CN114283938A - CNN-XGboost-based glioblastoma prognosis prediction method - Google Patents
CNN-XGboost-based glioblastoma prognosis prediction method Download PDFInfo
- Publication number
- CN114283938A CN114283938A CN202111409937.0A CN202111409937A CN114283938A CN 114283938 A CN114283938 A CN 114283938A CN 202111409937 A CN202111409937 A CN 202111409937A CN 114283938 A CN114283938 A CN 114283938A
- Authority
- CN
- China
- Prior art keywords
- image
- tumor
- network
- model
- feature map
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 208000005017 glioblastoma Diseases 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000004393 prognosis Methods 0.000 title claims abstract description 22
- 206010028980 Neoplasm Diseases 0.000 claims abstract description 44
- 238000000605 extraction Methods 0.000 claims abstract description 36
- 238000013135 deep learning Methods 0.000 claims abstract description 16
- 238000007781 pre-processing Methods 0.000 claims abstract description 12
- 210000004556 brain Anatomy 0.000 claims abstract description 10
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 230000011218 segmentation Effects 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 9
- 238000010837 poor prognosis Methods 0.000 claims description 7
- 201000010915 Glioblastoma multiforme Diseases 0.000 claims description 5
- 230000004931 aggregating effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 230000007306 turnover Effects 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 238000005315 distribution function Methods 0.000 claims description 2
- 230000004927 fusion Effects 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 239000000284 extract Substances 0.000 claims 1
- 238000011144 upstream manufacturing Methods 0.000 abstract description 7
- 238000013461 design Methods 0.000 abstract description 4
- 230000001537 neural effect Effects 0.000 abstract 1
- 238000010801 machine learning Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 238000011282 treatment Methods 0.000 description 3
- 206010018338 Glioma Diseases 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011369 optimal treatment Methods 0.000 description 2
- 238000004445 quantitative analysis Methods 0.000 description 2
- 208000032612 Glial tumor Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 201000007983 brain glioma Diseases 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000036210 malignancy Effects 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000010882 preoperative diagnosis Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 230000001225 therapeutic effect Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a CNN-XGboost-based glioblastoma prognosis prediction method, which comprises the following steps: s1 brain MR image basic preprocessing; s2 extracting a tumor feature map based on the mixed double-Gaussian model; s3 data enhancement; s4 deep learning feature extraction; s5, establishing a prognosis model based on the traditional mechanistic XGboost feature classifier. The method is used for fitting the intensity histogram of the MR image before the operation of the patient based on the double Gaussian models to extract the tumor feature map, so that the difficulty of feature extraction of a subsequent network model is reduced. Meanwhile, the invention designs a deep convolution characteristic extraction network which comprises an upstream branch and a downstream branch. The upstream branch is a simple feature extraction module which mainly focuses on extracting simple features of the segmented label image. The downstream branch is a depth feature extraction network of the tumor feature map, which mainly focuses on the tumor feature map obtained after preprocessing. In addition, the XGboost is combined to perform classified fitting on the features extracted by the deep neural convolution network, so that the overfitting phenomenon can be effectively relieved, and the performance such as the accuracy of prognosis prediction is improved.
Description
Technical Field
The invention relates to a medical technology, in particular to a CNN-XGboost-based glioblastoma multiforme prognosis prediction method.
Background
Brain glioma is one of the most serious malignancies, accounting for 70% of adult primary tumors. The World Health Organization (WHO) has classified gliomas into four grades, with the grade IV glioma known as Glioblastoma (GBM) being the most fatal, with short overall survival time and prompt therapeutic measures. If the patient with glioblastoma patient with poor prognosis result after operation can be divided into two groups with good and bad prognosis during preoperative diagnosis, a doctor can be guided to recommend the patient to make a follow-up diagnosis in time, and the patient with confirmed relapse can be treated and intervened or different diagnosis and treatment schemes are planned to avoid missing the optimal treatment opportunity. Therefore, it is of great significance to plan different treatment and management schemes for patients with poor prognosis effect predicted before operation.
Currently, the commonly used glioblastoma prognosis analysis method basically depends on only clinical indexes (such as age, KPS score value and the like) or a quantitative analysis method of some MR images (such as a traditional feature extraction method plus machine learning, a deep learning method and the like) is adopted. In the image quantitative analysis method, feature extraction is often directly performed on the MR image which is only subjected to conventional preprocessing, and if overlapped features can be decomposed by further processing operation on the input image data, difficulty in feature extraction of a subsequent model can be reduced, and prediction performance is improved. In terms of the method, deep learning and traditional machine learning methods have advantages. For the fact that the sample size of medical images is limited, overfitting phenomena of models are caused by high-dimensional depth features extracted from small sample data sets by deep learning, and therefore the models are not strong in generalization ability and low in prediction accuracy. However, the combination of feature engineering and conventional machine learning is difficult to achieve ideal prediction effects because the relationship may not be correctly characterized by the shallow features extracted by manual design.
Therefore, the inventor designs a CNN-XGboost-based glioblastoma prognosis prediction method. Firstly, the original MR image of a patient is processed to extract a tumor characteristic map as input image data besides the conventional image preprocessing operation. In addition, a deep convolutional neural network is designed and combined with a traditional machine learning method to exert respective advantages, so that an overfitting phenomenon is relieved, and the prediction performance of the model is improved.
Disclosure of Invention
The invention aims to provide a CNN-XGboost-based glioblastoma prognosis prediction method, which designs a method for fitting an intensity histogram of an MR image before a patient operation based on a double-Gaussian model and extracting a tumor feature map so as to decompose overlapped features and reduce the difficulty of feature extraction of a subsequent network model. Meanwhile, a deep convolutional neural network is designed, and the network comprises an upstream branch and a downstream branch. The upstream branch is a simple feature extraction module which mainly focuses on extracting simple features of the segmented label image. The downstream branch is a depth feature extraction network of the tumor feature map, which mainly focuses on the tumor feature map obtained after preprocessing. In addition, XGboost in traditional machine learning is combined, and the characteristics of the XGboost are screened and integrated, so that the overfitting risk is reduced, and the accuracy of prognosis prediction is improved.
Therefore, the method can realize individual prediction of poor prognosis of the glioblastoma multiforme, is helpful for guiding the patient to make a re-diagnosis in time, and intervenes or plans a new treatment scheme for the patient with confirmed relapse so as to avoid missing the optimal treatment opportunity.
In order to achieve the purpose, the invention provides a CNN-XGboost-based glioblastoma multiforme prognosis prediction method, which comprises the following steps:
s1: brain MR image basis preprocessing
And extracting a region of interest (ROI) from the mask of the obtained 2D image according to the existing segmentation label, and then carrying out standardization and normalization operations.
S2: tumor feature map extraction based on mixed double-Gaussian model
When image data processing is carried out, the brain MR image is formed by superposition of various features, and a mixed double-Gaussian model is one of mixed distribution models. Therefore, the characteristics of the mixed Gaussian distribution model can be separated by fitting the mixed Gaussian distribution model, and the mixed Gaussian distribution model is called a tumor characteristic map, so that the subsequent model learning is facilitated.
And aggregating the MR image intensity histogram of the glioblastoma into 2 classes by using double-Gaussian fitting, selecting the intermediate value of the 2 classes as an optimal threshold, and dividing the MR image of the glioblastoma into a tumor feature map consisting of a high-intensity subgraph and a low-intensity subgraph by using the threshold.
S3: data enhancement
To alleviate the overfitting phenomenon, the input ROI mask image, the high and low intensity tumor feature maps mentioned in step S2, and the corresponding segmentation label pictures are subjected to online data enhancement simultaneously. The method specifically comprises random vertical turnover, random horizontal turnover and random rotation.
S4: deep learning feature extraction
Constructing a tumor feature map (comprising high and low intensity subgraphs), a corresponding ROI segmentation label image and clinical information (KPS score value, age) of a patient, wherein the tumor feature map is input in the step S2, and a backbone network of the tumor feature map is used for extracting depth features of a 2D depth convolution neural network of Resnet 18;
the deep learning network is composed of two parallel branch feature extraction modules, and the upstream branch is a simple feature extraction module which mainly focuses on extracting simple features of the segmentation label images. The downstream branch is a depth feature extraction network of the tumor feature map, which mainly focuses on the tumor feature map obtained after preprocessing. The simple feature extraction module is a shallow CNN network which comprises a plurality of stacked 3 x 3 convolutional layers and a batch normalization layer, and the deep learning features are compressed into one-dimensional vectors which are recorded as L by using a global pooling layer in the last layer of the convolutional neural networkROI. And a main network of the depth feature extraction module of the downstream module is a Resnet series, and in order to relieve the overfitting problem by considering the characteristics of the medical data small sample, a Resnet18 network with ImageNet pre-training weight is selected as the main network of the downstream branch. The input tumor feature map image is considered to be a 2-channel image containing high and low intensities, so the number of input channels of the first layer convolution of the network is set to 2 here.
S5: establishment of prognosis model based on traditional mechanistic XGboost feature classifier
And performing fitting classification on the high-dimensional features extracted in the step of S4 by using XGboost.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Table 1 is the shallow convolutional neural network structure in the upstream branch of the present invention;
FIG. 1 is a flow chart of the operation of the present invention
FIG. 2 is a schematic diagram of the brain MR image tumor feature map extraction based on the double Gaussian mixture model
FIG. 3 is a model diagram of prognosis prediction model of glioblastoma based on CNN-XGboost;
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention comprises the following steps:
s1: brain MR image basis preprocessing
To perform a double-gaussian fit of the intensity distribution histogram of the subsequent MR images, the region of interest is first extracted from the 2D brain MR images. And masking the obtained 2D image according to the existing segmentation label to extract a region of interest (ROI), and then carrying out normalization and standardization operations.
In the formula xiPixels representing 2D images obtained from T2 sequence slicesPoint value, max (x)except_background),min (xexcept_background) Respectively representing the maximum and minimum values of pixels in the ROI region of the 2D image.
Wherein xnormTo obtain a 2D image matrix after normalization, muexcept_backgroundIs the mean value, σ, of the region of interest of the normalized 2D imageexcept_backgroundIs the standard deviation of the region of interest of the normalized 2D image.
S2: tumor feature map extraction based on mixed double-Gaussian model
When image data processing is performed, it is considered that the brain MR image is formed by superposition of various features, and a mixed double gaussian distribution is one of mixed distribution models. Therefore, the characteristics of the mixed Gaussian distribution model can be separated and called a tumor characteristic map by fitting the mixed Gaussian distribution model, so that the subsequent model learning is facilitated.
f(x)=G1(x)+G2(x)+n
Wherein,
after the pre-operation MR image of the glioblastoma patient is processed as described in step S1, it is assumed that the intensity distribution histogram follows a double gaussian distribution. Where f (x) is expressed as a double Gaussian fitting function of the intensity histogram, G1(x),G2(x) And n is noise, and is a distribution function of high and low intensity subgraphs to be extracted.
And aggregating the MR image intensity histogram of the glioblastoma into 2 classes by using double-Gaussian fitting, selecting the intermediate value of the 2 classes as an optimal threshold, and then extracting the MR image of the glioblastoma by using the threshold to obtain a tumor characteristic graph comprising a high-intensity subgraph and a low-intensity subgraph.
S3: data enhancement
On-line data enhancement is performed on the input ROI mask image, the high-low intensity tumor feature map mentioned in step S2.
In a preferred embodiment, the S3 specifically includes the following steps:
randomly and vertically overturning with the probability of 0.8;
randomly and horizontally turning, wherein the probability is 0.8;
randomly rotated at an angle between-20 ° and 20 °.
S4: deep learning feature extraction
Constructing a tumor feature map (comprising high-intensity subgraphs and low-intensity subgraphs), a corresponding ROI segmentation label image and clinical information (kps score value, age) of a patient, wherein the tumor feature map is input in the step S2, and a backbone network is used for extracting depth features of a 2D convolutional neural network of Resnet;
the deep learning network is composed of two parallel branch feature extraction modules, and the upstream branch is a simple feature extraction module which mainly focuses on extracting simple features of the segmentation label images. The downstream branch is a depth feature extraction network of the tumor feature map, which mainly focuses on the tumor feature map obtained after preprocessing. The simple feature extraction module is a shallow CNN network which comprises a plurality of stacked 3 x 3 convolutional layers and a batch normalization layer, and the deep learning features are compressed into one-dimensional vectors which are recorded as L by using a global pooling layer in the last layer of the convolutional neural networkROI. And a main network of the depth feature extraction module of the downstream module is a Resnet series, and a Resnet18 network with ImageNet pre-training weight is selected as a main network of a downstream path in order to relieve the overfitting problem by considering the characteristics of the medical data small sample. Considering that the input tumor feature map image is a 2-channel image containing high and low intensities, the number of input channels of the first layer convolution of the network is set to 2.
In a preferred embodiment, the model is predicted by using a cross-entropy loss function during model trainingPerformance measurement, CE (x)i) Represents a sample xiThe corresponding cross entropy loss function is shown as the formula:
wherein y isiRepresents a sample xiTrue patient prognosis, 0 indicates that the patient has a poor prognosis, and 1 indicates that the patient has no poor prognosis; y isiSample x representing model predictioniThe value of (3) for the prognosis is 0 or 1.
Lfusion=F(λ1LROI+λ2Lspatialmap)+(λ3Lkps+λ4Lage)
F is a full connection layer for feature dimension reduction, and is subjected to feature fusion with information such as input age, KPS score value and the like after dimension reduction, wherein the weight parameter is lambda1=1,λ2=1,λ3=0.8,λ40.8. And finally, optimizing the loss function shown by the formula by using an Adagrad algorithm until the model converges.
S5: establishment of prognosis model based on traditional mechanistic XGboost feature classifier
And (4) removing the last full connection layer from the deep convolution network designed in the step S4 to obtain characteristics, and performing classification fitting by using XGboost (eXtreme Gradient Boosting). XGboost is a tree integration model, there are a total of k trees, and the minimization objective function Obj, Obj is defined as follows:
wherein,is the predicted value of the ith sample, Ω (f)k) Representing the kth tree fkΩ is the formula notation of the computational complexity.
Compared with the prior art, the detection method for predicting the glioblastoma multiforme based on the CNN and the XGboost has the advantages of being simple in structure, low in cost and capable of achieving accurate detection. Based on a tumor feature map, a segmentation label image and clinical information of a patient as a part of an input image, depth features of the input image are extracted based on CNN, data enhancement is performed in the process, then the extracted depth features are classified and fitted by using XGboost, and compared with the method of directly performing feature engineering on an original MR image and performing feature learning by using other deep learning methods, the method has the advantage that performances such as the accuracy of glioblastoma prognosis detection are optimized.
Claims (5)
1. A CNN-XGboost-based glioblastoma multiforme prognosis prediction method is characterized by comprising the following steps:
s1: performing brain MR image basic preprocessing;
and extracting a region of interest (ROI) from the mask of the obtained 2D image according to the existing segmentation label, and then carrying out standardization and normalization operations.
S2: extracting a tumor feature map based on a mixed double-Gaussian model;
and aggregating the MR image intensity histogram of the glioblastoma into 2 classes by using double-Gaussian fitting, selecting the intermediate value of the 2 classes as an optimal threshold, and dividing the MR image of the glioblastoma into a tumor feature map consisting of a high-intensity subgraph and a low-intensity subgraph by using the threshold.
S3: data enhancement
The simultaneous online data enhancement of the input ROI mask image, the high and low intensity tumor feature maps mentioned in step S2, and the corresponding segmentation label pictures is performed. The method specifically comprises random vertical turnover, random horizontal turnover and random rotation.
S4: deep learning feature extraction
Constructing a tumor feature map (comprising high and low intensity subgraphs), a corresponding ROI segmentation label image and clinical information (KPS score value, age) of a patient, wherein the tumor feature map is input in the step S2, and a backbone network of the tumor feature map is used for extracting depth features of a 2D depth convolution neural network of Resnet 18;
s5: establishment of prognosis model based on traditional mechanistic XGboost feature classifier
And performing fitting classification on the high-dimensional features extracted in the step of S4 by using XGboost.
2. The brain MR image basis preprocessing method according to claim 1, characterized in that:
and (3) performing normalization and standardization operation on a region of interest (ROI) obtained by performing mask extraction according to the existing segmentation label. The normalization and normalization formulas are respectively as follows:
in the formula xiPixel point values, max (x), representing 2D images obtained from a T2 sequence sliceexcept_background),min(xexcept_background) Respectively representing the maximum and minimum values of pixels within the ROI region of the 2D image.
Wherein xnormTo obtain a 2D image matrix after normalization, muexcept_backgroundIs the mean value, σ, of the region of interest of the normalized 2D imageexcept_backgroundIs the standard deviation of the region of interest of the normalized 2D image.
3. The hybrid double-gaussian model-based tumor feature map extraction method according to claim 1, wherein the S2 specifically comprises the following steps:
when image data processing is performed, it is considered that the brain MR image is formed by superposition of various features, and a mixed double gaussian distribution is one of mixed distribution models. Therefore, the characteristics of the mixed Gaussian distribution model can be separated and called a tumor characteristic map by fitting the mixed Gaussian distribution model, so that the subsequent model learning is facilitated.
f(x)=G1(x)+G2(x)+n
Wherein,
after the pre-operation MR image of the glioblastoma patient is processed as described in step S1, it is assumed that the intensity distribution histogram follows a double gaussian distribution. Where f (x) is expressed as a double Gaussian fitting function of the intensity histogram, G1(x),G2(x) And n is noise, and is a distribution function of high and low intensity subgraphs to be extracted.
And aggregating the MR image intensity histogram of the glioblastoma into 2 classes by using double-Gaussian fitting, selecting the intermediate value of the 2 classes as an optimal threshold, and then extracting the MR image of the glioblastoma by using the threshold to obtain a tumor characteristic graph comprising a high-intensity subgraph and a low-intensity subgraph.
4. The data enhancement method of claim, wherein: in S3, the online data enhancement method includes: random vertical flipping, random horizontal flipping, random rotation.
5. The method for extracting deep learning features according to the claim, wherein the step S4 specifically includes the steps of:
the input of the deep learning feature extraction network is the tumor feature map (comprising high and low intensity subgraphs), the corresponding ROI segmentation label image and the clinical information (kps score value, age) of the patient mentioned in the step S2, and the backbone network extracts the deep features for the 2D convolutional neural network of Resnet;
deep learning feature extraction network feature extraction from two parallel branchesThe module is taken to constitute, and the upper branch is a simple feature extraction module which mainly focuses on extracting simple features of the segmentation label image. The downstream branch is a depth feature extraction network of the tumor feature map, which mainly focuses on the tumor feature map obtained after preprocessing. The simple feature extraction module is a shallow CNN network which comprises a plurality of stacked 3 x 3 convolutional layers and a batch normalization layer, and the deep learning features are compressed into one-dimensional vectors which are recorded as L by using a global pooling layer in the last layer of the convolutional neural networkROI. And a main network of the depth feature extraction module of the downstream module is a Resnet series, and a Resnet18 network with ImageNet pre-training weight is selected as a main network of a downstream path in order to relieve the overfitting problem by considering the characteristics of the medical data small sample. Considering that the input tumor feature map image is a 2-channel image containing high and low intensities, the number of input channels of the first layer convolution of the network is set to 2.
In model training, the prediction performance of the model is measured by using a cross entropy loss function, CE (x)i) Represents a sample xiThe corresponding cross entropy loss function is shown as the formula:
wherein y isiRepresents a sample xiTrue patient prognosis, 0 indicates that the patient has a poor prognosis, and 1 indicates that the patient has no poor prognosis; y isiSample x representing model predictioniThe value of (3) for the prognosis is 0 or 1.
Lfusion=F(λ1LROI+λ2Lspatialmap)+(λ3Lkps+λ4Lage)
F is a full connection layer for feature dimension reduction, and is subjected to feature fusion with information such as input age, KPS score value and the like after dimension reduction, wherein the weight parameter is lambda1=1,λ2=1,λ3=0.8,λ40.8. Finally using Adagrad algorithmThe loss function shown in the formula is optimized until the model converges.
The method of claim, wherein the prognostic model is established based on a conventional mechanistic XGboost feature classifier: and performing XGboost classification fitting on the features obtained by removing the last full connection layer by using the deep convolutional neural network designed in the step S4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111409937.0A CN114283938A (en) | 2021-11-25 | 2021-11-25 | CNN-XGboost-based glioblastoma prognosis prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111409937.0A CN114283938A (en) | 2021-11-25 | 2021-11-25 | CNN-XGboost-based glioblastoma prognosis prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114283938A true CN114283938A (en) | 2022-04-05 |
Family
ID=80869837
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111409937.0A Pending CN114283938A (en) | 2021-11-25 | 2021-11-25 | CNN-XGboost-based glioblastoma prognosis prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114283938A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114795258A (en) * | 2022-04-18 | 2022-07-29 | 浙江大学 | Child hip joint dysplasia diagnosis system |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106683081A (en) * | 2016-12-17 | 2017-05-17 | 复旦大学 | Brain glioma molecular marker nondestructive prediction method and prediction system based on radiomics |
US20190156159A1 (en) * | 2017-11-20 | 2019-05-23 | Kavya Venkata Kota Sai KOPPARAPU | System and method for automatic assessment of cancer |
US20200272864A1 (en) * | 2017-11-06 | 2020-08-27 | University Health Network | Platform, device and process for annotation and classification of tissue specimens using convolutional neural network |
CN112184658A (en) * | 2020-09-24 | 2021-01-05 | 上海健康医学院 | Method, medium, and electronic device for predicting non-small cell lung cancer prognostic survival |
-
2021
- 2021-11-25 CN CN202111409937.0A patent/CN114283938A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106683081A (en) * | 2016-12-17 | 2017-05-17 | 复旦大学 | Brain glioma molecular marker nondestructive prediction method and prediction system based on radiomics |
US20200272864A1 (en) * | 2017-11-06 | 2020-08-27 | University Health Network | Platform, device and process for annotation and classification of tissue specimens using convolutional neural network |
US20190156159A1 (en) * | 2017-11-20 | 2019-05-23 | Kavya Venkata Kota Sai KOPPARAPU | System and method for automatic assessment of cancer |
CN112184658A (en) * | 2020-09-24 | 2021-01-05 | 上海健康医学院 | Method, medium, and electronic device for predicting non-small cell lung cancer prognostic survival |
Non-Patent Citations (3)
Title |
---|
刘云鹏;刘光品;王仁芳;金冉;孙德超;邱虹;董晨;李瑾;洪国斌;: "深度学习结合影像组学的肝脏肿瘤CT分割", 中国图象图形学报, no. 10, 16 October 2020 (2020-10-16) * |
胡光亮;王艳;罗勇;周激流;: "基于卷积神经网络的鼻咽肿瘤MR图像分割", 计算机应用, no. 1, 30 June 2018 (2018-06-30) * |
郭畅宇: "基于深度学习的高级别脑胶质瘤预后算法研究", 医药卫生科技辑, no. 03, 15 March 2024 (2024-03-15), pages 070 - 359 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114795258A (en) * | 2022-04-18 | 2022-07-29 | 浙江大学 | Child hip joint dysplasia diagnosis system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111414862B (en) | Expression recognition method based on neural network fusion key point angle change | |
CN106203395B (en) | Face attribute recognition method based on multitask deep learning | |
CN111860171B (en) | Method and system for detecting irregular-shaped target in large-scale remote sensing image | |
CN110751098B (en) | Face recognition method for generating confrontation network based on illumination and posture | |
Wang et al. | Weakly supervised learning for whole slide lung cancer image classification | |
CN108898175A (en) | Area of computer aided model building method based on deep learning gastric cancer pathological section | |
CN107480702B (en) | Feature selection and feature fusion method for HCC pathological image recognition | |
CN108776774A (en) | A kind of human facial expression recognition method based on complexity categorization of perception algorithm | |
CN109815967A (en) | CNN ship seakeeping system and method based on Fusion Features | |
CN105069774B (en) | The Target Segmentation method of optimization is cut based on multi-instance learning and figure | |
WO2022127500A1 (en) | Multiple neural networks-based mri image segmentation method and apparatus, and device | |
CN101609485B (en) | Medical image diagnosing system and diagnosing method based on migrated nuclear matching tracing | |
CN110390282A (en) | A kind of finger vein identification method and system based on the loss of cosine center | |
Gehlot et al. | Ednfc-net: Convolutional neural network with nested feature concatenation for nuclei-instance segmentation | |
CN115409846A (en) | Colorectal cancer focus region lightweight segmentation method based on deep learning | |
CN112149689B (en) | Unsupervised domain adaptation method and system based on target domain self-supervised learning | |
CN115984937A (en) | Face expression recognition method based on optimized ResNet algorithm | |
CN114283938A (en) | CNN-XGboost-based glioblastoma prognosis prediction method | |
Wang et al. | CWC-transformer: a visual transformer approach for compressed whole slide image classification | |
CN111310820A (en) | Foundation meteorological cloud chart classification method based on cross validation depth CNN feature integration | |
CN116798093A (en) | Two-stage facial expression recognition method based on course learning and label smoothing | |
CN116912253B (en) | Lung cancer pathological image classification method based on multi-scale mixed neural network | |
吴雨林 et al. | Face recognition system based on CNN and LBP features for classifier optimization and fusion | |
CN117670895A (en) | Immunohistochemical pathological image cell segmentation method based on section re-staining technology | |
CN111127407B (en) | Fourier transform-based style migration forged image detection device and method |
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 |