CN114529772B - OCT three-dimensional image classification method, system, computer device and storage medium - Google Patents

OCT three-dimensional image classification method, system, computer device and storage medium Download PDF

Info

Publication number
CN114529772B
CN114529772B CN202210407731.2A CN202210407731A CN114529772B CN 114529772 B CN114529772 B CN 114529772B CN 202210407731 A CN202210407731 A CN 202210407731A CN 114529772 B CN114529772 B CN 114529772B
Authority
CN
China
Prior art keywords
scan
image
classification
determining
oct
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
CN202210407731.2A
Other languages
Chinese (zh)
Other versions
CN114529772A (en
Inventor
安林
秦嘉
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.)
Weiren Medical Foshan Co ltd
Guangdong Weiren Medical Technology Co ltd
Original Assignee
Weiren Medical Foshan Co ltd
Guangdong Weiren Medical Technology Co ltd
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 Weiren Medical Foshan Co ltd, Guangdong Weiren Medical Technology Co ltd filed Critical Weiren Medical Foshan Co ltd
Priority to CN202210407731.2A priority Critical patent/CN114529772B/en
Publication of CN114529772A publication Critical patent/CN114529772A/en
Application granted granted Critical
Publication of CN114529772B publication Critical patent/CN114529772B/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
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—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
    • G06T2207/10072—Tomographic images
    • G06T2207/10101—Optical tomography; Optical coherence tomography [OCT]
    • 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]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an OCT three-dimensional image classification method, a system, a computer device and a storage medium, wherein the method comprises the following steps: labeling the first B-scan image according to the image type of the first OCT three-dimensional image, and training to obtain a first B-scan classification model; classifying the first B-scan image according to the first B-scan classification model, and determining a third B-scan image and a second B-scan image according to whether the classification result is consistent with the B-scan label; screening out the similar B-scan images of the second B-scan image, resetting labels according to the B-scan labels of the similar B-scan images, and performing model optimization on the first B-scan classification model to obtain a second B-scan classification model for OCT three-dimensional image classification. The invention improves the accuracy of OCT three-dimensional image classification, and can be widely applied to the technical field of image processing.

Description

OCT three-dimensional image classification method, system, computer device and storage medium
Technical Field
The invention relates to the technical field of image processing, in particular to an OCT three-dimensional image classification method, an OCT three-dimensional image classification system, a computer device and a storage medium.
Background
Optical Coherence Tomography (OCT) is an imaging technique rapidly developed in recent years, which uses light Coherence to detect back-reflected or several scattered signals of incident weak coherent light at different depth levels of a biological tissue, and scans the biological tissue to obtain a three-dimensional structural image.
In the prior art, there are two general methods for classifying OCT three-dimensional images: firstly, a known type of OCT three-dimensional image is directly marked, then a neural network for OCT three-dimensional image classification is trained according to an obtained OCT three-dimensional image sample, the OCT three-dimensional image is usually composed of hundreds of B-scan (cross-section scanning) images, the requirement on system computing power for directly processing the high-resolution OCT three-dimensional image is high, and the efficiency of model training and classification is low; secondly, labeling a B-scan image of an OCT three-dimensional image of a known type, training a neural network for classifying the B-scan image, and determining a classification result of the OCT three-dimensional image according to the classification result of the B-scan image, however, the method usually directly adopts the type of the OCT three-dimensional image as a label of the B-scan image, but when the type of the OCT three-dimensional image is abnormal, only a part of the B-scan image has corresponding abnormal characteristics, and the other part of the B-scan image does not have the abnormal characteristics, so that the label of a training sample is inaccurate, the classification accuracy of the B-scan image is influenced, and the classification accuracy of the OCT three-dimensional image is further influenced.
Disclosure of Invention
The present invention aims to solve at least to some extent one of the technical problems of the prior art.
Therefore, an object of the embodiments of the present invention is to provide an accurate and efficient OCT three-dimensional image classification method.
Another object of an embodiment of the present invention is to provide an OCT three-dimensional image classification system.
In order to achieve the technical purpose, the technical scheme adopted by the embodiment of the invention comprises the following steps:
in one aspect, an embodiment of the present invention provides an OCT three-dimensional image classification method, including the following steps:
acquiring a plurality of first OCT three-dimensional images and corresponding image types, determining a plurality of first B-scan images of the first OCT three-dimensional images, labeling the first B-scan images according to the image types to obtain B-scan labels, and training according to the first B-scan images and the B-scan labels to obtain a first B-scan classification model;
performing image classification on the first B-scan image according to the first B-scan classification model, and determining a second B-scan image and a third B-scan image according to a classification result, wherein the second B-scan image is the first B-scan image of which the classification result is inconsistent with the corresponding B-scan label, and the third B-scan image is the first B-scan image of which the classification result is consistent with the corresponding B-scan label;
determining a second texture feature vector of the second B-scan image and a third texture feature vector of the third B-scan image, determining cosine similarity of the second texture feature vector and the third texture feature vector, and screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity;
performing label resetting on the second B-scan image according to the B-scan label of the similar B-scan image, and performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after label resetting to obtain a trained second B-scan classification model;
acquiring a plurality of B-scan images to be classified of a second OCT three-dimensional image to be classified, inputting the B-scan images to be classified into the second B-scan classification model to obtain a B-scan classification result of the B-scan images to be classified, and determining a classification result of the second OCT three-dimensional image according to the B-scan classification result.
Further, in an embodiment of the present invention, the step of training to obtain the first B-scan classification model according to the first B-scan image and the B-scan label specifically includes:
determining a plurality of training samples according to the first B-scan image and the B-scan label, and determining a training sample set according to the training samples;
inputting the training sample set into a pre-constructed capsule neural network for training to obtain the first B-scan classification model;
the capsule neural network comprises a convolution layer, a main capsule layer, a digital capsule layer and a full connection layer, wherein the convolution layer is used for performing convolution processing on the training samples to obtain a plurality of sample characteristic sub-images, the main capsule layer is used for determining a plurality of vector neurons according to the sample characteristic sub-images, the digital capsule layer is used for calculating the activation vectors of the vector neurons through a dynamic routing algorithm, and the full connection layer is used for determining the classification results of the training samples according to the modular lengths of the activation vectors.
Further, in an embodiment of the present invention, the step of determining the second texture feature vector of the second B-scan image specifically includes:
carrying out Gabor wavelet transform on the second B-scan image to obtain a second texture image;
and determining a first mean value and a first variance of the second texture image, and further determining the second texture feature vector according to the first mean value and the first variance.
Further, in an embodiment of the present invention, the step of screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity specifically includes:
when the cosine similarity is larger than or equal to a preset first threshold value, determining that a third B-scan image corresponding to the third texture feature vector is a similar B-scan image of a second B-scan image corresponding to the second texture feature vector;
or the like, or a combination thereof,
and sequencing the third texture feature vectors according to the cosine similarity to obtain a texture feature vector sequence, wherein the third texture feature vectors in the texture feature vector sequence are sequentially arranged from large to small according to the corresponding cosine similarity, and when the sequence value of the third texture feature vectors is smaller than or equal to a preset second threshold, determining that the third B-scan image corresponding to the third texture feature vectors is a similar B-scan image of the second B-scan image corresponding to the second texture feature vectors.
Further, in an embodiment of the present invention, the step of performing label resetting on the second B-scan image according to the B-scan label of the similar B-scan image specifically includes:
determining the frequency of occurrence of each type of B-scan label in the B-scan labels of the similar B-scan images;
and selecting the B-scan label with the highest occurrence frequency as an optimized label, and resetting the B-scan label of the second B-scan image as the optimized label.
Further, in an embodiment of the present invention, the step of performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after the label is reset to obtain a trained second B-scan classification model specifically includes:
determining a plurality of optimized samples according to the third B-scan image and the second B-scan image after the label is reset, and determining an optimized sample set according to the optimized samples;
inputting the optimized sample set into the first B-scan classification model for training to obtain an optimized B-scan classification model;
classifying the optimized samples according to the optimized B-scan classification model, determining the optimized samples with the classification results inconsistent with the corresponding B-scan labels as fourth B-scan images, and calculating a first ratio of the number of the fourth B-scan images to the total number of the optimized samples;
when the first ratio is larger than or equal to a preset third threshold, carrying out label resetting on the fourth B-scan image, re-determining an optimized sample set according to a label resetting result, returning and inputting the optimized sample set into the first B-scan classification model for training, and obtaining an optimized B-scan classification model;
and when the first ratio is smaller than a preset third threshold value, stopping training to obtain a trained second B-scan classification model.
Further, in an embodiment of the present invention, the B-scan classification result includes a normal B-scan picture and a plurality of abnormal B-scan pictures with different abnormal types, and the step of determining the classification result of the second OCT three-dimensional image according to the B-scan classification result specifically includes:
determining the number of the abnormal B-scan pictures according to the B-scan classification result, and calculating a second ratio of the number of the abnormal B-scan pictures to the total number of the B-scan pictures to be classified;
when the second ratio is larger than or equal to a preset fourth threshold, determining that the second OCT three-dimensional image is an abnormal OCT three-dimensional image, and determining the abnormal type of the second OCT three-dimensional image according to the abnormal type with the highest occurrence frequency in the abnormal B-scan picture;
and when the second ratio is smaller than a preset fourth threshold, determining that the second OCT three-dimensional image is a normal OCT three-dimensional image.
In another aspect, an embodiment of the present invention provides an OCT three-dimensional image classification system, including:
the first B-scan classification model training module is used for acquiring a plurality of first OCT three-dimensional images and corresponding image types, determining a plurality of first B-scan images of the first OCT three-dimensional images, labeling the first B-scan images according to the image types to obtain B-scan labels, and then training according to the first B-scan images and the B-scan labels to obtain a first B-scan classification model;
the first B-scan image classification module is used for performing image classification on the first B-scan image according to the first B-scan classification model, and determining a second B-scan image and a third B-scan image according to a classification result, wherein the second B-scan image is a first B-scan image with a classification result inconsistent with a corresponding B-scan label, and the third B-scan image is a first B-scan image with a classification result consistent with a corresponding B-scan label;
the similar B-scan image determining module is used for determining a second texture feature vector of the second B-scan image and a third texture feature vector of the third B-scan image, determining cosine similarity of the second texture feature vector and the third texture feature vector, and further screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity;
the label resetting and model optimizing module is used for resetting the label of the second B-scan image according to the B-scan label of the similar B-scan image, and performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after label resetting to obtain a trained second B-scan classification model;
the second OCT three-dimensional image classification module is used for acquiring a plurality of B-scan images to be classified of a second OCT three-dimensional image to be classified, inputting the B-scan images to be classified into the second B-scan classification model to obtain a B-scan classification result of the B-scan images to be classified, and further determining the classification result of the second OCT three-dimensional image according to the B-scan classification result.
In another aspect, an embodiment of the present invention provides a computer apparatus, including:
at least one processor;
at least one memory for storing at least one program;
when executed by the at least one processor, causes the at least one processor to implement a method of OCT three-dimensional image classification as previously described.
In another aspect, an embodiment of the present invention further provides a computer-readable storage medium, in which a processor-executable program is stored, where the processor-executable program is used to execute the OCT three-dimensional image classification method described above when executed by a processor.
Advantages and benefits of the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention:
according to the embodiment of the invention, a first B-scan image is pre-labeled according to the image type of a first OCT three-dimensional image, and a first B-scan classification model is obtained according to the pre-labeled first B-scan image training; classifying the first B-scan image according to the first B-scan classification model, and respectively determining a third B-scan image and a second B-scan image according to whether the classification result is consistent with the B-scan label; then screening out a similar B-scan image of the second B-scan image according to the cosine similarity of the texture feature vectors of the first B-scan image and the second B-scan image; performing label resetting on the second B-scan image according to the B-scan labels of the similar B-scan images, and performing model optimization on the first B-scan classification model to obtain a second B-scan classification model; and finally, classifying the B-scan image to be classified of the second OCT three-dimensional image according to the second B-scan classification model, and further determining the classification result of the second OCT three-dimensional image according to the classification result of the B-scan image to be classified. According to the embodiment of the invention, the training sample with the possibly inaccurate label is screened out according to the consistency of the model classification result and the B-scan label, and then the label is reset according to the label of the similar B-scan image, so that the reset label is more accurate, the accuracy of B-scan image classification is improved, and the accuracy of OCT three-dimensional image classification is further improved; determining a similar B-scan image of the second B-scan image according to the cosine similarity of the texture feature vectors of the two images, so that the similar B-scan image is highly similar to the second B-scan image in texture features, the accuracy of label resetting is improved to a certain extent, and the accuracy of B-scan image classification and OCT three-dimensional image classification is further improved; through the alternate iteration of label resetting and model optimization, labels and model parameters of the training samples are optimized, and the classification precision of the second B-scan classification model is improved, so that the accuracy of OCT three-dimensional image classification is further improved; in addition, the embodiment of the invention does not need to directly process the high-resolution OCT three-dimensional image, reduces the requirement on the calculation power of the system and improves the efficiency of model training and classification.
Drawings
In order to more clearly illustrate the technical solution in the embodiment of the present invention, the following description is made on the drawings required to be used in the embodiment of the present invention, and it should be understood that the drawings in the following description are only for convenience and clarity of describing some embodiments in the technical solution of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a flowchart illustrating steps of an OCT three-dimensional image classification method according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of a specific OCT three-dimensional image classification method according to an embodiment of the present invention;
fig. 3 is a structural block diagram of an OCT three-dimensional image classification system according to an embodiment of the present invention;
fig. 4 is a block diagram of a computer device according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention and are not to be construed as limiting the present invention. The step numbers in the following embodiments are provided only for convenience of illustration, the order between the steps is not limited at all, and the execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.
In the description of the present invention, the meaning of a plurality is two or more, if there is a description to the first and the second for the purpose of distinguishing technical features, it is not understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the precedence of the indicated technical features. Furthermore, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
Referring to fig. 1, an embodiment of the present invention provides an OCT three-dimensional image classification method, which specifically includes the following steps:
s101, acquiring a plurality of first OCT three-dimensional images and corresponding image types, determining a plurality of first B-scan images of the first OCT three-dimensional images, labeling the first B-scan images according to the image types to obtain B-scan labels, and training according to the first B-scan images and the B-scan labels to obtain a first B-scan classification model.
Specifically, the first OCT three-dimensional image is an OCT three-dimensional image of a known image type, the image type of the first OCT three-dimensional image can be determined to be a normal or specific abnormal type through manual labeling, and then all B-scan images (i.e., the first B-scan image) of the first OCT three-dimensional image are labeled according to the image type to obtain the B-scan label.
As can be understood, the OCT three-dimensional image comprises a large number of B-scan images, so that enough B-scan images can be obtained as subsequent training samples through manual marking of the OCT three-dimensional image; when the type of the OCT three-dimensional image is abnormal, only a part of the B-scan image often has corresponding abnormal features, and the other part of the B-scan image does not have the abnormal features, so that the label of the current training sample is not accurate.
Further as an optional implementation manner, the step of training to obtain the first B-scan classification model according to the first B-scan image and the B-scan label specifically includes:
a1, determining a plurality of training samples according to the first B-scan image and the B-scan label, and determining a training sample set according to the training samples;
a2, inputting a training sample set into a pre-constructed capsule neural network for training to obtain a first B-scan classification model;
the capsule neural network comprises a convolution layer, a main capsule layer, a digital capsule layer and a full-connection layer, wherein the convolution layer is used for performing convolution processing on a training sample to obtain a plurality of sample characteristic subgraphs, the main capsule layer is used for determining a plurality of vector neurons according to the sample characteristic subgraphs, the digital capsule layer is used for calculating an activation vector of each vector neuron through a dynamic routing algorithm, and the full-connection layer is used for determining a classification result of the training sample according to the modular length of the activation vector.
In particular, the capsule neural network is identified as a new neural network that can replace the conventional neural network, and the design of the capsule is more consistent with the principle of human neurons. Although the network can learn context information including the global context as the number of layers of the convolutional network increases, and then use the context information to predict, in practice, because the convolution is a local connection and parameter sharing, and the correlation and the mutual position relationship between the features are not considered, there is no spatial information actually available in the CNN, and therefore the correlation and the spatial position relationship between the features cannot be used as a determination condition in image classification and recognition. According to the capsule neural network provided by the embodiment of the invention, the main capsule layer and the digital capsule layer are added after the convolution layer, the scalar data of the convolution layer is subjected to dimension expansion, and the extracted features in the image are correlated, so that the final classification judgment can be favorably carried out integrally, and the accuracy of OCT three-dimensional image classification is further improved.
The capsule neural network used in the embodiments of the present invention is described below.
Specifically, neurons are the basic units of neural networks, and capsules are the basic units of capsule neural networks. When only neurons are used for prediction, the final output result only has specific classification probability values, and the specific recognition process of the neurons is difficult to explain; after combining a plurality of neurons with the same recognition characteristics in the form of capsules, a final prediction result can be output in the form of a multi-dimensional vector, the modular length of the multi-dimensional vector represents a corresponding prediction probability value, and the direction of the vector represents the orientation of specific features in an image, including position information such as the direction, size and width of a feature subgraph.
When the model is trained, firstly, the characteristic subgraph is led into the capsule neural network according to the label type, and then convolution operation is carried out through the convolution layer. The parameters of the convolutional layer consist of a series of iteratively learnable filters, each filter being small in width and height, and can be set to 3 x 3, with the input and data dimensions being consistent. As the filters move along the width and height of the image, a two-dimensional active subgraph is generated, with each filter having an entire set of filters, forming multiple active subgraphs.
Then capsule convolution and capsule output are carried out. The capsule output is the vector output of a set number of capsule layers, the classification result cannot be directly read, and the model length of the capsule needs to be taken in subsequent operation to judge the type. And removing the dimension with the dimension of 1 in the result vector output by the capsule to achieve the effect of consistent dimension. And then the processed result is judged by applying a full connection layer.
The fully connected layer acts as a "classifier" in subsequent operations. The last layer of the network is used as the input of the fully connected layer, and the Sigmoid activation function can be selected. And judging the final output dimension of the full-connection layer according to the number of the input samples and the number of the categories, displaying the final output vector result in a two-dimensional array form, wherein the length of the output vector represents the probability value of the corresponding category.
It can be understood that in the embodiment of the invention, after the capsule neural network is trained for the preset times, the first B-scan classification model can be obtained, and then training is continued after the label is reset.
S102, carrying out image classification on the first B-scan image according to the first B-scan classification model, and determining a second B-scan image and a third B-scan image according to the classification result, wherein the second B-scan image is the first B-scan image with the classification result inconsistent with the corresponding B-scan label, and the third B-scan image is the first B-scan image with the classification result consistent with the corresponding B-scan label.
Specifically, because the image category of the OCT three-dimensional image is adopted when the first B-scan image is labeled, and the label of the first B-scan image is not completely accurate, after the first B-scan classification model is obtained through preliminary training, the first B-scan image which may be inaccurate in label is screened out according to whether the classification result of the first B-scan image by the first B-scan classification model is consistent with the corresponding B-scan label, and is subjected to label resetting in subsequent steps.
S103, determining a second texture feature vector of the second B-scan image and a third texture feature vector of the third B-scan image, determining cosine similarity of the second texture feature vector and the third texture feature vector, and screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity.
Specifically, the texture feature is a global feature, which describes surface properties of an object corresponding to an image or an image region, and requires statistical calculation in a region including a plurality of pixel points. In pattern matching, the regional characteristics have great superiority, and the situation that the matching cannot be successful due to local deviation is avoided; in addition, as a statistical feature, the texture feature has rotation invariance and has strong resistance to noise.
Cosine similarity, also called cosine similarity, is measured by measuring the cosine value of the angle between two vectors, and can be calculated by the following formula:
Figure 504903DEST_PATH_IMAGE001
wherein similarity represents cosine similarity, and theta represents an included angle between vectors A and B.
When the two vectors have the same direction, the cosine similarity value is 1; when the included angle of the two vectors is 90 degrees, the cosine similarity value is 0; the cosine similarity has a value of-1 when the two vectors point in completely opposite directions. Cosine similarity is usually used in high dimensional space, so its value range is [ -1, 1 ]. For high-dimensional feature vectors, their euclidean distance in the feature space is typically large, and if cosine similarity is used, the angle between them may be small, and thus the similarity is high. In addition, in the fields of texts, images, videos and the like, the feature dimensionality of a researched object is often high, the cosine similarity still keeps the properties of' 1 when the cosine similarity is the same, 0 when the cosine similarity is orthogonal, and-1 when the cosine similarity is opposite under the high-dimensional condition, the value of the Euclidean distance is influenced by the dimensionality, the range is not fixed, and the meaning is fuzzy.
It can be understood that, in the embodiment of the present invention, texture feature extraction is performed on the second B-scan image and the third B-scan image to obtain a second texture feature vector and a third texture feature vector, then for each second texture feature vector, cosine similarity between all third texture feature vectors and the third texture feature vector is calculated, and then similar B-scan images of the second B-scan image are screened out according to the magnitude of the cosine similarity.
As a further optional implementation, the step of determining a second texture feature vector of a second B-scan image specifically includes:
b1, carrying out Gabor wavelet transform on the second B-scan image to obtain a second texture image;
and B2, determining a first mean value and a first variance of the second texture image, and further determining a second texture feature vector according to the first mean value and the first variance.
In particular, the Gabor wavelet is an important texture feature extraction method, and by utilizing the orthogonality of its basis functions, the Gabor wavelet can not only effectively extract texture features, but also eliminate redundant information.
Gabor wavelets can be viewed as a wavelet transform where the mother wavelet is a Gabor function. Assuming that I (x, y) represents a pixel point of the second B-scan image, the two-dimensional discrete Gabor wavelet transform of the image can be expressed as:
Figure 484402DEST_PATH_IMAGE002
wherein, Wmn(x, y) represents a pixel point of the second texture image, x and y represent the position of the pixel in the image, M and N represent the scale and direction of the wavelet transform respectively, the value range of M is 0,1, ⋯, the value range of M, N is 0,1, ⋯, N, gmnRepresenting the Gabor wavelet transform function, x1And y1Mask size variables representing Gabor filters, represent complex conjugate numbers.
The first mean and the first variance may be expressed as:
Figure 298774DEST_PATH_IMAGE003
Figure 183554DEST_PATH_IMAGE004
wherein, mumnDenotes the first mean value, σmnRepresenting a first variance.
The second texture feature vector may be represented as:
TFC=[μ00，σ00，μ01，σ01，⋯，μMN，σMN]
where TFC represents a second texture feature vector.
Similarly, a third texture feature vector may also be obtained based on the above calculation process, which is not described herein again.
Further as an optional implementation manner, the step of screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity specifically includes:
when the cosine similarity is larger than or equal to a preset first threshold value, determining that a third B-scan image corresponding to the third texture feature vector is a similar B-scan image of a second B-scan image corresponding to the second texture feature vector;
or the like, or a combination thereof,
and sequencing the third texture feature vectors according to the cosine similarity to obtain a texture feature vector sequence, sequentially arranging the third texture feature vectors in the texture feature vector sequence from large to small according to the corresponding cosine similarity, and when the sequencing value of the third texture feature vectors is smaller than or equal to a preset second threshold value, determining that the third B-scan image corresponding to the third texture feature vectors is a similar B-scan image of the second B-scan image corresponding to the second texture feature vectors.
Specifically, there are two embodiments for screening similar B-scan images, the first is to determine that the corresponding third B-scan image is a similar B-scan image when the cosine similarity is greater than or equal to a first threshold, and this has an advantage that the similarity of the similar B-scan images can be controlled according to the setting of the first threshold; and the second method is to sort the images from big to small according to the cosine similarity, and select the third B-scan image in the front several positions as the similar B-scan image.
In the embodiment of the invention, a first threshold value is used for screening a third B-scan image with larger cosine similarity, and the first threshold value can be 0.75; the second threshold is used for screening a third B-scan image with a smaller rank value in the texture feature vector sequence, and the second threshold can be 100, 200, 300 and the like; in practical application, a proper screening mode and a proper threshold value can be selected according to specific situations.
And S104, performing label resetting on the second B-scan image according to the B-scan labels of the similar B-scan images, and performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after label resetting to obtain a trained second B-scan classification model.
As a further optional implementation, the step of performing label resetting on the second B-scan image according to the B-scan labels of the similar B-scan images specifically includes:
c1, determining the frequency of occurrence of each type of B-scan label in the B-scan labels of the similar B-scan images;
and C2, selecting the B-scan label with the highest occurrence frequency as an optimization label, and resetting the B-scan label of the second B-scan image as the optimization label.
Specifically, the process of label resetting is a process of label reassignment, in the embodiment of the present invention, for each second B-scan image, a B-scan label with the highest frequency of occurrence in B-scan labels of similar B-scan images is selected as an optimized label, and the label of the second B-scan image is reset to a new optimized label, so that the label resetting of the second B-scan image in the iteration is completed. And when the labels of all the second B-scan images are reset, performing model optimization.
Further as an optional implementation manner, the step of performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after the label is reset to obtain a trained second B-scan classification model specifically includes:
d1, determining a plurality of optimized samples according to the third B-scan image and the second B-scan image after the label is reset, and determining an optimized sample set according to the optimized samples;
d2, inputting the optimized sample set into the first B-scan classification model for training to obtain an optimized B-scan classification model;
d3, classifying the optimized samples according to the optimized B-scan classification model, determining the optimized samples with the classification results inconsistent with the corresponding B-scan labels as fourth B-scan images, and calculating a first ratio of the number of the fourth B-scan images to the total number of the optimized samples;
d4, when the first ratio is larger than or equal to a preset third threshold value, carrying out label resetting on the fourth B-scan image, re-determining an optimized sample set according to a label resetting result, returning and inputting the optimized sample set into the first B-scan classification model for training, and obtaining the optimized B-scan classification model;
and D5, stopping training when the first ratio is smaller than a preset third threshold value, and obtaining a trained second B-scan classification model.
In the embodiment of the invention, the convergence condition of model optimization is that the ratio of the number of the optimized samples with inconsistent classification results and corresponding B-scan labels to the total number of the optimized samples is lower than a preset third threshold (such as 0.1 percent), and when the convergence condition is reached, the optimization can be stopped and a trained second B-scan classification model is output; when the convergence condition is not reached, the label resetting needs to be performed on the optimized sample of which the classification result is inconsistent with the corresponding B-scan label again, and the model optimization needs to be performed again until the convergence condition is reached.
S105, obtaining a plurality of B-scan images to be classified of the second OCT three-dimensional image to be classified, inputting the B-scan images to be classified into a second B-scan classification model to obtain a B-scan classification result of the B-scan images to be classified, and determining a classification result of the second OCT three-dimensional image according to the B-scan classification result.
As a further optional implementation manner, the B-scan classification result includes a normal B-scan picture and a plurality of abnormal B-scan pictures of different abnormal types, and the step of determining the classification result of the second OCT three-dimensional image according to the B-scan classification result specifically includes:
e1, determining the number of the abnormal B-scan pictures according to the B-scan classification result, and calculating a second ratio of the number of the abnormal B-scan pictures to the total number of the B-scan pictures to be classified;
e2, when the second ratio is larger than or equal to a preset fourth threshold, determining that the second OCT three-dimensional image is an abnormal OCT three-dimensional image, and determining the abnormal type of the second OCT three-dimensional image according to the abnormal type with the highest occurrence frequency in the abnormal B-scan picture;
and E3, when the second ratio is smaller than a preset fourth threshold value, determining that the second OCT three-dimensional image is the normal OCT three-dimensional image.
Specifically, because the second OCT three-dimensional image to be classified may include a large number of normal B-scan images and a small number of abnormal B-scan images, if the image type of the second OCT three-dimensional image is determined directly according to the type with the highest occurrence frequency in the classification result of the B-scan images, the second OCT three-dimensional image which originally belongs to an abnormality may be classified as normal, therefore, a fourth threshold (e.g., ten percent) is preset in the embodiment of the present invention to determine whether the second OCT three-dimensional image is abnormal, when the ratio of the number of abnormal B-scan images to the total number of the B-scan images to be classified is greater than or equal to the threshold, the second OCT three-dimensional image may be determined as abnormal, and then the abnormal type of the second OCT three-dimensional image is determined according to the abnormal type with the highest occurrence frequency in the abnormal B-scan images, thereby further improving the accuracy of OCT three-dimensional image classification.
The method steps of the embodiment of the present invention are described above, and the specific implementation flow of the OCT three-dimensional image classification method of the present invention is further described below with reference to a specific embodiment.
As shown in fig. 2, which is a schematic view of a specific flow chart of an OCT three-dimensional image classification method according to an embodiment of the present invention, in the embodiment of the present invention, a first B-scan image is pre-labeled according to an image type of the first OCT three-dimensional image, a training sample set is determined according to the pre-labeled first B-scan image, and the training sample set is input to a capsule neural network to train to obtain a first B-scan classification model; classifying the first B-scan image according to the first B-scan classification model, and respectively determining a third B-scan image and a second B-scan image according to whether the classification result is consistent with the B-scan label; carrying out Gabor wavelet transformation on the third B-scan image and the second B-scan image to obtain a third texture feature vector and a second texture feature vector, and screening out a similar B-scan image of the second B-scan image according to the cosine similarity of the third texture feature vector and the second texture feature vector; performing label resetting on the second B-scan image according to the B-scan labels of the similar B-scan images, performing model optimization on the first B-scan classification model, and iterating to perform label resetting and model optimization until a preset convergence condition is achieved to obtain a second B-scan classification model; and finally, classifying the B-scan image to be classified of the second OCT three-dimensional image according to the second B-scan classification model, and further determining the classification result of the second OCT three-dimensional image according to the classification result of the B-scan image to be classified.
It can be understood that the embodiment of the invention screens out the training sample with the possibly inaccurate label according to the consistency of the model classification result and the B-scan label, and then resets the label according to the label of the similar B-scan image, so that the reset label is more accurate, the accuracy of B-scan image classification is improved, and the accuracy of OCT three-dimensional image classification is further improved; determining a similar B-scan image of the second B-scan image according to the cosine similarity of the texture feature vectors of the two images, so that the similar B-scan image is highly similar to the second B-scan image in texture features, the accuracy of label resetting is improved to a certain extent, and the accuracy of B-scan image classification and OCT three-dimensional image classification is further improved; through the alternate iteration of label resetting and model optimization, the labels and model parameters of the training samples are optimized, and the classification precision of the second B-scan classification model is improved, so that the accuracy of OCT three-dimensional image classification is further improved; in addition, the embodiment of the invention does not need to directly process the high-resolution OCT three-dimensional image, reduces the requirement on the calculation power of the system and improves the efficiency of model training and classification.
Referring to fig. 3, an embodiment of the present invention provides an OCT three-dimensional image classification system, including:
the first B-scan classification model training module is used for acquiring a plurality of first OCT three-dimensional images and corresponding image types, determining a plurality of first B-scan images of the first OCT three-dimensional images, labeling the first B-scan images according to the image types to obtain B-scan labels, and then training according to the first B-scan images and the B-scan labels to obtain a first B-scan classification model;
the first B-scan image classification module is used for performing image classification on the first B-scan image according to the first B-scan classification model, and determining a second B-scan image and a third B-scan image according to the classification result, wherein the second B-scan image is a first B-scan image with the classification result inconsistent with the corresponding B-scan label, and the third B-scan image is a first B-scan image with the classification result consistent with the corresponding B-scan label;
the similar B-scan image determining module is used for determining a second texture feature vector of a second B-scan image and a third texture feature vector of a third B-scan image, determining cosine similarity of the second texture feature vector and the third texture feature vector, and further screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity;
the label resetting and model optimizing module is used for resetting the labels of the second B-scan image according to the B-scan labels of the similar B-scan images, performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after the labels are reset, and obtaining a trained second B-scan classification model;
and the second OCT three-dimensional image classification module is used for acquiring a plurality of B-scan images to be classified of a second OCT three-dimensional image to be classified, inputting the B-scan images to be classified into the second B-scan classification model to obtain a B-scan classification result of the B-scan images to be classified, and further determining the classification result of the second OCT three-dimensional image according to the B-scan classification result.
The contents in the method embodiments are all applicable to the system embodiments, the functions specifically implemented by the system embodiments are the same as those in the method embodiments, and the beneficial effects achieved by the system embodiments are also the same as those achieved by the method embodiments.
Referring to fig. 4, an embodiment of the present invention provides a computer apparatus, including:
at least one processor;
at least one memory for storing at least one program;
when the at least one program is executed by the at least one processor, the at least one program causes the at least one processor to implement the foregoing method of classifying an OCT three-dimensional image.
The contents in the method embodiments are all applicable to the device embodiments, the functions specifically implemented by the device embodiments are the same as those in the method embodiments, and the beneficial effects achieved by the device embodiments are also the same as those achieved by the method embodiments.
Embodiments of the present invention further provide a computer-readable storage medium, in which a program executable by a processor is stored, and the program executable by the processor is used for executing the foregoing OCT three-dimensional image classification method when executed by the processor.
The computer-readable storage medium of the embodiment of the invention can execute the OCT three-dimensional image classification method provided by the embodiment of the method of the invention, can execute any combination implementation steps of the embodiment of the method, and has corresponding functions and beneficial effects of the method.
Embodiments of the present invention also disclose a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and executed by the processor, causing the computer device to perform the method illustrated in fig. 1.
In alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flow charts of the present invention are provided by way of example in order to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed, and in which sub-operations described as part of larger operations are performed independently.
Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise indicated to the contrary, one or more of the functions and/or features described above may be integrated in a single physical device and/or software module, or one or more functions and/or features may be implemented in separate physical devices or software modules. It will also be understood that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be understood within the ordinary skill of an engineer given the nature, function, and interrelationships of the modules. Accordingly, those skilled in the art can, using ordinary skill, practice the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative of and not intended to limit the scope of the invention, which is to be determined from the appended claims along with their full scope of equivalents.
The above-described functions, if implemented in the form of software functional units and sold or used as a separate product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention or a part thereof which substantially contributes to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the above methods of the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Further, the computer readable medium could even be paper or another suitable medium upon which the above described program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the foregoing description of the specification, reference to the description of "one embodiment/example," "another embodiment/example," or "certain embodiments/examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (9)

1. An OCT three-dimensional image classification method is characterized by comprising the following steps:
acquiring a plurality of first OCT three-dimensional images and corresponding image types, determining a plurality of first B-scan images of the first OCT three-dimensional images, labeling the first B-scan images according to the image types to obtain B-scan labels, and training according to the first B-scan images and the B-scan labels to obtain a first B-scan classification model;
performing image classification on the first B-scan image according to the first B-scan classification model, and determining a second B-scan image and a third B-scan image according to a classification result, wherein the second B-scan image is a first B-scan image of which the classification result is inconsistent with a corresponding B-scan label, and the third B-scan image is a first B-scan image of which the classification result is consistent with the corresponding B-scan label;
determining a second texture feature vector of the second B-scan image and a third texture feature vector of the third B-scan image, determining cosine similarity of the second texture feature vector and the third texture feature vector, and screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity;
performing label resetting on the second B-scan image according to the B-scan labels of the similar B-scan images, and performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after label resetting to obtain a trained second B-scan classification model;
acquiring a plurality of B-scan images to be classified of a second OCT three-dimensional image to be classified, inputting the B-scan images to be classified into the second B-scan classification model to obtain a B-scan classification result of the B-scan images to be classified, and further determining a classification result of the second OCT three-dimensional image according to the B-scan classification result;
the step of performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after label resetting to obtain a trained second B-scan classification model specifically comprises the following steps:
determining a plurality of optimized samples according to the third B-scan image and the second B-scan image after the label is reset, and determining an optimized sample set according to the optimized samples;
inputting the optimized sample set into the first B-scan classification model for training to obtain an optimized B-scan classification model;
classifying the optimized samples according to the optimized B-scan classification model, determining the optimized samples with the classification results inconsistent with the corresponding B-scan labels as fourth B-scan images, and calculating a first ratio of the number of the fourth B-scan images to the total number of the optimized samples;
when the first ratio is larger than or equal to a preset third threshold, carrying out label resetting on the fourth B-scan image, re-determining an optimized sample set according to a label resetting result, returning and inputting the optimized sample set into the first B-scan classification model for training, and obtaining an optimized B-scan classification model;
and when the first ratio is smaller than a preset third threshold, stopping training to obtain a trained second B-scan classification model.
2. The OCT three-dimensional image classification method of claim 1, wherein the step of training the first B-scan image and the B-scan label to obtain the first B-scan classification model specifically comprises:
determining a plurality of training samples according to the first B-scan image and the B-scan label, and determining a training sample set according to the training samples;
inputting the training sample set into a pre-constructed capsule neural network for training to obtain the first B-scan classification model;
the capsule neural network comprises a convolution layer, a main capsule layer, a digital capsule layer and a full connection layer, wherein the convolution layer is used for performing convolution processing on the training samples to obtain a plurality of sample characteristic subgraphs, the main capsule layer is used for determining a plurality of vector neurons according to the sample characteristic subgraphs, the digital capsule layer is used for calculating the activation vectors of the vector neurons through a dynamic routing algorithm, and the full connection layer is used for determining the classification results of the training samples according to the modular length of the activation vectors.
3. The OCT three-dimensional image classification method of claim 1, wherein said step of determining said second texture feature vector of said second B-scan image comprises:
carrying out Gabor wavelet transform on the second B-scan image to obtain a second texture image;
and determining a first mean value and a first variance of the second texture image, and further determining the second texture feature vector according to the first mean value and the first variance.
4. The OCT three-dimensional image classification method according to claim 1, wherein the step of selecting a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity specifically comprises:
when the cosine similarity is larger than or equal to a preset first threshold value, determining that a third B-scan image corresponding to the third texture feature vector is a similar B-scan image of a second B-scan image corresponding to the second texture feature vector;
or the like, or a combination thereof,
and sequencing the third texture feature vectors according to the cosine similarity to obtain a texture feature vector sequence, wherein the third texture feature vectors in the texture feature vector sequence are sequentially arranged from large to small according to the corresponding cosine similarity, and when the sequence value of the third texture feature vectors is smaller than or equal to a preset second threshold, determining that the third B-scan image corresponding to the third texture feature vectors is a similar B-scan image of the second B-scan image corresponding to the second texture feature vectors.
5. The OCT three-dimensional image classification method according to claim 1, wherein the step of performing a label resetting on the second B-scan image according to the B-scan label of the similar B-scan image specifically comprises:
determining the frequency of occurrence of each type of B-scan tag among B-scan tags of the similar B-scan images;
and selecting the B-scan label with the highest frequency of occurrence as an optimization label, and resetting the B-scan label of the second B-scan image as the optimization label.
6. The OCT three-dimensional image classification method according to any one of claims 1 to 5, wherein the B-scan classification result comprises a normal B-scan picture and a plurality of abnormal B-scan pictures of different abnormal types, and the step of determining the classification result of the second OCT three-dimensional image according to the B-scan classification result specifically comprises:
determining the number of the abnormal B-scan pictures according to the B-scan classification result, and calculating a second ratio of the number of the abnormal B-scan pictures to the total number of the B-scan pictures to be classified;
when the second ratio is larger than or equal to a preset fourth threshold, determining that the second OCT three-dimensional image is an abnormal OCT three-dimensional image, and determining the abnormal type of the second OCT three-dimensional image according to the abnormal type with the highest occurrence frequency in the abnormal B-scan picture;
and when the second ratio is smaller than a preset fourth threshold, determining that the second OCT three-dimensional image is a normal OCT three-dimensional image.
7. An OCT three-dimensional image classification system, comprising:
the first B-scan classification model training module is used for acquiring a plurality of first OCT three-dimensional images and corresponding image types, determining a plurality of first B-scan images of the first OCT three-dimensional images, labeling the first B-scan images according to the image types to obtain B-scan labels, and then training according to the first B-scan images and the B-scan labels to obtain a first B-scan classification model;
the first B-scan image classification module is used for performing image classification on the first B-scan image according to the first B-scan classification model, and determining a second B-scan image and a third B-scan image according to a classification result, wherein the second B-scan image is a first B-scan image with a classification result inconsistent with a corresponding B-scan label, and the third B-scan image is a first B-scan image with a classification result consistent with a corresponding B-scan label;
the similar B-scan image determining module is used for determining a second texture feature vector of the second B-scan image and a third texture feature vector of the third B-scan image, determining cosine similarity of the second texture feature vector and the third texture feature vector, and further screening a plurality of similar B-scan images of the second B-scan image from the third B-scan image according to the cosine similarity;
the label resetting and model optimizing module is used for resetting the label of the second B-scan image according to the B-scan label of the similar B-scan image, and performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after label resetting to obtain a trained second B-scan classification model;
the second OCT three-dimensional image classification module is used for acquiring a plurality of B-scan images to be classified of a second OCT three-dimensional image to be classified, inputting the B-scan images to be classified into the second B-scan classification model to obtain a B-scan classification result of the B-scan images to be classified, and further determining a classification result of the second OCT three-dimensional image according to the B-scan classification result;
the step of performing model optimization on the first B-scan classification model according to the third B-scan image and the second B-scan image after the label is reset to obtain a trained second B-scan classification model specifically comprises the following steps:
determining a plurality of optimized samples according to the third B-scan image and the second B-scan image after the label is reset, and determining an optimized sample set according to the optimized samples;
inputting the optimized sample set into the first B-scan classification model for training to obtain an optimized B-scan classification model;
classifying the optimized samples according to the optimized B-scan classification model, determining the optimized samples with the classification results inconsistent with the corresponding B-scan labels as fourth B-scan images, and calculating a first ratio of the number of the fourth B-scan images to the total number of the optimized samples;
when the first ratio is larger than or equal to a preset third threshold, carrying out label resetting on the fourth B-scan image, re-determining an optimized sample set according to a label resetting result, returning and inputting the optimized sample set into the first B-scan classification model for training, and obtaining an optimized B-scan classification model;
and when the first ratio is smaller than a preset third threshold value, stopping training to obtain a trained second B-scan classification model.
8. A computer apparatus, comprising:
at least one processor;
at least one memory for storing at least one program;
when executed by the at least one processor, cause the at least one processor to implement a method of OCT three-dimensional image classification as claimed in any one of claims 1 to 6.
9. A computer-readable storage medium in which a processor-executable program is stored, wherein the processor-executable program, when executed by a processor, is adapted to perform an OCT three-dimensional image classification method according to any one of claims 1 to 6.
CN202210407731.2A 2022-04-19 2022-04-19 OCT three-dimensional image classification method, system, computer device and storage medium Active CN114529772B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210407731.2A CN114529772B (en) 2022-04-19 2022-04-19 OCT three-dimensional image classification method, system, computer device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210407731.2A CN114529772B (en) 2022-04-19 2022-04-19 OCT three-dimensional image classification method, system, computer device and storage medium

Publications (2)

Publication Number Publication Date
CN114529772A CN114529772A (en) 2022-05-24
CN114529772B true CN114529772B (en) 2022-07-15

Family

ID=81628009

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210407731.2A Active CN114529772B (en) 2022-04-19 2022-04-19 OCT three-dimensional image classification method, system, computer device and storage medium

Country Status (1)

Country Link
CN (1) CN114529772B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102750289A (en) * 2011-04-19 2012-10-24 富士通株式会社 Tag group classifying method and equipment as well as data mixing method and equipment
CN103150574A (en) * 2013-01-05 2013-06-12 南京邮电大学 Image spam detection method based on nearest tag propagation algorithm
CN109583297A (en) * 2018-10-25 2019-04-05 清华大学 Retina OCT volume data identification method and device
EP3617943A1 (en) * 2018-08-30 2020-03-04 Topcon Corporation Multivariate and multi-resolution retinal image anomaly detection system
CN113076963A (en) * 2021-06-07 2021-07-06 腾讯科技（深圳）有限公司 Image recognition method and device and computer readable storage medium

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7229881B2 (en) * 2018-08-14 2023-02-28 キヤノン株式会社 MEDICAL IMAGE PROCESSING APPARATUS, TRAINED MODEL, MEDICAL IMAGE PROCESSING METHOD AND PROGRAM
CN112102315B (en) * 2020-11-02 2021-02-19 腾讯科技（深圳）有限公司 Medical image processing method, medical image processing device, computer equipment and storage medium

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102750289A (en) * 2011-04-19 2012-10-24 富士通株式会社 Tag group classifying method and equipment as well as data mixing method and equipment
CN103150574A (en) * 2013-01-05 2013-06-12 南京邮电大学 Image spam detection method based on nearest tag propagation algorithm
EP3617943A1 (en) * 2018-08-30 2020-03-04 Topcon Corporation Multivariate and multi-resolution retinal image anomaly detection system
CN109583297A (en) * 2018-10-25 2019-04-05 清华大学 Retina OCT volume data identification method and device
CN113076963A (en) * 2021-06-07 2021-07-06 腾讯科技（深圳）有限公司 Image recognition method and device and computer readable storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
面向OCT影像的易损斑块自动识别方法;冯子健 等;《电子世界》;20190430(第4期);第68-70页 *

Also Published As

Publication number Publication date
CN114529772A (en) 2022-05-24

Similar Documents

Publication Publication Date Title
Huang et al. Surface defect saliency of magnetic tile
Tabernik et al. Segmentation-based deep-learning approach for surface-defect detection
Islam et al. Revisiting salient object detection: Simultaneous detection, ranking, and subitizing of multiple salient objects
Simonovsky et al. Dynamic edge-conditioned filters in convolutional neural networks on graphs
He et al. Supercnn: A superpixelwise convolutional neural network for salient object detection
CN107103326B (en) Collaborative significance detection method based on super-pixel clustering
US7406200B1 (en) Method and system for finding structures in multi-dimensional spaces using image-guided clustering
Robinson Sketch2code: Generating a website from a paper mockup
Ali et al. A hybrid geometric spatial image representation for scene classification
Liu et al. Evaluation of LBP and deep texture descriptors with a new robustness benchmark
Metre et al. An overview of the research on texture based plant leaf classification
WO2014205231A1 (en) Deep learning framework for generic object detection
CN113272827A (en) Validation of classification decisions in convolutional neural networks
Andrearczyk et al. Texture segmentation with fully convolutional networks
Neena et al. Image classification using an ensemble-based deep CNN
Goswami Impact of deep learning in image processing and computer vision
CN111144425B (en) Method and device for detecting shot screen picture, electronic equipment and storage medium
CN113449784A (en) Image multi-classification method, device, equipment and medium based on prior attribute map
Avola et al. Real-time deep learning method for automated detection and localization of structural defects in manufactured products
Cosentino et al. The geometry of self-supervised learning models and its impact on transfer learning
Rele et al. Machine Learning based Brain Tumor Detection using Transfer Learning
Ghosh et al. Automated defective pin detection for recycled microelectronics identification
Abdallah et al. Facial-expression recognition based on a low-dimensional temporal feature space
CN114529772B (en) OCT three-dimensional image classification method, system, computer device and storage medium
KR102178238B1 (en) Apparatus and method of defect classification using rotating kernel based on machine-learning

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
PE01 Entry into force of the registration of the contract for pledge of patent right
PE01 Entry into force of the registration of the contract for pledge of patent right

Denomination of invention: OCT 3D image classification method, system, computer device and storage medium

Effective date of registration: 20230227

Granted publication date: 20220715

Pledgee: Foshan rural commercial bank Limited by Share Ltd.

Pledgor: GUANGDONG WEIREN MEDICAL TECHNOLOGY Co.,Ltd.

Registration number: Y2023980033570