learning lung lesion tissue detection system based on MaskScoring R-CNN network | |
| CN111985536B (en) | Based on weak supervised learning gastroscopic pathology image Classification method | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| CN109636808B (en) | Lung lobe segmentation method based on full convolution neural network | |
| Liu et al. | A framework of wound segmentation based on deep convolutional networks | |
| CN110974306B (en) | System for discernment and location pancreas neuroendocrine tumour under ultrasonic endoscope | |
| CN112132166B (en) | Intelligent analysis method, system and device for digital cell pathology image | |
| Pan et al. | Mitosis detection techniques in H&E stained breast cancer pathological images: A comprehensive review | |
| CN111340827A (en) | Lung CT image data processing and analyzing method and system | |
| CN111476794B (en) | Cervical pathological tissue segmentation method based on UNET | |
| CN112215217B (en) | Digital image recognition method and device for simulating doctor to read film | |
| CN112085113B (en) | Severe tumor image recognition system and method | |
| CN112102332A (en) | Cancer WSI segmentation method based on local classification neural network | |
| CN115546605A (en) | Training method and device based on image labeling and segmentation model | |
| Kromp et al. | Deep Learning architectures for generalized immunofluorescence based nuclear image segmentation | |
| CN115206495A (en) | Renal cancer pathological image analysis method and system based on CoAtNet deep learning and intelligent microscopic device | |
| CN115880266B (en) | Intestinal polyp detection system and method based on deep learning | |
| EP4118617A1 (en) | Automated detection of tumors based on image processing | |
| CN112837325A (en) | Medical image processing method, device, electronic equipment and medium | |
| Azli et al. | Ultrasound image segmentation using a combination of edge enhancement and kirsch’s template method for detecting follicles in ovaries | |
| CN105956587A (en) | Method for automatically extracting meniscus from knee-joint magnetic resonance image sequence based on shape constraint | |
| Cheng et al. | Improved faster RCNN for white blood cells detection in blood smear image | |
| JPWO2019044579A1 (en) | Pathological diagnostic equipment, image processing methods and programs | |
| CN112990367A (en) | Image processing method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |