CN112348106B - Breast ultrasonic image classification method based on key point learning - Google Patents

Breast ultrasonic image classification method based on key point learning Download PDF

Info

Publication number
CN112348106B
CN112348106B CN202011290015.8A CN202011290015A CN112348106B CN 112348106 B CN112348106 B CN 112348106B CN 202011290015 A CN202011290015 A CN 202011290015A CN 112348106 B CN112348106 B CN 112348106B
Authority
CN
China
Prior art keywords
malignant
probability
breast
malignancy
benign
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
Application number
CN202011290015.8A
Other languages
Chinese (zh)
Other versions
CN112348106A (en
Inventor
杨国武
陈琴
曹占涛
王刚
张孟华
孔令宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202011290015.8A priority Critical patent/CN112348106B/en
Publication of CN112348106A publication Critical patent/CN112348106A/en
Application granted granted Critical
Publication of CN112348106B publication Critical patent/CN112348106B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • G06V2201/032—Recognition of patterns in medical or anatomical images of protuberances, polyps nodules, etc.

Abstract

The invention discloses a breast ultrasound image classification method based on key point learning, which comprises the steps of preprocessing an ultrasound image set, dividing an ultrasound image training set into m subsets, and training m-1 benign and malignant prediction models based on a neural network; then, according to a BI-RADS rating standard, a learning method of 6 key points is given, and the input breast ultrasound image is subjected to BI-RADS rated multi-level classification prediction on each model; and finally, generating a multi-level classification prediction result of BI-RADS rating by voting, and adjusting corresponding benign and malignant probability values.

Description

Breast ultrasonic image classification method based on key point learning
Technical Field
The invention relates to the field of image processing, in particular to a breast ultrasound image classification method based on key point learning.
Background
Given an ultrasound image of the breast, the physician first diagnoses the image to determine whether a tumor is present in the image and whether it is benign or malignant, thereby giving a large direction for subsequent treatment. The existing technologies for realizing the same function are roughly as follows:
The first method comprises the following steps: identification suggestions are given based on a medical comprehensive assistance system in the business software.
The medical comprehensive auxiliary system basically works as follows: a provider of the commercial medical comprehensive auxiliary system judges whether an image is benign or malignant according to local data of the provider by using various decision-based processes, and a decision-making method of the system generally extracts manual features of the provider of commercial software according to an input image and processes the features algorithmically according to the manual features to obtain a result.
The disadvantages of this method are as follows: the non-open source of the commercial software causes the image distinguishing and diagnosing process to be not transparent, the principle of the image processing cannot be known, the extraction based on various characteristics is time-consuming, the image distinguishing usually needs to wait for a long time, and new data cannot be utilized to learn new modes, so that the distinguishing level is always the same, and the long-term use is not facilitated.
And the second method comprises the following steps: the image is identified based on a traditional machine learning method or deep learning, the traditional machine learning method needs to extract features manually, common image features include image edges, image brightness, direction gradient Histograms (HOGs), statistical information is used for describing and representing the image, the features can describe common images to a certain extent, and then the traditional machine learning method or the deep learning is used for predicting.
The disadvantages of this method are as follows: the method has the advantages that only a prediction result under a certain category is given, a probability value under the prediction is not given, and the prediction result given by deep learning generally cannot represent a true probability value of the prediction. While the general medical assistance system needs to use the prediction result with higher probability value and the very reliable prediction result as the assistance means, the traditional method obviously cannot meet the core requirement of medical assistance.
Disclosure of Invention
Aiming at the defects in the prior art, the breast ultrasound image classification method based on the key point learning provided by the invention solves the technical problem that the effect and the interpretability of the existing breast ultrasound tumor identification method based on the deep learning are balanced.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that: a breast ultrasound image classification method based on key point learning comprises the following steps:
S1, collecting breast ultrasound images, and preprocessing and marking benign and malignant breast ultrasound images to obtain a breast ultrasound image set;
s2, dividing the breast ultrasound image set into a plurality of subsets, training and verifying a plurality of benign and malignant prediction neural network models by adopting the plurality of subsets, and obtaining benign and malignant probabilities of a plurality of local test sets based on the benign and malignant prediction neural network models;
s3, constructing 6 probability key points in a malignant probability point set according to the benign and malignant probability of each local test set and the BI-RADS standard, and calculating the category of the breast ultrasound image;
s4, generating a BI-RADS rating multi-level classification prediction result by voting according to the category of the breast ultrasound image, adjusting the corresponding malignancy probability value of the breast tumor, and obtaining a BI-RADS classification result of the breast ultrasound image based on the malignancy probability value of the breast tumor.
Further, step S1 includes the following substeps:
s11, acquiring a breast ultrasonic image, and performing preprocessing work of image size unification and image tumor position interception on the breast ultrasonic image to obtain a preprocessed breast ultrasonic image;
s12, performing benign and malignant labeling on the preprocessed breast ultrasound image to obtain a breast ultrasound image set.
Further, step S2 includes the following substeps:
s21, dividing the breast ultrasound image set into 1-m subsets and a local test set;
s22, taking the kth sub-set as a verification set, taking the residual sub-set as a training set, constructing the kth training verification pair to obtain m training verification pairs, and training and verifying the kth benign and malignant prediction neural network model by adopting the kth training verification pair to obtain m trained benign and malignant prediction neural network models;
s23, testing the trained m benign and malignant prediction neural network models by using the local test set to obtain the benign and malignant probability of the m local test sets
Figure BDA0002783508170000031
Wherein k is more than or equal to 1 and less than or equal to m.
Further, step S3 includes the following substeps:
s31, probability of goodness of malignancy for m local test sets
Figure BDA0002783508170000032
Sorting according to the probability of benign or malignant of each sort
Figure BDA0002783508170000033
And BI-RADS criteria, setting the rating BI ═ 3,4a,4b,4c,5} for breast ultrasound images as { c ═ c1,c2,c3,c4,c5A malignancy probability point set B corresponding to each level { 0%, 2%, 10%, 50%, 95%, 100% };
s32, calculating a boundary probability value calibration function Z (B) according to the malignancy probability point set B ═ 0%, 2%, 10%, 50%, 95%, 100% }i)：
S33 probability of goodness of malignancy from local test set
Figure BDA0002783508170000034
Calibrating function Z (b) at boundary probability valuesi) In the range ofTo category c of breast ultrasound imagesi；
Figure BDA0002783508170000035
Further, the boundary probability value calibration function Z (b) is calculated in step S32i) The formula of (1) is:
Figure BDA0002783508170000041
Figure BDA0002783508170000042
Figure BDA0002783508170000043
V＝{s1,…,st}bi∈B
wherein the boundary probability value calibrates the function Z (b)i) For the corresponding 6 probability key points in the malignant probability point set B,
Figure BDA0002783508170000044
in order to correspond to the key point of 2%,
Figure BDA0002783508170000045
in order to correspond to the 10% key point,
Figure BDA0002783508170000046
to correspond to 95% of the keypoints, the keypoints corresponding to 50% are still 0.5, i is 0,1,2,3,4,5, k1For prediction as 3-class subset V1Maximum value of the number of middle samples, k2For prediction as 4a type subset V2Maximum value of the number of middle samples, k3Prediction as class 5 subset V5Maximum value of middle sample number, V1、V2And V3Is a subset of V, V is a validation set, { s1,…,stIs a verification set VT breast ultrasound image samples in (1),
Figure BDA0002783508170000047
to predict the subset of samples to be of class 3,
Figure BDA0002783508170000048
to predict the subset of samples for class 4a,
Figure BDA0002783508170000049
to predict a subset of samples that are of class 5,
Figure BDA00027835081700000410
is a V1The proportion of the medium-malignant tumors is,
Figure BDA00027835081700000411
is a V2The proportion of the medium-malignant tumors is,
Figure BDA00027835081700000412
is a V3Proportion of moderate to malignant tumors, biIs an element in the malignant probability point set B.
Further, step S4 includes the following substeps:
s41, predicting the probability of goodness or malice of the neural network model and each local test set according to the goodness or malice
Figure BDA00027835081700000413
Voting the category of the breast ultrasound image, and taking the category with high voting rate as a prediction result of BI-RADS rating
Figure BDA00027835081700000414
S42 predicting results from BI-RADS
Figure BDA00027835081700000415
Calculating the malignancy probability value of the breast tumor;
s43, obtaining a classification result of the breast ultrasound image BI-RADS according to the malignancy probability value of the breast tumor.
Further, the formula for calculating the malignancy probability value of the breast tumor in step S42 is:
Figure BDA0002783508170000051
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0002783508170000052
is the probability value of the malignancy of the breast tumor,
Figure BDA0002783508170000053
the average result of the malignancy probability values of the models after voting in accordance with the voting results,
Figure BDA0002783508170000054
pooled samples are predicted as BI-RADS ratings for local testing
Figure BDA0002783508170000055
The corresponding malignancy rate is at the upper bound,
Figure BDA0002783508170000056
pooled samples are predicted as BI-RADS ratings for local testing
Figure BDA0002783508170000057
The corresponding lower bound of malignancy.
In conclusion, the beneficial effects of the invention are as follows:
1. the invention constructs a mammary gland ultrasonic identification method by using a mammary gland-based ultrasonic image and provides a probability value on each identification result, thereby improving the identification accuracy and interpretability and being more appropriate to an actual scene during application; the method adopts an end-to-end flow, and the processing of the image is faster than that of the traditional learning method based on the characteristics because a large amount of manual characteristics in the traditional image recognition are not used.
2. The invention uses the neural network as a tool, converts the general benign and malignant classification problem into a method for realizing multi-classification through a probability value calibration method, and realizes more accurate image identification of multiple classes.
3. The method realizes the determination of the BI-RADS category defined by the American medical imaging society by using the probability value calibration, can reach the BI-RADS classification standard recommended by the American imaging society, and can meet the recommended requirements on all levels of malignancy risks, so that the method can be applied to breast ultrasound diagnosis and identification work in a large scale, thereby realizing the more deep understanding of the identification of the image and being beneficial to guiding the medical image identification task in practical application.
4. The method can be suitable for other similar medical image recognition tasks, different tasks may need to determine different feature extraction methods, if an available training data set is large, a residual error network or a deeper deep learning framework can be directly used, and if the available training set is small, shallow neural networks such as AlexNet can be directly used for extracting features.
Drawings
Fig. 1 is a flowchart of a breast ultrasound image classification method based on keypoint learning.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
As shown in fig. 1, a breast ultrasound image classification method based on keypoint learning includes the following steps:
s1, collecting breast ultrasound images, and preprocessing and marking benign and malignant breast ultrasound images to obtain a breast ultrasound image set;
step S1 includes the following substeps:
s11, acquiring a breast ultrasonic image, and carrying out pretreatment work of image size unification and image tumor part interception on the breast ultrasonic image to obtain a pretreated breast ultrasonic image;
s12, performing benign and malignant labeling on the preprocessed breast ultrasound image to obtain a breast ultrasound image set.
S2, dividing the breast ultrasound image set into a plurality of subsets, training and verifying a plurality of benign and malignant prediction neural network models by adopting the plurality of subsets, and obtaining benign and malignant probabilities of a plurality of local test sets based on the benign and malignant prediction neural network models;
Step S2 includes the following substeps:
s21, dividing the breast ultrasound image set into 1-m subsets and a local test set;
s22, taking the kth sub-set as a verification set and the residual sub-set as a training set, constructing the kth training verification pair to obtain m training verification pairs, and training and verifying the kth benign and malignant prediction neural network model by adopting the kth training verification pair to obtain m trained benign and malignant prediction neural network models;
s23, testing the trained m benign and malignant prediction neural network models by using the local test set to obtain the benign and malignant probabilities of the m local test sets
Figure BDA0002783508170000071
Wherein k is more than or equal to 1 and less than or equal to m.
S3, constructing 6 probability key points in the malignancy probability point set according to the benign and malignant probability of each local test set and the BI-RADS standard, and calculating the category of the breast ultrasound image;
step S3 includes the following substeps:
s31 probability of benign and malignant of m local test sets
Figure BDA0002783508170000072
Sorting according to the probability of benign or malignant of each sort
Figure BDA0002783508170000073
And BI-RADS criteria, setting the rating of breast ultrasound images to be BI ═ {3,4a,4b,4c,5} as { c1,c2,c3,c4,c5Each grade of the malignancy probability point set B is { 0%, 2%, 10%, 50%, 95%, 100% };
s32, calculating a boundary probability value calibration function Z (B) according to the malignancy probability point set B ═ 0%, 2%, 10%, 50%, 95%, 100% } i)：
In step S32, a boundary probability value calibration function Z (b) is calculatedi) The formula of (1) is as follows:
Figure BDA0002783508170000074
Figure BDA0002783508170000075
Figure BDA0002783508170000076
V＝{s1,…,st}bi∈B
wherein the boundary probability value calibrates the function Z (b)i) For the corresponding 6 probabilistic keypoints in the malignant probabilistic point set B,
Figure BDA0002783508170000077
in order to correspond to the 2% key point,
Figure BDA0002783508170000078
in order to correspond to the 10% key point,
Figure BDA0002783508170000079
to correspond to 95% of the keypoints, the keypoints corresponding to 50% are still 0.5, i is 0,1,2,3,4,5, k1For prediction as 3-class subset V1Maximum value of the number of middle samples, k2For prediction as 4a type subset V2Maximum value of the number of middle samples, k3Prediction as class 5Subset V5Maximum value of middle sample number, V1、V2And V3Is a subset of V, V is a validation set, { s1,…,stThe t breast ultrasound image samples in the validation set V,
Figure BDA0002783508170000081
to predict the subset of samples to be of class 3,
Figure BDA0002783508170000082
to predict the subset of samples for class 4a,
Figure BDA0002783508170000083
to predict a subset of samples that are of class 5,
Figure BDA0002783508170000084
is a V1The proportion of the medium-malignant tumors is,
Figure BDA0002783508170000085
is a V2The proportion of the medium-malignant tumors is,
Figure BDA0002783508170000086
is a V3Proportion of moderate to malignant tumors, biFor elements in the malignancy probability point set B, {3,4a,4B,4c,5} is the name in the BI-RADS standard.
S33 probability of goodness of malignancy from local test set
Figure BDA0002783508170000087
Calibrating function Z (b) at boundary probability valuesi) The range in (1), the category c of the breast ultrasound image is obtainedi；
Figure BDA0002783508170000088
S4, according to the category c of the breast ultrasound image iGenerating a multi-level classification prediction result of BI-RADS rating by using voting, and adjusting malignancy probability value of corresponding breast tumor, and based on the resultAnd obtaining the malignant probability value of the breast tumor to obtain a breast ultrasound image BI-RADS classification result.
Step S4 includes the following substeps
S41, predicting the probability of goodness or malice of the neural network model and each local test set according to the goodness or malice
Figure BDA0002783508170000089
Voting the categories of the breast ultrasound images, and taking the categories with high voting rate as the prediction result of BI-RADS rating
Figure BDA00027835081700000810
S42 predicting results based on BI-RADS
Figure BDA00027835081700000811
Calculating the malignancy probability value of the breast tumor;
the formula for calculating the malignancy probability value of the breast tumor in step S42 is:
Figure BDA00027835081700000812
wherein the content of the first and second substances,
Figure BDA00027835081700000813
is the probability value of the malignancy of the breast tumor,
Figure BDA00027835081700000814
the average result of the malignancy probability values of the models after voting in accordance with the voting results,
Figure BDA00027835081700000815
pooled samples are predicted as BI-RADS ratings for local testing
Figure BDA00027835081700000816
The corresponding malignancy rate is at the upper bound,
Figure BDA00027835081700000817
pooled samples are predicted as BI-RADS ratings for local testing
Figure BDA00027835081700000818
The corresponding lower bound of malignancy.
S43, obtaining a classification result of the breast ultrasound image BI-RADS according to the malignancy probability value of the breast tumor.

Claims (2)

1. A breast ultrasound image classification method based on key point learning is characterized by comprising the following steps:
S1, collecting breast ultrasound images, and preprocessing and marking benign and malignant breast ultrasound images to obtain a breast ultrasound image set;
s2, dividing the breast ultrasound image set into a plurality of subsets, training and verifying a plurality of benign and malignant prediction neural network models by adopting the plurality of subsets, and obtaining benign and malignant probabilities of a plurality of local test sets based on the benign and malignant prediction neural network models;
step S2 includes the following substeps:
s21, dividing the breast ultrasound image set into 1-m subsets and a local test set;
s22, taking the kth sub-set as a verification set, taking the residual sub-set as a training set, constructing the kth training verification pair to obtain m training verification pairs, and training and verifying the kth benign and malignant prediction neural network model by adopting the kth training verification pair to obtain m trained benign and malignant prediction neural network models;
s23, testing the trained m benign and malignant prediction neural network models by using the local test set to obtain the benign and malignant probabilities of the m local test sets
Figure FDA0003634518000000011
Wherein k is more than or equal to 1 and less than or equal to m;
s3, constructing 6 probability key points in the malignancy probability point set according to the benign and malignant probability of each local test set and the BI-RADS standard, and calculating the category of the breast ultrasound image;
Step S3 includes the following substeps:
s31, probability of goodness of malignancy for m local test sets
Figure FDA0003634518000000012
Sorting according to the probability of benign or malignant of each sort
Figure FDA0003634518000000013
And BI-RADS criteria, setting the rating BI ═ 3, 4a, 4b, 4c, 5} for breast ultrasound images as { c ═ c1，c2，c3，c4，c5Each grade of the malignancy probability point set B is { 0%, 2%, 10%, 50%, 95%, 100% };
s32, calculating a boundary probability value calibration function Z (B) according to the malignancy probability point set B ═ 0%, 2%, 10%, 50%, 95%, 100% }i)：
In step S32, a boundary probability value calibration function Z (b) is calculatedi) The formula of (1) is:
Figure FDA0003634518000000021
Figure FDA0003634518000000022
Figure FDA0003634518000000023
V＝{s1，…，st}，bi∈B
wherein the boundary probability value calibrates the function Z (b)i) For the corresponding 6 probability key points in the malignant probability point set B,
Figure FDA0003634518000000024
in order to correspond to the key point of 2%,
Figure FDA0003634518000000025
in order to correspond to the 10% key point,
Figure FDA0003634518000000026
as the key point corresponding to 95%, i ═ 0, 1, 2, 3, 4, 5, k1For prediction as 3-class subset V1Maximum value of the number of middle samples, k2For prediction as 4a type subset V2Maximum value of the number of middle samples, k3Prediction as class 5 subset V5Maximum value of middle sample number, V1、V2And V3Is a subset of V, V is a validation set, { s1，…，stThe t breast ultrasound image samples in the validation set V,
Figure FDA0003634518000000027
to predict the subset of samples to be of class 3,
Figure FDA0003634518000000028
to predict the subset of samples for class 4a,
Figure FDA0003634518000000029
To predict the subset of samples that are of class 5,
Figure FDA00036345180000000210
is a V1The proportion of the medium-malignant tumors is,
Figure FDA00036345180000000211
is a V2The proportion of the medium-malignant tumors is,
Figure FDA00036345180000000212
is a V3Proportion of moderate to malignant tumors, biElements in the malignant probability point set B;
s33 probability of goodness of malignancy from local test set
Figure FDA00036345180000000213
Calibrating function Z (b) at boundary probability valuesi) The range in (1), the category c of the breast ultrasound image is obtainedi；
Figure FDA00036345180000000214
S4, generating a BI-RADS rating multi-level classification prediction result by voting according to the category of the breast ultrasound image, adjusting the corresponding malignancy probability value of the breast tumor, and obtaining a BI-RADS classification result of the breast ultrasound image based on the malignancy probability value of the breast tumor;
step S4 includes the following substeps:
s41, predicting the probability of goodness or malice of the neural network model and each local test set according to the goodness or malice
Figure FDA0003634518000000031
Voting the categories of the breast ultrasound images, and taking the categories with high voting rate as the prediction result of BI-RADS rating
Figure FDA0003634518000000032
S42 predicting results based on BI-RADS
Figure FDA0003634518000000033
Calculating the malignancy probability value of the breast tumor;
the formula for calculating the malignancy probability value of the breast tumor in step S42 is:
Figure FDA0003634518000000034
wherein the content of the first and second substances,
Figure FDA0003634518000000035
is the probability value of the malignancy of the breast tumor,
Figure FDA0003634518000000036
the average result of the malignancy probability values of the models after voting in accordance with the voting results,
Figure FDA0003634518000000037
Samples pooled for local testing are predicted as BI-RADS ratings
Figure FDA0003634518000000038
The corresponding upper bound of the malignancy rate is reached,
Figure FDA0003634518000000039
samples pooled for local testing are predicted as BI-RADS ratings
Figure FDA00036345180000000310
The lower bound of the corresponding malignancy;
s43, obtaining a breast ultrasound image BI-RADS classification result according to the malignancy probability value of the breast tumor.
2. The method for classifying breast ultrasound images based on keypoint learning according to claim 1, wherein said step S1 comprises the following substeps:
s11, acquiring a breast ultrasonic image, and performing preprocessing work of image size unification and image tumor position interception on the breast ultrasonic image to obtain a preprocessed breast ultrasonic image;
and S12, performing benign and malignant labeling on the preprocessed breast ultrasound image to obtain a breast ultrasound image set.
CN202011290015.8A 2020-11-18 2020-11-18 Breast ultrasonic image classification method based on key point learning Active CN112348106B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011290015.8A CN112348106B (en) 2020-11-18 2020-11-18 Breast ultrasonic image classification method based on key point learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011290015.8A CN112348106B (en) 2020-11-18 2020-11-18 Breast ultrasonic image classification method based on key point learning

Publications (2)

Publication Number Publication Date
CN112348106A CN112348106A (en) 2021-02-09
CN112348106B true CN112348106B (en) 2022-06-28

Family

ID=74364111

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011290015.8A Active CN112348106B (en) 2020-11-18 2020-11-18 Breast ultrasonic image classification method based on key point learning

Country Status (1)

Country Link
CN (1) CN112348106B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115410137B (en) * 2022-11-01 2023-04-14 杭州新中大科技股份有限公司 Double-flow worker labor state identification method based on space-time characteristics

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110264462A (en) * 2019-06-25 2019-09-20 电子科技大学 A kind of breast ultrasound tumour recognition methods based on deep learning
CN111402205A (en) * 2020-02-27 2020-07-10 华南理工大学 Mammary gland tumor data cleaning method based on multilayer perceptron
CN111743574A (en) * 2020-07-12 2020-10-09 西北工业大学 Method for constructing knowledge graph of breast cancer ultrasonic image high confidence entity relationship

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10580137B2 (en) * 2018-01-30 2020-03-03 International Business Machines Corporation Systems and methods for detecting an indication of malignancy in a sequence of anatomical images
CN110838110A (en) * 2019-11-05 2020-02-25 张峰 System for identifying benign and malignant tumor based on ultrasonic imaging

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110264462A (en) * 2019-06-25 2019-09-20 电子科技大学 A kind of breast ultrasound tumour recognition methods based on deep learning
CN111402205A (en) * 2020-02-27 2020-07-10 华南理工大学 Mammary gland tumor data cleaning method based on multilayer perceptron
CN111743574A (en) * 2020-07-12 2020-10-09 西北工业大学 Method for constructing knowledge graph of breast cancer ultrasonic image high confidence entity relationship

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Wenjing Yang等.Confidence Calibration on Multiclass Classification in Medical Imaging.《2020 IEEE International Conference on Data Mining (ICDM)》.2021, *
陈永东.基于超声征象打分特征的乳腺肿瘤分类研究.《中国优秀硕士学位论文全文数据库 (医药卫生科技辑)》.2018,(第06期), *
黄冬玲等.高频超声结合BI-RADS分级在乳腺良恶性病变中的应用价值.《医学信息》.2019,第32卷(第10期), *

Also Published As

Publication number Publication date
CN112348106A (en) 2021-02-09

Similar Documents

Publication Publication Date Title
CN108805196B (en) Automatic incremental learning method for image recognition
US8676731B1 (en) Data extraction confidence attribute with transformations
CN109815976A (en) A kind of certificate information recognition methods, device and equipment
CN108171203B (en) Method and device for identifying vehicle
CN109492750B (en) Zero sample image classification method based on convolutional neural network and factor space
CN113947607A (en) Cancer pathology image survival prognosis model construction method based on deep learning
CN112926045B (en) Group control equipment identification method based on logistic regression model
CN109460471B (en) Method for establishing fiber category map library based on self-learning mode
CN112348106B (en) Breast ultrasonic image classification method based on key point learning
CN113033667B (en) Ultrasound image two-stage deep learning breast tumor classification method and device
CN113177554B (en) Thyroid nodule identification and segmentation method, system, storage medium and equipment
CN110059705A (en) A kind of OCR recognition result decision method and equipment based on modeling
CN110110734A (en) Opener recognition methods, information processing equipment and storage medium
CN117036904A (en) Attention-guided semi-supervised corn hyperspectral image data expansion method
CN115357718B (en) Method, system, device and storage medium for discovering repeated materials of theme integration service
CN116434273A (en) Multi-label prediction method and system based on single positive label
CN113870194B (en) Breast tumor ultrasonic image processing device with fusion of deep layer characteristics and shallow layer LBP characteristics
CN113158878B (en) Heterogeneous migration fault diagnosis method, system and model based on subspace
CN104636758B (en) A kind of SAR image suitability Forecasting Methodology based on support vector regression
CN115511798A (en) Pneumonia classification method and device based on artificial intelligence technology
CN110427973B (en) Classification method for ambiguity-oriented annotation samples
CN114863163A (en) Method and system for cell classification based on cell image
CN113988226A (en) Data desensitization validity verification method and device, computer equipment and storage medium
CN104778479A (en) Image classification method and system based on sparse coding extractor
CN111221995B (en) Sequence matching method based on big data and set theory

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