CN111612750A - Overlapping chromosome segmentation network based on multi-scale feature extraction - Google Patents
Overlapping chromosome segmentation network based on multi-scale feature extraction Download PDFInfo
- Publication number
- CN111612750A CN111612750A CN202010401332.6A CN202010401332A CN111612750A CN 111612750 A CN111612750 A CN 111612750A CN 202010401332 A CN202010401332 A CN 202010401332A CN 111612750 A CN111612750 A CN 111612750A
- Authority
- CN
- China
- Prior art keywords
- path
- res
- convolution
- module
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a multi-scale U-shaped convolution neural network MACS Net aiming at the problems that target segmentation areas in overlapped chromosome images are different in size and not obvious in distinction and the like. Introducing a multilayer cavity convolution and same-step pooling technology at the bottommost layer of the UNet to realize detection of target segmentation areas with different sizes and extraction of features; convolutional block connection is introduced between UNet codecs, and semantic information difference of UNet codecs is relieved. The intersection ratio (IoU) of the chromosome overlapping region is taken as an evaluation index, and the result shows that the division IoU of MACS Net in the chromosome overlapping part reaches 0.9860, which is improved by 2.78% compared with 0.9593 of UNet, and the MACS Net respectively shows more ideal noise robustness in data sets of salt and pepper, Gaussian and Poisson noise pollution.
Description
Technical Field
The invention belongs to the field of medical image segmentation, and particularly relates to a multi-scale feature extraction and image segmentation method which can be used for segmentation of overlapped chromosomes in a medical image and subsequent karyotype analysis.
Background
There are 23 pairs of chromosomes, including 22 pairs of autosomes and 1 pair of sex chromosomes, in the healthy cell nucleus of humans. As a gene vector, chromosomal abnormality accounts for more than 50% of spontaneous abortion, stillbirths and early deaths, and is also an important cause of many congenital diseases, and the neonatal morbidity is about 1%. Chromosomal abnormalities, including quantitative variations and morphological structural aberrations, may occur on each chromosome. Therefore, how to discriminate the abnormality of chromosome number and morphological structure has become a key approach for genetic disease diagnosis, especially early diagnosis.
The chromosome shows a clear and stable form in the metaphase stage of mitosis, so the karyotype analysis usually takes the metaphase chromosome as a research object, and the diagnosis of the morphological structure and the number variation of the chromosome is performed by analyzing, comparing, sequencing and numbering according to the characteristics of the chromosome length, the position of the centromere, the proportion of the long arm and the short arm, the existence of the satellite and the like by various banding techniques, and the karyotype analysis is an important auxiliary means for diagnosing genetic diseases. Common banding techniques include chromosomal banding and Fluorescence In Situ Hybridization (FISH). The FISH technology, which was introduced at the end of the 70's 20 th century, enables better visualization of target DNA under a fluorescent microscope by combining probes of fluorescent substances with DNA in chromosomes. The FISH technique allows for structural detection of the stained portion, thereby facilitating detection of factors such as chromosomal structure deletions, additions and variations that lead to genetic diseases. However, even though the chromosomes are numbered in the same way, the flexible material can show different curved forms in the cell nucleus at different times, and clustering phenomena can be generated due to contact and overlapping of the chromosomes. At present, chromosome karyotype analysis firstly segments chromosomes, then classifies and numbers, and finally identifies whether abnormalities occur through means such as morphological analysis.
Chromosome segmentation is used as the first step of karyotyping to directly determine the accuracy and reliability of subsequent chromosome classification and abnormality detection. Furthermore, statistics show that up to 40% of chromosomes are touching and overlapping at metaphase, and that touching and overlapping of two chromosomes is the most common, but at present, the segmentation of overlapping chromosomes is heavily dependent on manual work. For example, Sharma and the like distribute the data sets to various mass-market platforms in a crowdsourcing mode for manual segmentation, and then summarize the data sets to complete subsequent classification and anomaly identification. The manual segmentation method is heavily dependent on the experience of the operator and is time-consuming and labor-intensive. Therefore, how to automatically and effectively segment chromosomes, especially overlapped chromosomes, has become a key link for karyotyping.
Most of the traditional automatic segmentation methods are implemented based on geometric morphology. For example, Balaji et al first extracts the edges of overlapping chromosomes and calculates their curvatures based on the Otsu thresholding method, finds intersections (pits) and tangents therein, and then implements semi-automatic segmentation of overlapping chromosomes by Voronoi diagrams and Delaunay triangulation. Somasundalam et al first separate a single chromosome by using a multi-objective geodesic contour method, for overlapping chromosomes, identify cut points on an image by a curvature function, draw a hypothesis line on an overlapping region by using the obtained cut points, and finally segment the overlapping chromosomes. Yilmaz et al propose a method of thresholding and watershed segmentation to separate individual chromosomes and chromosome clusters, calculate out the cut points of the chromosome clusters through a curvature function, and finally segment overlapped chromosomes through an optimal geodesic path between the cut points. The method for determining the intersection point of the chromosome overlapping part for segmentation based on the curvature has the problems of false judgment and missing judgment of effective pits, and the accuracy rate is to be improved generally.
With the development of deep learning in recent years, a batch of high-performance deep convolutional neural networks appear, and the pixel-level segmentation of images can be realized. The full convolution network adopts a full convolution layer as an output layer of the network, introduces transposition convolution and realizes pixel-level semantic segmentation of an image for the first time; the UNet adopts a U-shaped structure, the network is divided into an encoding part and a decoding part, and the encoding part and the decoding part are fused through hop connection, can be segmented by utilizing the shallow characteristic and the deep characteristic of the image, and can achieve an ideal segmentation effect only by using a small amount of data; the dense convolutional network establishes direct connection among all layers, for each layer, the output characteristics of all the layers in front are used as the input of the dense convolutional network, and the output of the dense convolutional network is also used as the input of all the subsequent layers, so that the maximum information flow among all the layers of the network is ensured, the gradient disappearance problem of the deep network is relieved, the propagation of the characteristics is enhanced, the multiplexing of the characteristics reduces the parameter quantity of the network, and the transmission efficiency of the information and the gradient in the network is improved. The target areas segmented by the chromosome images are different in size, and the semantic features are not rich enough, so that the analysis of the overlapped chromosome images is difficult. UNet effectively improves the accuracy of the segmentation of the overlapped chromosomes by the strong feature extraction capability of UNet. If the detection capability of the target segmentation area can be improved, and the detail segmentation of the overlapped part is expected to further improve the segmentation accuracy.
Disclosure of Invention
The invention aims to provide an overlapped chromosome segmentation method based on a multi-scale U-shaped convolutional neural network aiming at the defects in the prior art so as to improve the segmentation precision of overlapped chromosomes.
The technical idea of the invention is as follows: the invention establishes a network structure MACS Net (MAC + SSPM UNet, MACS Net) for multi-scale division of overlapped chromosomes based on UNet, better extracts multi-scale spatial features of overlapped parts of chromosomes by designing MAC and SSPM modules, and integrates Res Path module to more fully utilize context information and semantic information in the network, thereby effectively improving the division of overlapped regions of chromosomes with different sizes.
The implementation scheme comprises the following steps:
(1) performing data amplification on the overlapped chromosome images;
(1a) amplifying the overlapping chromosome images to 128 x 128 size;
(1b) generating a pixel-level category label image with a corresponding size;
(2) constructing a synchronous long Pooling Module (SSPM);
(2a) unifying the step length of each pooling layer, and setting the step length to be 2;
(2b) designing pooling sizes with different sizes in each pooling layer, and reducing each pooled characteristic diagram to 1 dimension through 1 × 1 convolution;
(2c) obtaining a multi-scale feature map with the same size as the original map through 2 times of upsampling, finally stacking each feature map and outputting through 1 × 1 convolution operation;
(3) constructing a Multi-layer hole Convolution Module (MAC);
(3a) the MAC is provided with five branches, wherein the four branches only reserve one layer of cavity convolution, the number of filling cavities of each cavity convolution module is increased one by one, and the fifth branch is not operated;
(3b) applying a 1 × 1 convolution on three branches to perform linear correction, and finally adding the outputs of the five branches;
(4) constructing a Res Path module;
(4a) adding a series of convolution blocks on a Path of simple jump connection to form a Res Path, thereby relieving the difference of semantic information between an encoder and a decoder;
(4b) MACS Net adopts five Res Path modules to replace the original five hop connections, which are respectively marked as Res Path I, Res Path II, Res Path III, Res Path IV and Res Path V;
(4c) the most convolution blocks are designed in the first connection, and the number of the convolution blocks in other paths is reduced one by one;
(5) constructing MACS Net;
(5a) the invention is based on Unet, and proposes MACS Net network by setting SSPM, MAC and Res Path modules, etc., the network body is composed of 27 standard convolution layers, 5 pooling layers and 5 up-sampling layers;
(5b) replacing the original convolution module with the MAC module in (4) and the SSPM module in (3) at the lowest layer of the network;
(5c) the network of the invention adopts Res Path module in (5) to realize the jump connection between the coder and the decoder;
(6) training the MACS Net network;
in order to effectively avoid over-learning and under-learning and comprehensively consider the calculation cost, the invention develops a 5-fold cross-validation experiment and counts IoU scores of each region of a test set of the cross-validation experiment for final performance evaluation. The network uniformly adopts an Adam optimizer to minimize an objective function, which is an optimization method which has better performance and can adaptively adjust the learning rate.
Compared with the prior art, the invention has the following advantages:
1. the invention realizes the segmentation of the overlapped chromosomes with higher precision;
2. the MAC module designed in the MACS Net provided by the invention shows better noise robustness;
3. the SSPM module designed in the MACS Net provided by the invention shows more stable data generalization capability;
4. the Res Path module integrated in the MACS Net relieves semantic information difference between the codecs and improves the overall segmentation effect of the network.
Drawings
FIG. 1 is a MACS Net network architecture; FIG. 2 is a block diagram of an SSPM module; FIG. 3 is a diagram of a MAC module architecture; FIG. 4 is a diagram of a ResPath V module; FIG. 5 is an overlay chromosome image and category label map.
Detailed description of the preferred embodiments
The invention is described in further detail below with reference to the following figures and specific examples:
step 1, performing data amplification on the overlapped chromosome image;
1a) amplifying the size of the overlapped chromosome images to 128 x 128 size;
1b) a pixel level class label image of a corresponding size is generated as shown in fig. 5. Wherein (a) is a chromosomeαAndβoverlapping composite images, (b) - (e) are their corresponding category label images, and the light color regions in (b) and (c) correspond to chromosomes, respectivelyαAndβcorresponding to the overlapping area, (d) corresponding to the background area;
step 2, constructing an SSPM module as shown in FIG. 2;
2a) unifying the step length of each pooling layer, and setting the step length to be 2;
2b) considering that the size of the feature map at the bottommost layer of the network is 4 multiplied by 4, the pooling sizes of different sizes are designed to be 2, 3 and 4 in each pooling layer, and each feature map after pooling is reduced to 1 dimension through 1 multiplied by 1 convolution;
2c) obtaining a multi-scale feature map with the same size as the original map through 2 times of upsampling, finally stacking each feature map and outputting through 1 × 1 convolution operation;
step 3, constructing an MSC module, as shown in fig. 3;
3a) the MAC is provided with five branches, wherein the four branches only reserve one layer of cavity convolution, the number of filling cavities of the cavity convolution modules in each branch is increased one by one, and the fifth branch is not operated;
3b) applying a 1 × 1 convolution to the three branches with the number of holes of 2, 3 and 4 to perform linear correction, and finally adding the outputs of the five branches;
step 4, constructing a Res Path module;
4a) adding a series of convolution blocks on a Path of simple jump connection to form a Res Path, thereby relieving the difference of semantic information between an encoder and a decoder;
4b) MACS Net adopts five Res Path modules to replace the original five-hop connection, which are respectively marked as Res Path I, Res Path II, Res Path III, Res Path V and Res Path V, wherein the Res Path V module is shown in figure 4;
4c) considering that the most information difference exists in Res Path I, the most convolution blocks are designed, the number of the convolution blocks in other paths is reduced one by one, and the configuration parameters of each link Path are shown in Table 1;
step 5, constructing MACS Net, as shown in FIG. 1;
5a) the invention is based on Unet, and proposes MACS Net network by setting SSPM, MAC and Res Path modules, etc., the network body is composed of 27 standard convolution layers, 5 pooling layers and 5 up-sampling layers;
5b) replacing the original convolution module with the MAC in the step 4 and the SSPM in the step 3 at the lowest layer of the network to extract richer multi-scale spatial features;
5c) the network of the invention adopts Res Path module in step 5 to realize the jump connection between the coder and the decoder, and fully utilizes the context information and semantic information in the network while extracting the space characteristic;
and 6, training the MACS Net network.
In order to effectively avoid over-learning and under-learning and comprehensively consider the problem of calculation cost, 5-fold cross validation experiments are carried out on a data set, all overlapped chromosome images are divided into 5 parts in each group of experiments, each part of data is respectively used as a test set, the other 4 parts of data are used as training sets, 5 models are respectively trained, and IoU scores of all regions of the test sets are counted for final performance evaluation. The network uniformly adopts an Adam optimizer to minimize an objective function, which is an optimization method which has better performance and can adaptively adjust the learning rate.
The technical effects of the invention are further explained by combining simulation tests as follows:
the experimental environment of the invention is configured as follows: the computer processor is Intel (R) Xeon (R) W-2175 CPU @2.50GHz, 64GB operating memory, NVIDIA GeForce RTX 2080Ti GPU, Keras framework.
TABLE 1 Res Path parameter Table
In summary, the invention provides a MACS Net network for extracting multi-scale features and relieving semantic information difference, and high-precision segmentation of overlapped chromosomes is realized. The network is specially designed with an MAC module for fixing the number of the cavity convolution layers and an SSPM module for fixing the step length of the pooling, and adopts a Res Path module to realize jump connection, thereby improving the feature extraction capability and the detection capability of the network on multi-scale targets, and remarkably improving the performance of overlapping chromosome segmentation. By taking the IoU score of the chromosome overlapping part as an evaluation index, the division IoU of MACSNet in the chromosome overlapping part reaches 0.9860, and is improved by 2.78 percent compared with the currently most commonly used UNet (0.9593).
Claims (6)
1. A method for segmenting overlapping chromosomes based on a multi-scale U-shaped convolutional neural network is characterized by comprising the following steps:
(1) performing data amplification on the overlapped chromosome images;
(1a) amplifying the overlapping chromosome images to 128 x 128 size;
(1b) generating a pixel-level category label image with a corresponding size;
(2) constructing a synchronous long Pooling Module (SSPM);
(2a) unifying the step length of each pooling layer, and setting the step length to be 2;
(2b) respectively designing pooling sizes with different sizes in each pooling layer, and reducing each pooled characteristic diagram to 1 dimension through 1 × 1 convolution;
(2c) obtaining a multi-scale feature map with the same size as the original map through 2 times of upsampling, finally stacking each feature map and outputting through 1 × 1 convolution operation;
(3) constructing a Multi-layer hole Convolution Module (MAC);
(3a) the MAC is provided with five branches, wherein only one layer of cavity convolution is reserved in the four branches, the number of filling cavities of the cavity convolution module in each branch is increased one by one, and the fifth branch is not operated;
(3b) applying a 1 × 1 convolution to the three branches with the number of holes larger than 1 to perform linear correction, and finally adding the outputs of the five branches;
(4) constructing a Res Path module;
(4a) adding a series of convolution blocks on a Path of simple jump connection to form a Res Path, thereby relieving the difference of semantic information between an encoder and a decoder;
(4b) MACS Net adopts five Res Path modules to replace the original five hop connections, which are respectively marked as Res Path I, Res Path II, Res Path III, Res Path IV and Res Path V;
(4c) the most convolution blocks are designed in the first connection, and the number of the convolution blocks in other paths is reduced one by one;
(5) constructing MACS Net;
(5a) the invention provides a MACS Net network by setting SSPM, MAC and Res Path modules and the like based on UNet, wherein a network main body consists of 27 standard convolution layers, 5 pooling layers and 5 up-sampling layers;
(5b) replacing the original convolution module with the MAC module in (4) and the SSPM module in (3) at the lowest layer of the network;
(5c) the network of the invention adopts Res Path module in (5) to realize the jump connection between the coder and the decoder;
(6) training the MACS Net network;
in order to effectively avoid over-learning and under-learning and comprehensively consider the calculation cost, the invention develops a 5-fold cross validation experiment and counts IoU scores of each region of a test set of the cross validation experiment for final performance evaluation; the network uniformly adopts an Adam optimizer to minimize an objective function, which is an optimization method which has better performance and can adaptively adjust the learning rate.
2. The method of claim 1, wherein in step (2b), different sizes of pooling sizes 2, 3 and 4 are designed at each pooling level, taking into account the size of the network's lowest level feature map as 4 x 4.
3. The method of claim 1, wherein in step (3b) a further 1 x 1 convolution is applied to the three branches with the number of holes being 2, 3 and 4 to perform the linear correction.
4. The method of claim 1, wherein step (4c) takes into account the most information differences in Res Path i, and therefore the most volume blocks are designed.
5. The method according to claim 1, wherein the SSPM module is used in step (5b) to replace the original convolution module of the UNet, so that richer multi-scale spatial features can be extracted;
and (5) replacing simple hop connection by using a Res Path module in the step (5c), and fully utilizing context information and semantic information in the network while extracting the spatial features.
6. The method of claim 1, wherein the 5-fold cross validation experiment in step (6) is performed by dividing all overlapped chromosome images into 5 parts in each experiment, and training 5 models respectively by using each part of data as a test set and using the other 4 parts of data as a training set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010401332.6A CN111612750B (en) | 2020-05-13 | 2020-05-13 | Overlapping chromosome segmentation network based on multi-scale feature extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010401332.6A CN111612750B (en) | 2020-05-13 | 2020-05-13 | Overlapping chromosome segmentation network based on multi-scale feature extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111612750A true CN111612750A (en) | 2020-09-01 |
| CN111612750B CN111612750B (en) | 2023-08-11 |
Family
ID=72203173
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010401332.6A Active CN111612750B (en) | 2020-05-13 | 2020-05-13 | Overlapping chromosome segmentation network based on multi-scale feature extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111612750B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112215847A (en) * | 2020-09-30 | 2021-01-12 | 武汉大学 | Method for automatically segmenting overlapped chromosomes based on counterstudy multi-scale features |
| CN112819792A (en) * | 2021-02-03 | 2021-05-18 | 杭州高斯洪堡科技有限公司 | DualNet-based urban area change detection method |
| CN113989502A (en) * | 2021-10-25 | 2022-01-28 | 湖南自兴智慧医疗科技有限公司 | Chromosome segmentation identification method and device based on graph convolution neural network and electronic equipment |
| CN114519723A (en) * | 2021-12-24 | 2022-05-20 | 上海海洋大学 | Meteorite crater automatic extraction method based on pyramid image segmentation |
| CN116129123A (en) * | 2023-02-27 | 2023-05-16 | 中国矿业大学 | End-to-end chromosome segmentation method based on uncertainty calibration and region decomposition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109191471A (en) * | 2018-08-28 | 2019-01-11 | 杭州电子科技大学 | Based on the pancreatic cell image partition method for improving U-Net network |
| CN111127449A (en) * | 2019-12-25 | 2020-05-08 | 汕头大学 | Automatic crack detection method based on encoder-decoder |
| CN111145170A (en) * | 2019-12-31 | 2020-05-12 | 电子科技大学 | Medical image segmentation method based on deep learning |
-
2020
- 2020-05-13 CN CN202010401332.6A patent/CN111612750B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109191471A (en) * | 2018-08-28 | 2019-01-11 | 杭州电子科技大学 | Based on the pancreatic cell image partition method for improving U-Net network |
| CN111127449A (en) * | 2019-12-25 | 2020-05-08 | 汕头大学 | Automatic crack detection method based on encoder-decoder |
| CN111145170A (en) * | 2019-12-31 | 2020-05-12 | 电子科技大学 | Medical image segmentation method based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| 宁霄;赵鹏;: "基于U-Net卷积神经网络的年轮图像分割算法" * |
| 张成成;宋婕萍;徐淑琴;李卉;徐闰红;王小艳;张立;游齐靖;张凯;林浩添;: "基于深度卷积神经网络对中期染色体分类的应用研究" * |
| 谢源;苗玉彬;张舒;: "基于空洞全卷积网络的叶片分割算法" * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112215847A (en) * | 2020-09-30 | 2021-01-12 | 武汉大学 | Method for automatically segmenting overlapped chromosomes based on counterstudy multi-scale features |
| CN112819792A (en) * | 2021-02-03 | 2021-05-18 | 杭州高斯洪堡科技有限公司 | DualNet-based urban area change detection method |
| CN113989502A (en) * | 2021-10-25 | 2022-01-28 | 湖南自兴智慧医疗科技有限公司 | Chromosome segmentation identification method and device based on graph convolution neural network and electronic equipment |
| CN113989502B (en) * | 2021-10-25 | 2024-06-07 | 湖南自兴智慧医疗科技有限公司 | Chromosome segmentation recognition method and device based on graph convolution neural network and electronic equipment |
| CN114519723A (en) * | 2021-12-24 | 2022-05-20 | 上海海洋大学 | Meteorite crater automatic extraction method based on pyramid image segmentation |
| CN114519723B (en) * | 2021-12-24 | 2024-05-28 | 上海海洋大学 | Pyramid image segmentation-based meteorite crater automatic extraction method |
| CN116129123A (en) * | 2023-02-27 | 2023-05-16 | 中国矿业大学 | End-to-end chromosome segmentation method based on uncertainty calibration and region decomposition |
| CN116129123B (en) * | 2023-02-27 | 2024-01-05 | 中国矿业大学 | End-to-end chromosome segmentation method based on uncertainty calibration and region decomposition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111612750B (en) | 2023-08-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111612750A (en) | Overlapping chromosome segmentation network based on multi-scale feature extraction | |
| CN110472616B (en) | Image recognition method and device, computer equipment and storage medium | |
| CN110415233A (en) | Pavement crack rapid extracting method based on two step convolutional neural networks | |
| CN110120040A (en) | Sectioning image processing method, device, computer equipment and storage medium | |
| CN112085741B (en) | Gastric cancer pathological section segmentation algorithm based on deep learning | |
| CN106021984A (en) | Whole-exome sequencing data analysis system | |
| CN110705403A (en) | Cell sorting method, cell sorting device, cell sorting medium, and electronic apparatus | |
| CN110738637B (en) | Automatic classification system for breast cancer pathological sections | |
| CN113177927B (en) | Bone marrow cell classification and identification method and system based on multiple features and multiple classifiers | |
| CN113222933A (en) | Image recognition system applied to renal cell carcinoma full-chain diagnosis | |
| CN110852330A (en) | Behavior identification method based on single stage | |
| CN110414317B (en) | Full-automatic leukocyte classification counting method based on capsule network | |
| CN112233085A (en) | Cervical cell image segmentation method based on pixel prediction enhancement | |
| CN117036288A (en) | Tumor subtype diagnosis method for full-slice pathological image | |
| CN113658199B (en) | Regression correction-based chromosome instance segmentation network | |
| CN113408480B (en) | Artificial intelligent auxiliary diagnosis system for blood diseases based on bone marrow cell morphology | |
| CN117670895A (en) | Immunohistochemical pathological image cell segmentation method based on section re-staining technology | |
| CN117576131A (en) | Weak supervision cell nucleus segmentation method and device based on edge optimization and feature denoising | |
| CN110717916B (en) | Pulmonary embolism detection system based on convolutional neural network | |
| CN112508860A (en) | Artificial intelligence interpretation method and system for positive check of immunohistochemical image | |
| CN113158950B (en) | Automatic segmentation method for overlapped chromosomes | |
| CN109887603A (en) | A kind of computer-assisted medical data processing system and method | |
| CN112381838B (en) | Automatic image cutting method for digital pathological section image | |
| CN114022485A (en) | Computer-aided diagnosis method for colorectal cancer based on small sample learning | |
| CN114418949A (en) | Pulmonary nodule detection method based on three-dimensional U-shaped network and channel attention |
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 |