CN112508850B - Deep learning-based method for detecting malignant area of thyroid cell pathological section - Google Patents
Deep learning-based method for detecting malignant area of thyroid cell pathological section Download PDFInfo
- Publication number
- CN112508850B CN112508850B CN202011243793.1A CN202011243793A CN112508850B CN 112508850 B CN112508850 B CN 112508850B CN 202011243793 A CN202011243793 A CN 202011243793A CN 112508850 B CN112508850 B CN 112508850B
- Authority
- CN
- China
- Prior art keywords
- model
- pathological section
- training
- pathological
- image
- 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
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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
- G06T2207/10061—Microscopic image from scanning electron microscope
-
- 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
- G06T2207/20024—Filtering details
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Abstract
The invention discloses a method for detecting a malignant area of a thyroid cell pathological section based on deep learning, which mainly comprises the following steps: carrying out pathological section on the thyroid cell; digitally processing the image of the pathological section on a microscope, and smearing the pathological section with different staining agents to obtain a colorful pathological section; cutting the complete pathological section into a cutting block with a proper size as the input of a deep neural network model; screening out invalid sections of the pathological section; classifying the benign and malignant pathological sections after the primary screening of the blocks by adopting a weak supervision learning method; constructing a machine learning method based on a random forest by using a false positive removing scheme to remove false positive from the prediction result of benign and malignant classification; therefore, the accuracy of detection can be further improved. And a high-risk area display step of pathological sections: normalizing the probability of the malignancy prediction of each block and mapping the normalized probability into an original graph to generate a thermodynamic diagram; and the pathological doctor can visually display the information more intuitively.
Description
Technical Field
The invention relates to the field of medical image processing, in particular to a method for detecting a malignant region of a thyroid cell pathological section based on deep learning.
Background
The pathological section is used as a gold standard for pathological diagnosis and plays an important role in clinic and scientific research. The thyroid cell pathological screening is in a popularizing stage, and the pathological screening method of cell puncture has the advantages of small trauma, low risk and quick diagnosis, and is popularized in a plurality of hospitals. However, the cellular pathology screening is late, so that pathologists are relatively few and experienced doctors are scarce, and the overstock of cases needing diagnosis is caused. Meanwhile, in the diagnosis of the cytopathology section, the target cells are not obtained or are too few, so that the diagnosis cannot be performed.
Based on the above, the deep learning method is provided, a high-risk malignant cell area is pre-extracted on a pathological section, a doctor is assisted to diagnose and make a decision, and through the process, the workload of the pathological doctor is greatly reduced, the misdiagnosis rate is also reduced, the working efficiency of the pathological doctor is improved, and the pathological doctor can concentrate on more complex cases.
The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.
Disclosure of Invention
The invention aims to provide a method for detecting a malignant area of a thyroid cell pathological section based on deep learning, which adopts a method of multi-stage training from coarse to fine, firstly eliminates obviously invalid blocks, classifies the valid blocks into benign and malignant areas, and further improves the detection accuracy by removing false positive.
In order to achieve the aim, the invention provides a thyroid cell pathological section malignant area detection method based on deep learning, which comprises the following steps of a section step, a pretreatment step, an image sampling step, a section preliminary screening step, a benign and malignant classification step, a suspicious area post-treatment step and a pathological section high-risk area display step; slicing: carrying out pathological section on the thyroid cell; a pretreatment step: digitally processing the image of the pathological section on a microscope, and smearing the pathological section with different staining agents to obtain a colorful pathological section; an image sampling step: cutting the complete pathological section into a cutting block with a proper size as the input of a deep neural network model; a dicing preliminary screening step: screening out invalid sections of the pathological section; and (3) classifying benign and malignant diseases: classifying the benign and malignant pathological sections after the primary screening of the blocks by adopting a weak supervision learning method; post-processing the suspicious region: constructing a machine learning method based on a random forest by using a false positive removing scheme to remove false positive from the prediction result of benign and malignant classification; and a high-risk area display step of pathological sections: the probability of malignancy prediction for each tile is normalized and mapped into the original image, the grayscale image is converted to a color image and a thermodynamic image is generated.
In a preferred embodiment, the slicing step includes converting the color pathological section image from RGB space to gray space and smoothing the image by gaussian filtering.
In a preferred embodiment, the image sampling step cuts the complete pathological section into a cut block with a proper size by using an overlap sampling method based on an overlap area of 50%, and the resolution of the cut block is 512x512 pixels.
In a preferred embodiment, the sections that are not valid for the section preliminary screening step are sections that do not have the required resolution or do not contain sufficient thyroid cells.
In a preferred embodiment, in the preliminary screening step of the cut blocks, the initial weight obtained by training image network with lightweight resnet18 classification network is used as the initial weight to perform the preliminary screening of the cut blocks, and the cut blocks with the prediction probability lower than the threshold value at resnet18 are screened out.
In a preferred embodiment, the step of classifying the benign and malignant diseases comprises a learning method adopting weak supervision, and fully utilizing unlabeled data to participate in training, wherein the specific training process comprises a first-stage training and a second-stage training; the first stage of training comprises: on 140 slices diagnosed as papillary thyroid carcinoma, 1203 malignant PTC cut-off images were marked by a thyroid cell pathologist as positive classes, and 1203 normal cut-off images were randomly cut off on 120 normal slices as negative classes; according to the following steps of 4: 1: 1, distributing a training set test set and a verification set in proportion, participating in the training of the model 1, and selecting resnet18 to construct the model 1; performing iterative training on the model, and updating the weight of the model to obtain the model with optimal performance on the test set; predicting the blocks of the cell pathological section of the thyroid without the label by using a model with optimal performance, and setting the labels of the blocks with the prediction probability greater than a given threshold value as positive case labels; setting the block label with the prediction probability smaller than a given threshold value as a negative example label; adding the marked data of the positive example label and the negative example label into a training set; the second stage of training comprises: selecting a resnet34 model as a second-stage training model 2, performing iterative training on the model 2, updating the weight to obtain an optimal model of the model 2 on a test set, setting a pseudo label for a label-free cut block by adopting the same method as that of the first stage, and expanding the training set; the first-stage training and the second-stage training respectively build model 3, model 4 and model 5 gradually expanding data by resnet50, resnet101 and resnet152 so as to further improve the performance of the models; wherein, the accuracy of model 1, model 2, model 3, model 4 and model 5 on the verification set is 0.78, 0.80, 0.83, 0.87 and 0.95 respectively.
In a preferred embodiment, in the prediction stages of model 1 to model 5, the overlapped regions are removed by using a maximum suppression technique, and of the intersected blocks, the block with the highest probability of malignancy is selected as an output.
In a preferred embodiment, the model input for false positive removal in the post-processing step of the suspicious region comprises: the probability of a cut predicted by model 5 of the image module; model 5 features extracted from the penultimate layer; extracting image characteristics of the pathological section of the thyroid cell based on the local binary pattern; and coordinate information of the center point of the cut piece in the image.
In a preferred embodiment, the probability of malignancy prediction for each slice in the high-risk region display step of the pathological section is normalized to 0-255.
In order to achieve the above object, the present invention provides a deep learning-based detection system for malignant regions of thyroid cell pathological sections, which can be used in the detection method, wherein the detection system comprises an apparatus part, a hardware part and a software part; the equipment part mainly comprises a microscope and the like for providing an image source; the hardware part mainly comprises a GPU, a CPU and the like; the software part mainly comprises Keras, Pythrch, tensorflow, Caffe or Paddle and the like.
Compared with the prior art, the method for detecting the malignant area of the thyroid cell pathological section based on deep learning has the following beneficial effects: the method adopts a multi-stage training method from coarse to fine, obviously invalid blocks are removed, the effective blocks are classified into benign and malignant blocks, and the detection accuracy is further improved by removing false positive. The method has the advantages that the high-risk malignant cell area is pre-extracted on the pathological section, a doctor is assisted to diagnose and make a decision, and through the process, the workload of the pathological doctor is greatly reduced, the misdiagnosis rate is reduced, the working efficiency of the pathological doctor is improved, and the pathological doctor can concentrate on more complex cases.
Drawings
FIG. 1 is a flow chart of a detection method according to an embodiment of the invention.
Detailed Description
The following detailed description of the present invention is provided in conjunction with the accompanying drawings, but it should be understood that the scope of the present invention is not limited to the specific embodiments.
Throughout the specification and claims, unless explicitly stated otherwise, the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated element or component but not the exclusion of any other element or component.
As shown in fig. 1, the method for detecting a malignant region of a thyroid cell pathological section based on deep learning according to a preferred embodiment of the present invention mainly includes the following steps of sectioning, preprocessing, image sampling, preliminary screening of a section, classifying benign and malignant regions, post-processing of a suspicious region, displaying a high-risk region of a pathological section, and the like.
1: pretreatment of
The purpose of preprocessing is to strengthen the characteristics of the image, the image of the pathological section on the microscope is smeared by different staining agents after being processed digitally, so that a colorful pathological section is obtained, and the doctor can observe and diagnose conveniently. However, different staining methods and operator procedures can lead to different staining differences of the different sections, which can interfere with diagnostic analysis using machine learning and deep learning methods. In order to reduce the influence, in the image preprocessing stage, the image is firstly converted from the RGB space to the gray scale space; meanwhile, due to the fact that stains and the like exist on the glass slide, noise is introduced to the cell pathological section, and therefore the image is smoothed by the Gaussian filtering method, and the influence of noise is reduced.
2: image sampling
The complete thyroid cell pathological section is generally very large, and huge memory is consumed and the cost is very high when the complete section is sent to a deep neural network for training. Therefore, the present embodiment adopts an overlap sampling method based on an overlap area of 0.5 (50% overlap) (which can ensure that all possible positions are collected and at least one suspicious region can be completely presented), and cuts out a block with a suitable size (in the present embodiment, a cut block with a resolution of 512 × 512 is selected) as an input of the deep neural network model.
3: preliminary screening of cut pieces
In the cytopathological section of the thyroid, the resolution of some areas can not meet the requirement; some regions do not contain enough thyroid cells, which is not significant for diagnosis, and the cut blocks are easy to screen without professional doctors, so the embodiment trains a lightweight classification network of resnet18 (a specific model training method: manually labeling 1000 blocks with insufficient resolution and insufficient thyroid cells as negative classes, a block with clear visual field and thyroid cells in the visual field satisfying the next workflow is taken as a positive example, migration training is performed based on the initial weight obtained by training resnet18 in imagenet as an initial weight, a model with the best effect on the verification set is selected as a preliminary screening model of the cut blocks, in the prediction stage, the cut blocks with the prediction probability of resnet18 lower than a threshold value (0.2 set in the embodiment) are removed), part of invalid cut blocks are screened, and the rest cut blocks are sent to a good and malignant classification module, further detection is performed.
4: classification of benign and malignant disease
The benign and malignant classification module is a core part of the embodiment, can screen thyroid malignant cell masses which are focused by doctors, and utilizes the pathological section high-risk area display module to draw a thermodynamic diagram of malignant cell clusters, so that diagnosis by pathological doctors is facilitated.
Because pathological sections of thyroid cells are large and experienced pathologists are few, marking malignant cell masses on pathological sections is a very time-consuming matter. On the basis that the unlabeled thyroid cell pathological section is easy to obtain, in the benign and malignant classification module, a weak supervision learning method is adopted in the embodiment, data which are not labeled are fully utilized to participate in training, and the specific process is as follows:
1) on 140 sections diagnosed with Papillary Thyroid Carcinoma (PTC), 1203 malignant PTC section images were marked as positive classes by the thyroid cytopathologist, and 1203 sections were randomly cut out on 120 normal sections as negative classes. According to the following steps of 4: 1: the proportion of 1 assigns a training set test set and a validation set, and participates in the training of the Model (Model)1 (the Model (Model)1 is selected as Resnet 18). And (5) carrying out iterative training on the model, and updating the weight of the model to obtain the model with the optimal performance on the test set. The model predicts the blocks of the cytopathological section of the unlabeled thyroid, and the label of the block whose prediction probability is greater than a given threshold is set as the positive case label (in this embodiment, the threshold is set to 0.9). The slice label with the prediction probability smaller than a given threshold is set as a negative example label. These labeled data are added to the training set. (in the present embodiment, the number of positive and negative examples is extended by 0.25 times the original data amount).
2) As the training set is increased, in this embodiment, the resnet34 Model is selected as the second-stage training Model (Model)2, iterative training is performed on the Model (Model)2, and the weight is updated, so that the optimal Model of the Model (Model)2 on the test set is obtained. And setting a pseudo label for the label-free cut block by using the same method as the first stage, and expanding the training set.
3) Similar to the method in the first step and the second step, the Model (Model)3, the Model (Model)4 and the Model (Model)5 are respectively constructed by the resnet50, the resnet101 and the resnet152, so that the performance of the Model is further improved.
4) The accuracy rates of the Model (Model)1, the Model (Model)2, the Model (Model)3, the Model (Model)4 and the Model (Model)5 on the verification set are respectively 0.78, 0.80, 0.83, 0.87 and 0.95. The weak supervised learning method can effectively improve the classification performance of the benign and malignant cells.
In the prediction phase of models (Model)1 to (Model)5, a very large suppression technique (NMS) is used to eliminate the overlapping regions: and selecting the block with the highest malignancy probability from the intersected blocks as output.
5: post-processing of suspicious regions
In this module, this embodiment proposes a solution for removing false positives, that is, a random forest-based machine learning method is constructed to remove false positives from prediction results of a Model (Model)5 for good and malignant classification, where inputs of the Model are from four parts: the probability of a cut predicted by a Model (Model)5 of the image module; model 5 reciprocal second layer extracted features; extracting image features of the pathological section of the thyroid cell based on a Local Binary Pattern (LBP); and coordinate information of the center point of the cut piece in the image (normalized to [0, 1 ]). And training the random forest model. The model eliminates the blocks with the probability lower than a certain threshold (the threshold of the embodiment is set to be 0.2), and further lightens the work of pathologists. This threshold is dynamically adjustable in this embodiment, and can be manually adjusted lower if the pathologist wants to learn more about the malignant cell pathology slices and make further evaluations.
6: high-risk area display of pathological section
Normalizing the probability of malignancy prediction of each cutting block to [0,255], mapping the probability into an original image, converting a gray scale image into a color image to generate a thermodynamic diagram, and presenting the thermodynamic diagram to a cytopathologist, and providing a visualization suggestion for the pathologist, particularly a doctor with less experience: which areas are of major concern. The method of the embodiment can transfer the experience learned from experienced pathologists to young doctors to help the young doctors to grow up.
The module 4 (classification step of benign and malignant regions) and the module 5 (post-processing step of suspicious regions) in the present embodiment may work in series or may work independently. When the doctor needs to display the result more quickly, suspicious malignant regions can be displayed and output in a high-to-low order according to the probability based on the threshold set by the doctor. If the doctor wants more accurate results, the doctor can select the results based on the threshold (0.5 is selected in the embodiment) and the prediction probability based on the module 5 is displayed from high to low.
The detection system based on the deep learning of the malignant area of the thyroid cell pathological section can be used for the detection method, and comprises an equipment part, a hardware part and a software part; the equipment part mainly comprises a microscope and the like for providing an image source; the hardware part mainly comprises a GPU, a CPU and the like; the software part mainly comprises Keras, Pythrch, tensorflow, Caffe or Paddle and the like.
In some embodiments, the present example development use is Keras: version 2.1.6, but the invention is not limited to this version. Meanwhile, the deep learning framework capable of being used in an extensible manner can be applied to the development system of the invention, but is not limited to, the pytore, tensorflow, Caffe or folder.
In some embodiments, the hardware developed by the GPU of this embodiment uses GeForce GTX 1080ti, but the present invention is not limited to this type of GPU. May be any stand-alone graphics card including, without limitation, GeForce RTX 1060, GeForce RTX 2080ti, GeForce P6000 (more graphics cards may view the english world web).
In some embodiments, the hardware developed by the CPU of this embodiment uses Intel (R) Xeon (R) CPU E5-2640 v4@2.40 GHz. More server CPUs refer to the CPU types of the servers of Inter and AMD.
In some embodiments, the pathological pictures of the present example use HE stained images, and other staining protocols include, but are not limited to, CK7, P40, CD56, hexamine silver, TTF1, PAS, and the like.
In conclusion, the detection method of the malignant area of the thyroid cell pathological section based on deep learning has the following beneficial effects: the method adopts a multi-stage training method from coarse to fine, obviously invalid blocks are removed, the effective blocks are classified into benign and malignant blocks, and the detection accuracy is further improved by removing false positive. The method has the advantages that the high-risk malignant cell area is pre-extracted on the pathological section, a doctor is assisted to diagnose and make a decision, and through the process, the workload of the pathological doctor is greatly reduced, the misdiagnosis rate is reduced, the working efficiency of the pathological doctor is improved, and the pathological doctor can concentrate on more complex cases.
The foregoing descriptions of specific exemplary embodiments of the present invention have been presented for purposes of illustration and description. It is not intended to limit the invention to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. The exemplary embodiments were chosen and described in order to explain certain principles of the invention and its practical application to enable one skilled in the art to make and use various exemplary embodiments of the invention and various alternatives and modifications as are suited to the particular use contemplated. It is intended that the scope of the invention be defined by the claims and their equivalents.
Claims (9)
1. A method for detecting a malignant area of a thyroid cell pathological section based on deep learning is characterized by comprising the following steps:
slicing: carrying out pathological section on the thyroid cell;
a pretreatment step: digitally processing the image of the pathological section on a microscope, and smearing the pathological section with different staining agents to obtain a colorful pathological section;
an image sampling step: cutting the complete pathological section into a cutting block with a proper size as the input of a deep neural network model;
a dicing preliminary screening step: screening out invalid sections of the pathological section;
and (3) classifying benign and malignant diseases: classifying the benign and malignant pathological sections after the preliminary screening of the blocks by adopting a weak supervision learning method, wherein the classifying step of the benign and malignant pathological sections comprises the step of adopting the weak supervision learning method, fully utilizing data which is not labeled to participate in training, and the specific training process comprises the following steps:
the first stage of training comprises:
on 140 slices diagnosed as papillary thyroid carcinoma, 1203 malignant PTC cut-off images were marked by a thyroid cell pathologist as positive classes, and 1203 normal cut-off images were randomly cut off on 120 normal slices as negative classes;
according to the following steps of 4: 1: 1, distributing a training set test set and a verification set in proportion, and participating in the training of the model 1, wherein the training method of the model 1 is selected as resnet 18;
performing iterative training on the model, and updating the weight of the model to obtain the model with optimal performance on the test set; predicting the blocks of the cell pathological section of the thyroid without the label by using the model with the optimal performance, and setting the labels of the blocks with the prediction probability greater than a given threshold value as positive case labels;
setting the block label with the prediction probability smaller than a given threshold value as a negative example label; and
adding the labeled data of the positive example label and the negative example label into a training set; and
the second stage of training comprises:
selecting a resnet34 model as a second-stage training model 2, performing iterative training on the model 2, updating the weight to obtain an optimal model of the model 2 on a test set, setting a pseudo label for a label-free cut block by adopting the same method as that of the first stage, and expanding the training set;
the first-stage training and the second-stage training respectively build model 3, model 4 and model 5 gradually expanding data by resnet50, resnet101 and resnet152 so as to further improve the performance of the models;
wherein, the accuracy of the model 1, the model 2, the model 3, the model 4 and the model 5 on the verification set is respectively 0.78, 0.80, 0.83, 0.87 and 0.95;
post-processing the suspicious region: constructing a machine learning method based on a random forest by using a false positive removing scheme to remove false positive from the prediction result of benign and malignant classification;
displaying the high-risk area of the pathological section: the probability of malignancy prediction for each tile is normalized and mapped into the original image, the grayscale image is converted to a color image and a thermodynamic image is generated.
2. The detection method as claimed in claim 1, wherein the slicing step includes converting the color pathological section image from RGB space to gray space and smoothing the image by gaussian filtering.
3. The detection method according to claim 1, wherein the image sampling step cuts out a cut piece with a proper size from the complete pathological section by using an overlap sampling method based on an overlap area of 50%, and the resolution of the cut piece is 512x512 pixels.
4. The assay of claim 1 wherein the sections that are not valid in the section preliminary screening step are areas that do not have the required resolution or do not contain sufficient thyroid cells.
5. The detection method as claimed in claim 4, wherein in the preliminary screening step of the cut blocks, the preliminary screening of the cut blocks is performed by using initial weights obtained by training a light-weighted classification network of resnet18 in imagenet as initial weights, and the cut blocks with the prediction probability lower than a threshold value in resnet18 are screened.
6. The detection method according to claim 1, wherein in the prediction stages of model 1 to model 5, the overlapped region is eliminated by using a maximum suppression technique, and of the intersected blocks, the block with the highest probability of malignancy is selected as an output.
7. The detection method according to claim 1, wherein the model input for false positive removal in the post-processing step of the suspicious region comprises:
the probability of a cut predicted by model 5 of the image module;
model 5 features extracted from the penultimate layer;
extracting image characteristics of the pathological section of the thyroid cell based on the local binary pattern; and
coordinate information of the center point of the cut piece in the image.
8. The detection method according to claim 1, wherein the probability of malignancy prediction of each slice in the step of displaying the high-risk region of the pathological section is normalized to 0-255.
9. A detection system of a malignant region of a thyroid cell pathological section based on deep learning, which is applied to the detection method according to any one of claims 1 to 8, characterized in that the detection system comprises an equipment part, a hardware part and a software part; the apparatus portion includes a microscope providing a source of images; the hardware part comprises a GPU and a CPU; the software portion includes Keras, Pythrch, tensorflow, Caffe or Paddle.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011243793.1A CN112508850B (en) | 2020-11-10 | 2020-11-10 | Deep learning-based method for detecting malignant area of thyroid cell pathological section |
| PCT/CN2021/091988 WO2022100034A1 (en) | 2020-11-10 | 2021-05-07 | Detection method for malignant region of thyroid cell pathological section based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011243793.1A CN112508850B (en) | 2020-11-10 | 2020-11-10 | Deep learning-based method for detecting malignant area of thyroid cell pathological section |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112508850A CN112508850A (en) | 2021-03-16 |
| CN112508850B true CN112508850B (en) | 2021-07-20 |
Family
ID=74955773
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011243793.1A Active CN112508850B (en) | 2020-11-10 | 2020-11-10 | Deep learning-based method for detecting malignant area of thyroid cell pathological section |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112508850B (en) |
| WO (1) | WO2022100034A1 (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112508850B (en) * | 2020-11-10 | 2021-07-20 | 广州柏视医疗科技有限公司 | Deep learning-based method for detecting malignant area of thyroid cell pathological section |
| CN113538435B (en) * | 2021-09-17 | 2022-02-18 | 北京航空航天大学 | Pancreatic cancer pathological image classification method and system based on deep learning |
| CN114743195B (en) * | 2022-04-13 | 2022-12-09 | 赛维森(广州)医疗科技服务有限公司 | Thyroid cell pathology digital image recognizer training method and image processing method |
| CN115222734B (en) * | 2022-09-20 | 2023-01-17 | 山东大学齐鲁医院 | Image analysis method and system for gastric mucosa enteroepithelization |
| CN115909006B (en) * | 2022-10-27 | 2024-01-19 | 武汉兰丁智能医学股份有限公司 | Mammary tissue image classification method and system based on convolution transducer |
| CN115661815B (en) * | 2022-12-07 | 2023-09-12 | 赛维森(广州)医疗科技服务有限公司 | Pathological image classification method and device based on global feature mapping |
| CN116453114B (en) * | 2022-12-14 | 2024-03-05 | 西南医科大学附属口腔医院 | Pathological image analysis method, equipment and system based on deep learning |
| CN115880262B (en) * | 2022-12-20 | 2023-09-05 | 桂林电子科技大学 | Weak supervision pathological image tissue segmentation method based on online noise suppression strategy |
| CN116448760B (en) * | 2023-03-21 | 2023-10-20 | 上海华维可控农业科技集团股份有限公司 | Agricultural intelligent monitoring system and method based on machine vision |
| CN116313115B (en) * | 2023-05-10 | 2023-08-15 | 浙江大学 | Drug action mechanism prediction method based on mitochondrial dynamic phenotype and deep learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7440601B1 (en) * | 2003-10-10 | 2008-10-21 | The United States Of America As Represented By The Department Of Health And Human Services | Automated identification of ileocecal valve |
| CN103854028A (en) * | 2008-12-22 | 2014-06-11 | 赛利恩影像股份有限公司 | Method and system of automated detection of lesions in medical images |
| CN109670510A (en) * | 2018-12-21 | 2019-04-23 | 万达信息股份有限公司 | A kind of gastroscopic biopsy pathological data screening system and method based on deep learning |
| CN110060245A (en) * | 2019-04-15 | 2019-07-26 | 青岛大学 | A kind of gastric cancer lymph node dyeing pathological image automatic identifying method based on deep neural network |
| CN110309860A (en) * | 2019-06-06 | 2019-10-08 | 昆明理工大学 | The method classified based on grade malignancy of the convolutional neural networks to Lung neoplasm |
| CN110838118A (en) * | 2019-09-24 | 2020-02-25 | 上海联影智能医疗科技有限公司 | System and method for anomaly detection in medical procedures |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3432263B1 (en) * | 2017-07-17 | 2020-09-16 | Siemens Healthcare GmbH | Semantic segmentation for cancer detection in digital breast tomosynthesis |
| CN108520518A (en) * | 2018-04-10 | 2018-09-11 | 复旦大学附属肿瘤医院 | A kind of thyroid tumors Ultrasound Image Recognition Method and its device |
| CN111612752A (en) * | 2020-05-15 | 2020-09-01 | 江苏省人民医院（南京医科大学第一附属医院） | Ultrasonic image thyroid nodule intelligent detection system based on fast-RCNN |
| CN112508850B (en) * | 2020-11-10 | 2021-07-20 | 广州柏视医疗科技有限公司 | Deep learning-based method for detecting malignant area of thyroid cell pathological section |
-
2020
- 2020-11-10 CN CN202011243793.1A patent/CN112508850B/en active Active
-
2021
- 2021-05-07 WO PCT/CN2021/091988 patent/WO2022100034A1/en active Application Filing
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7440601B1 (en) * | 2003-10-10 | 2008-10-21 | The United States Of America As Represented By The Department Of Health And Human Services | Automated identification of ileocecal valve |
| CN103854028A (en) * | 2008-12-22 | 2014-06-11 | 赛利恩影像股份有限公司 | Method and system of automated detection of lesions in medical images |
| CN109670510A (en) * | 2018-12-21 | 2019-04-23 | 万达信息股份有限公司 | A kind of gastroscopic biopsy pathological data screening system and method based on deep learning |
| CN110060245A (en) * | 2019-04-15 | 2019-07-26 | 青岛大学 | A kind of gastric cancer lymph node dyeing pathological image automatic identifying method based on deep neural network |
| CN111145176A (en) * | 2019-04-15 | 2020-05-12 | 青岛大学 | Method and system for automatically identifying lymph node staining pathological image of gastric cancer based on deep neural network |
| CN110309860A (en) * | 2019-06-06 | 2019-10-08 | 昆明理工大学 | The method classified based on grade malignancy of the convolutional neural networks to Lung neoplasm |
| CN110838118A (en) * | 2019-09-24 | 2020-02-25 | 上海联影智能医疗科技有限公司 | System and method for anomaly detection in medical procedures |
Non-Patent Citations (3)
| Title |
|---|
| An integrated approach for medical abnormality detection using deep patch convolutional neural networks;Pengcheng Xi et al.;《The Visual Computer》;20191209;1869-1882 * |
| Automatic diagnosis for thyroid nodules in ultrasound images by deep neural networks;Lituan Wang et al.;《Medical Image Analysis》;20200204;1-13 * |
| 联合多种超声检查方法在甲状腺结节良恶性鉴别中的应用;邓龙;《中国优秀硕士学位论文全文数据库 医药卫生科技辑》;20181015;E060-11 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112508850A (en) | 2021-03-16 |
| WO2022100034A1 (en) | 2022-05-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112508850B (en) | Deep learning-based method for detecting malignant area of thyroid cell pathological section | |
| CN108765408B (en) | Method for constructing cancer pathological image virtual disease case library and multi-scale cancer detection system based on convolutional neural network | |
| Landau et al. | Artificial intelligence in cytopathology: a review of the literature and overview of commercial landscape | |
| Vaickus et al. | Automating the Paris System for urine cytopathology—A hybrid deep‐learning and morphometric approach | |
| CN110245657B (en) | Pathological image similarity detection method and detection device | |
| CN111986150B (en) | The method comprises the following steps of: digital number pathological image Interactive annotation refining method | |
| Chang et al. | Nuclear segmentation in H&E sections via multi-reference graph cut (MRGC) | |
| CN106056595A (en) | Method for automatically identifying whether thyroid nodule is benign or malignant based on deep convolutional neural network | |
| CN111145209A (en) | Medical image segmentation method, device, equipment and storage medium | |
| CN101551854B (en) | A processing system of unbalanced medical image and processing method thereof | |
| CN111488921A (en) | Panoramic digital pathological image intelligent analysis system and method | |
| CN112365471B (en) | Cervical cancer cell intelligent detection method based on deep learning | |
| CN107492084B (en) | Typical clustering cell nucleus image synthesis method based on randomness | |
| CN107169497A (en) | A kind of tumor imaging label extracting method based on gene iconography | |
| Baykal et al. | Modern convolutional object detectors for nuclei detection on pleural effusion cytology images | |
| CN112990214A (en) | Medical image feature recognition prediction model | |
| Yang et al. | A generalized deep learning-based framework for assistance to the human malaria diagnosis from microscopic images | |
| CN111047559A (en) | Method for rapidly detecting abnormal area of digital pathological section | |
| CN113222903A (en) | Full-section histopathology image analysis method and system | |
| CN116468690B (en) | Subtype analysis system of invasive non-mucous lung adenocarcinoma based on deep learning | |
| CN117152506A (en) | Triple negative breast cancer immunophenotype prediction method and system based on multi-scale characteristics | |
| CN116758336A (en) | Medical image intelligent analysis system based on artificial intelligence | |
| CN114973244B (en) | System and method for automatically identifying mitosis of H & E staining pathological image of breast cancer | |
| CN113177602B (en) | Image classification method, device, electronic equipment and storage medium | |
| CN115439850A (en) | Image-text character recognition method, device, equipment and storage medium based on examination sheet |
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 |