CN112541906B - Data processing method and device, electronic equipment and storage medium - Google Patents
Data processing method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN112541906B CN112541906B CN202011498021.2A CN202011498021A CN112541906B CN 112541906 B CN112541906 B CN 112541906B CN 202011498021 A CN202011498021 A CN 202011498021A CN 112541906 B CN112541906 B CN 112541906B
- Authority
- CN
- China
- Prior art keywords
- label
- focus
- lesion
- fundus image
- image data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003860 storage Methods 0.000 title claims abstract description 12
- 238000003672 processing method Methods 0.000 title abstract description 7
- 238000013499 data model Methods 0.000 claims abstract description 109
- 238000000034 method Methods 0.000 claims abstract description 72
- 238000009499 grossing Methods 0.000 claims abstract description 39
- 230000008569 process Effects 0.000 claims abstract description 27
- 230000003902 lesion Effects 0.000 claims description 124
- 238000012545 processing Methods 0.000 claims description 63
- 230000006870 function Effects 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 16
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000002372 labelling Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 abstract description 6
- 230000001629 suppression Effects 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 10
- 201000010099 disease Diseases 0.000 description 4
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 208000032843 Hemorrhage Diseases 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 201000011066 hemangioma Diseases 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 208000010412 Glaucoma Diseases 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 208000030533 eye disease Diseases 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Abstract
The embodiment of the invention provides a data processing method and device, electronic equipment and a storage medium, wherein the method comprises the following steps: presetting a plurality of focus labels and a label smooth value corresponding to each focus label in a data model; acquiring fundus image data; and calling the data model to process the fundus image data so as to determine a plurality of target focus labels corresponding to the fundus image data according to the label smoothing value. According to the embodiment of the invention, the application optimization of the multi-label data model is realized, and the plurality of target focus labels corresponding to the fundus image data are determined according to the preset label smooth value corresponding to each focus label, so that the reality and uncertainty can be considered, the application of the multi-label data model is more consistent with the actual situation, and the noise suppression effect is achieved.
Description
Technical Field
The present invention relates to the field of image technologies, and in particular, to a method and an apparatus for data processing, an electronic device, and a storage medium.
Background
With the development of medical imaging, a technology based on deep learning has been increasingly used for medical image diagnosis, and is a trend, such as lesion determination with respect to a fundus image.
In the prior art, because the judgment of the focus is not absolute, on one hand, the judgment results of a plurality of doctors on the same fundus map are different, on the other hand, a certain focus is uncertain, and only a certain probability is possibly a certain disease, and when a training set for deep learning is labeled, more labeled focuses, fewer labeled focuses or a label is wrongly labeled, so that the problem of relatively prominent and common noise in medical science is a problem which needs to be solved urgently at present.
Disclosure of Invention
In view of the above problems, it is proposed to provide a method and apparatus for data processing, an electronic device, and a storage medium, which overcome the above problems or at least partially solve the above problems, including:
a method of data processing, the method comprising:
presetting a plurality of focus labels and a label smooth value corresponding to each focus label in a data model;
acquiring fundus image data;
and calling the data model to process the fundus image data so as to determine a plurality of target focus labels corresponding to the fundus image data according to the label smoothing value.
Optionally, the data model determines the plurality of target lesion labels in the following manner:
aiming at each focus label, performing two-classification processing on the fundus image data by combining the label smooth value to obtain two-classification processing results;
and according to the two classification processing results, determining a plurality of target focus labels corresponding to the fundus image data from the plurality of focus labels.
Optionally, the two classification processing results include a lesion probability value, and the determining a plurality of target lesion labels corresponding to the fundus image data from the plurality of lesion labels according to the two classification processing results includes:
judging whether the lesion probability value meets a preset condition or not according to each lesion label;
and when the focus probability value is judged to meet the preset condition, determining the focus label as a target focus label corresponding to the fundus image data.
Optionally, the preset condition includes:
the lesion probability value is greater than a preset probability value corresponding to the lesion label.
Optionally, before the invoking the data model to process the fundus image data to determine a plurality of target lesion labels corresponding to the fundus image data according to the label smoothing value, the method further includes:
and adjusting the loss function of the data model by adopting the label smoothing value.
Optionally, the data model is a neural network-based data model.
Optionally, each lesion label corresponds to a lesion category.
An apparatus for data processing, the apparatus comprising:
the label smooth value presetting module is used for presetting a plurality of focus labels and a label smooth value corresponding to each focus label in the data model;
the fundus image data acquisition module is used for acquiring fundus image data;
and the target lesion label determination modules are used for calling the data model to process the fundus image data so as to determine a plurality of target lesion labels corresponding to the fundus image data according to the label smoothing value.
An electronic device comprising a processor, a memory and a computer program stored on the memory and capable of running on the processor, the computer program, when executed by the processor, implementing a method of data processing as described above.
A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the method of data processing as described above.
The embodiment of the invention has the following advantages:
in the embodiment of the invention, a plurality of focus labels and label smooth values corresponding to each focus label are preset in a data model, fundus image data is obtained, the data model is called to process the fundus image data, so that a plurality of target focus labels corresponding to the fundus image data are determined according to the label smooth values, application optimization of a multi-label data model is realized, and a plurality of target focus labels corresponding to the fundus image data are determined according to the preset label smooth values corresponding to each focus label, so that the reality and the uncertainty can be considered, the application of the multi-label data model is more in line with the actual situation, and the noise suppression effect is achieved.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the description of the present invention will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
FIG. 1 is a flow chart illustrating steps of a method for data processing according to an embodiment of the present invention;
FIG. 2 is a flow chart illustrating steps of another method for data processing according to an embodiment of the present invention;
FIG. 3 is a flow chart of steps in another method of data processing according to an embodiment of the invention;
fig. 4 is a schematic structural diagram of a data processing apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, the present invention is described in detail with reference to the accompanying drawings and the detailed description thereof. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without inventive step based on the embodiments of the present invention, are within the scope of protection of the present invention.
Referring to fig. 1, a flowchart illustrating steps of a data processing method according to an embodiment of the present invention is shown, which may specifically include the following steps:
the data model may be a data model based on a neural network, for example, a multi-layer convolutional neural network may be used as an extractor of features, and a classifier may be combined to determine a disease included in the fundus image data through the data model.
As an example, each lesion label may correspond to a lesion category, and a label smoothing value corresponding to each lesion label may be preset, and the label smoothing value may be used to control the smoothing strength of the lesion label, for example, the label smoothing value may be represented as α k (k may be the lesion category to which the lesion label corresponds).
In specific implementation, a plurality of focus labels and a label smoothing value corresponding to each focus label can be preset in the data model, and then when the fundus image data is processed, an output result of the data model can be obtained according to the label smoothing values.
In one example, since there are few cases where a plurality of lesions simultaneously appear in fundus image data, or when there are a plurality of lesions in fundus image data, only one more serious lesion needs to be output, for such a case, a classification method of a single lesion may be employed, so that a lesion class with the highest probability is taken as an output result for a plurality of lesion classes.
For the classification method of a single lesion, if the output of the penultimate layer of the neural network in the data model is x and the weight of the last layer is w, the prediction of the neural network can be represented as follows:
wherein p is k The probability of class k in the data model; x can be the output result of the penultimate layer of the neural network in the data model; w may be the weight value of the last layer of the neural network in the data model.
Furthermore, under the classification method of single lesions, the loss function can be expressed as follows:
wherein p is k The probability of class k in the data model; y is k =1, i.e. if the k-th class is the correct lesion category, the lesion label value may be 1, while the remaining lesion categories have a lesion label value of 0.
In practical application, because the problem that the judgment of the focus condition in the fundus image data is approximate to a single focus is not strict enough, a multi-class and multi-label multi-focus classification output mode can be adopted for the condition that multiple focuses exist.
Because the multi-label can have various characteristics for one data, that is, a plurality of focus labels can be correspondingly output, the multi-classification data model of the multi-focus can identify one data and output results representing that the data belongs to various focus categories, for example, a plurality of disease types can be reflected in a medical image, for example, a plurality of eye diseases such as hemangioma, fundus hemorrhage, glaucoma and the like can be reflected through fundus image data, and when the fundus image data is identified and processed by adopting the multi-classification data model, the output results can correspond to two focus categories such as hemangioma and fundus hemorrhage for the fundus image data.
Aiming at the classification method of multiple focuses, if the output of the last layer of the reciprocal of the neural network in the data model is x, the focus label y k Can be expressed as follows:
y k =Sigmoid(x k )
the Sigmoid function form can be expressed as follows, and the value range is (0, 1):
y=1/(1+e -x )
furthermore, under the classification method of multiple lesions, the loss function can be expressed as follows:
compared with a single classification data model of a single label, the single classification data model only can select the focus class with the largest probability value to output, and the probability output aiming at each focus class can be influenced by other focus classes, so that the probability value of the focus class with the large output probability is amplified, the probability values of other focus classes are further ignored, and aiming at a multi-label multi-classification data model, the multi-label can be regarded as a plurality of two-classification problems, namely, whether the multi-label belongs to the focus class corresponding to the current focus label or not is judged aiming at each focus class, and different focus classes can be independent of each other.
In another example, since the determination of the focus is not absolute, on one hand, the determination results of multiple doctors on the same fundus image may be different, and on the other hand, there is a case where a certain focus is uncertain and only has a certain probability of being a certain disease, and when labeling the training set of deep learning, there are multiple labeled focuses, few labeled focuses, or a label labeling error, then the algorithm in the multi-label multi-classification data model may be improved accordingly for the existing noise problem.
For the algorithm in the multi-label multi-classification data model, the correct lesion class is set to be y =1, the incorrect lesion class is set to be y =0, and there may be a plurality of lesion classes y =1, but from the viewpoint of label noise, the above method for setting the lesion label is not practical because the probability of the lesion label is actually not 1 when a certain lesion in certain fundus image data cannot be completely determined.
Furthermore, a Label of a Lesion (LS) may be processed as follows when the Label is used in a method for classifying a single lesion to suppress noise:
if the α is larger, the probability of the largest focus label is smaller, and the probability of the smallest focus label is larger, that is, the difference between different focus labels can be reduced, thereby increasing uncertainty.
Therefore, the anti-noise algorithm of Label Smooth can be applied to a multi-Label multi-classification data model, and the output result of the data model can be obtained according to the Label Smooth value when the fundus image data is processed by presetting a plurality of focus labels and the Label Smooth value corresponding to each focus Label.
in a specific implementation, fundus image data may be acquired, the fundus image data may be further processed by using a multi-label multi-classification data model, and an output result of the data model may be obtained according to a label smoothing value, for example, the fundus image data may be a color image, a single-channel grayscale image, or a binary image.
After acquiring the fundus image data, the data model can be called to process the fundus image data, and then a plurality of target focus labels corresponding to the fundus image data can be determined according to the label smoothing value to be used as an output result of the data model for the fundus image data.
Specifically, when fundus image data is processed by the multi-label multi-classification data model, a plurality of lesion classes can be output according to the label smoothing value, and lesion labels corresponding to the output lesion classes can be used as a plurality of target lesion labels corresponding to the fundus image data.
In the embodiment of the invention, a plurality of focus labels and label smooth values corresponding to each focus label are preset in a data model, fundus image data is obtained, the data model is called to process the fundus image data, so that a plurality of target focus labels corresponding to the fundus image data are determined according to the label smooth values, application optimization of a multi-label data model is realized, and a plurality of target focus labels corresponding to the fundus image data are determined according to the preset label smooth values corresponding to each focus label, so that the reality and the uncertainty can be considered, the application of the multi-label data model is more in line with the actual situation, and the noise suppression effect is achieved.
Referring to fig. 2, a flowchart illustrating steps of another data processing method according to an embodiment of the present invention is shown, which may specifically include the following steps:
in a specific implementation, a plurality of focus labels and a label smoothing value corresponding to each focus label may be preset in the data model, so that when the fundus image data is processed, an output result of the data model may be obtained according to the label smoothing value.
in a specific implementation, fundus image data may be acquired to further process the fundus image data using a multi-label multi-classification data model, and an output of the data model is obtained based on the label smoothing values.
after fundus image data are obtained, a data model can be called to process the fundus image data, and then the fundus image data can be subjected to two-classification processing by combining a label smooth value aiming at each focus label to obtain two-classification processing results.
Specifically, when the Label Smooths (LS) method is adopted for the classification data model of a single lesion, α acts on all lesion labels, and when the Label Smooths (LS) method is adopted for the classification data model of multiple lesions with multiple labels, because the problem is converted into multiple binary problems, that is, if the original N lesion categories are present, the problem can be converted into N binary problems, and the Label smooths can act on each binary problem respectively, that is, whether the problem is a certain lesion category can be judged, and further, for each lesion Label, the Label smoothing value can be combined to perform binary processing on the fundus image data, so as to obtain a secondary classification processing result.
In an example, for a plurality of lesion labels, a Label smoothing value corresponding to each lesion Label may be preset, that is, for each lesion Label, a Label smoothing value may apply a Label Smooth to a binary problem that the lesion Label is or is not; the same label smoothing value may also be used for multiple lesion labels.
After the result of the two-classification processing is obtained, a plurality of target lesion labels corresponding to the fundus image data may be determined from the plurality of lesion labels as an output result of the data model for the fundus image data, based on the result of the two-classification processing.
By applying the Label smooths to the multi-Label multi-lesion classification data model, on one hand, the problem of lesion judgment of the fundus image data is considered to be more in line with the actual situation as the multi-Label problem, and the multi-Label problem is also a method considering the reality and uncertainty, so that the application of the Label smooths is more reasonable on the basis.
On the other hand, each focus category in the multi-Label multi-focus classification data model can be decoupled, namely the output of each focus category can avoid mutual influence, after the Label Smooth is applied, a Label Smooth value can be respectively set according to each focus category, and then a plurality of target focus labels corresponding to the fundus image data can be determined according to the Label Smooth value to serve as the output result of the data model.
In the embodiment of the invention, a plurality of focus labels and a label smooth value corresponding to each focus label are preset in a data model, fundus image data are obtained, the data model is called to process fundus image data, two classification processing is carried out on the fundus image data by combining the label smooth value aiming at each focus label to obtain two classification processing results, a plurality of target focus labels corresponding to the fundus image data are determined from the focus labels according to the two classification processing results, application optimization of a multi-label data model is realized, and the target focus labels corresponding to the fundus image data are determined according to the preset label smooth value corresponding to each focus label, so that the actual and uncertainty can be considered, the application of the multi-label data model is more consistent with the actual situation, and the noise suppression effect is achieved.
Referring to fig. 3, a flowchart illustrating steps of another data processing method according to an embodiment of the present invention is shown, which may specifically include the following steps:
in a specific implementation, a plurality of focus labels and a label smoothing value corresponding to each focus label may be preset in the data model, so that when the fundus image data is processed, an output result of the data model may be obtained according to the label smoothing value.
in a specific implementation, fundus image data may be acquired to further process the fundus image data using a multi-label multi-classification data model, and an output of the data model is obtained based on the label smoothing values.
in practical application, the loss function of the data model may be adjusted by using a preset label smoothing value, and may be adjusted as follows:
wherein, when the multi-focus classification data model with multiple labels adopts a Label Smooth (LS) method, the focus labels can be replaced by
after fundus image data are obtained, a data model can be called to process the fundus image data, and then two classification processing can be carried out on the fundus image data according to each focus label and the label smooth value, so that two classification processing results are obtained.
When the Label Smooth (LS) method is adopted in the multi-labeled multi-lesion classification data model, the lesions can be labeled in the following wayAnd (3) processing:
where k =0or 1.
305, judging whether the lesion probability value meets a preset condition or not according to each lesion label, wherein the two classification processing results comprise lesion probability values;
after the two classification processing results are obtained, since the two classification processing results may include a lesion probability value, it may be determined whether the lesion probability value satisfies a preset condition for each lesion label, where the preset condition may include that the lesion probability value is greater than a preset probability value corresponding to the lesion label.
For example, a threshold may be preset for each lesion label as a preset probability value corresponding to the lesion label, and a judgment condition may be preset such that the lesion probability value is greater than the preset probability value corresponding to the lesion label, and then, after performing two classification processes on the fundus image data in combination with the label smoothing value for each lesion label, the lesion probability values of the obtained plurality of lesion labels may be judged to determine whether the lesion probability value of each lesion label is greater than the preset probability value corresponding to the lesion label.
And step 306, when the lesion probability value is judged to meet the preset condition, determining that the lesion label is a target lesion label corresponding to the fundus image data.
In a specific implementation, when it is determined that the lesion probability value satisfies the preset condition, the lesion label may be determined as a target lesion label corresponding to the fundus image data, so as to serve as an output result of the data model for the fundus image data.
For example, after two classification processes are performed on fundus image data by combining a label smooth value for each focus label, focus probability values of a plurality of focus labels can be obtained, then, a judgment can be performed on the focus probability values of the plurality of focus labels, for each focus label, when the focus probability value is greater than a preset probability value, the focus label can be judged as a focus label to be output by the data model, and then, a plurality of focus labels to be output for fundus image data can be obtained from the plurality of focus labels, that is, focus labels corresponding to a plurality of focus categories screened out can be used as an output result of the data model.
In the embodiment of the invention, a plurality of focus labels and a label smooth value corresponding to each focus label are preset in a data model, fundus image data are obtained, the label smooth value is adopted, a loss function of the data model is adjusted, the data model is called to process the fundus image data, the fundus image data are subjected to two-classification processing aiming at each focus label and the label smooth value, two-classification processing results are obtained, the two-classification processing results comprise focus probability values, whether the focus probability values meet preset conditions or not is judged aiming at each focus label, when the focus probability values meet the preset conditions is judged, the focus labels are determined to be target focus labels corresponding to the fundus image data, application optimization of the multi-label data model is realized, the plurality of target focus labels corresponding to the fundus image data are determined according to the preset label smooth value corresponding to each focus label, actual sum can be considered, the application of the multi-label data model is more consistent with actual conditions, and the noise suppression effect is achieved.
It should be noted that, for simplicity of description, the method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present invention is not limited by the illustrated order of acts, as some steps may occur in other orders or concurrently in accordance with the embodiments of the present invention. Further, those of skill in the art will appreciate that the embodiments described in the specification are presently preferred and that no particular act is required to implement the embodiments of the invention.
Referring to fig. 4, a schematic structural diagram of a data processing apparatus according to an embodiment of the present invention is shown, which may specifically include the following modules:
a label smooth value presetting module 401, configured to preset a plurality of lesion labels and a label smooth value corresponding to each lesion label in the data model;
a fundus image data acquisition module 402 for acquiring fundus image data;
a multiple target lesion label determining module 403, configured to invoke the data model to process the fundus image data, so as to determine multiple target lesion labels corresponding to the fundus image data according to the label smoothing value.
In an embodiment of the present invention, the data model determines the plurality of target lesion labels as follows:
for each focus label, performing two-classification processing on the fundus image data by combining the label smoothing value to obtain two-classification processing results;
and according to the two classification processing results, determining a plurality of target lesion labels corresponding to the fundus image data from the plurality of lesion labels.
In an embodiment of the present invention, the determining, according to the two classification processing results, a plurality of target lesion labels corresponding to the fundus image data from the plurality of lesion labels includes:
judging whether the lesion probability value meets a preset condition or not according to each lesion label;
and when the focus probability value is judged to meet the preset condition, determining the focus label as a target focus label corresponding to the fundus image data.
In an embodiment of the present invention, the preset conditions include:
the lesion probability value is greater than a preset probability value corresponding to the lesion label.
In an embodiment of the present invention, the method further includes:
and the loss function adjusting module is used for adjusting the loss function of the data model by adopting the label smooth value.
In an embodiment of the invention, the data model is a data model based on a neural network.
In one embodiment of the present invention, each lesion label corresponds to a lesion category.
In the embodiment of the invention, a plurality of focus labels and label smooth values corresponding to each focus label are preset in a data model, fundus image data are obtained, the data model is called to process fundus image data, a plurality of target focus labels corresponding to the fundus image data are determined according to the label smooth values, application optimization of a multi-label data model is realized, and a plurality of target focus labels corresponding to the fundus image data are determined according to the preset label smooth values corresponding to each focus label, so that the reality and uncertainty can be considered, the application of the multi-label data model is more consistent with the actual situation, and the noise suppression effect is achieved.
An embodiment of the present invention also provides an electronic device, which may include a processor, a memory, and a computer program stored in the memory and capable of running on the processor, and when executed by the processor, the computer program implements the method for processing data as above.
An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the above data processing method.
For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.
The embodiments in the present specification are all described in a progressive manner, and each embodiment focuses on differences from other embodiments, and portions that are the same and similar between the embodiments may be referred to each other.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, apparatus, or computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.
Embodiments of the present invention are described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing terminal to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing terminal to cause a series of operational steps to be performed on the computer or other programmable terminal to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications of these embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the embodiments of the invention.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrases "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or terminal device that comprises the element.
The method and apparatus for data processing, the electronic device, and the storage medium provided above are introduced in detail, and a specific example is applied in this document to explain the principle and the implementation of the present invention, and the description of the above embodiment is only used to help understand the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (8)
1. A method of data processing, the method comprising:
presetting a plurality of focus labels and a label smooth value corresponding to each focus label in a data model; the plurality of lesion labels are a plurality of features for one datum; an output layer in the data model adopts a Sigmoid function, and the data model applies Label Smooth;
acquiring fundus image data;
calling the data model to process the fundus image data so as to determine a plurality of target focus labels corresponding to the fundus image data according to the label smoothing value;
wherein k =0or 1; alpha (alpha) ("alpha") k Labeling each lesion with y k A corresponding label smoothing value for controlling a smoothing intensity of the lesion label;
wherein the data model determines the plurality of target lesion labels as follows:
for each focus label, performing two-classification processing on the fundus image data by combining the label smooth value to obtain two-classification processing results, wherein the two-classification processing results comprise focus probability values;
according to the two classification processing results, a plurality of target focus labels corresponding to the fundus image data are determined from the plurality of focus labels, and the target focus labels comprise: judging whether the lesion probability value meets a preset condition or not according to each lesion label; when the focus probability value is judged to meet a preset condition, determining the focus label as a target focus label corresponding to the fundus image data;
wherein the output of the last layer of the neural network reciprocal in the data model is x, and the focus label y k Expressed in the following way:
y k =Sigmoid(x k )
the Sigmoid function form can be expressed as follows, and the value range is (0, 1):
y=1/(1+e -x )。
2. the method according to claim 1, wherein the preset condition comprises:
the lesion probability value is greater than a preset probability value corresponding to the lesion label.
3. The method of claim 1, further comprising, prior to said invoking the data model to process the fundus image data to determine a plurality of target lesion labels corresponding to the fundus image data based on the label smoothing value:
and adjusting the loss function of the data model by adopting the label smoothing value.
4. The method of claim 1, wherein the data model is a neural network-based data model.
5. The method of claim 1, wherein each lesion label corresponds to a lesion category.
6. An apparatus for data processing, the apparatus comprising:
the label smooth value presetting module is used for presetting a plurality of focus labels and a label smooth value corresponding to each focus label in the data model; the plurality of lesion labels are a plurality of features for one datum; an output layer in the data model adopts a Sigmoid function, and the data model applies Label Smooth;
the fundus image data acquisition module is used for acquiring fundus image data;
a plurality of target lesion label determination modules, configured to invoke the data model to process the fundus image data, so as to determine a plurality of target lesion labels corresponding to the fundus image data according to the label smoothing value;
wherein k =0or 1; alpha (alpha) ("alpha") k Labeling each lesion with y k A corresponding label smoothing value, the label smoothing value being used to control the smoothing strength of the lesion label;
the apparatus is further configured to: determining the plurality of target lesion labels as follows:
for each focus label, performing two-classification processing on the fundus image data by combining the label smooth value to obtain two-classification processing results, wherein the two-classification processing results comprise focus probability values;
according to the two classification processing results, a plurality of target lesion labels corresponding to the fundus image data are determined from the plurality of lesion labels, and the target lesion labels include: judging whether the lesion probability value meets a preset condition or not according to each lesion label; when the focus probability value is judged to meet a preset condition, determining the focus label as a target focus label corresponding to the fundus image data;
the output of the last layer of the neural network reciprocal in the data model is x, and the focus label y k Expressed in the following way:
y k =Sigmoid(x k )
the Sigmoid function form can be expressed as follows, and its value range is (0, 1):
y=1/(1+e -x )。
7. an electronic device comprising a processor, a memory and a computer program stored on the memory and capable of running on the processor, the computer program, when executed by the processor, implementing a method of data processing according to any one of claims 1 to 5.
8. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out a method of data processing according to any one of claims 1 to 5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011498021.2A CN112541906B (en) | 2020-12-17 | 2020-12-17 | Data processing method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011498021.2A CN112541906B (en) | 2020-12-17 | 2020-12-17 | Data processing method and device, electronic equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112541906A CN112541906A (en) | 2021-03-23 |
CN112541906B true CN112541906B (en) | 2022-10-25 |
Family
ID=75018781
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011498021.2A Active CN112541906B (en) | 2020-12-17 | 2020-12-17 | Data processing method and device, electronic equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112541906B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108765371A (en) * | 2018-04-25 | 2018-11-06 | 浙江大学 | The dividing method of unconventional cell in a kind of pathological section |
CN110223281A (en) * | 2019-06-06 | 2019-09-10 | 东北大学 | A kind of Lung neoplasm image classification method when in data set containing uncertain data |
CN110786824A (en) * | 2019-12-02 | 2020-02-14 | 中山大学 | Coarse marking fundus oculi illumination bleeding lesion detection method and system based on bounding box correction network |
Family Cites Families (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7983459B2 (en) * | 2006-10-25 | 2011-07-19 | Rcadia Medical Imaging Ltd. | Creating a blood vessel tree from imaging data |
WO2012077130A1 (en) * | 2010-12-06 | 2012-06-14 | Indian Institute Of Technology, Kharagpur | Method and system to detect the microcalcifications in x-ray images using nonlinear energy operator. |
WO2017079568A1 (en) * | 2015-11-06 | 2017-05-11 | Google Inc. | Regularizing machine learning models |
CN110349156B (en) * | 2017-11-30 | 2023-05-30 | 腾讯科技(深圳)有限公司 | Method and device for identifying lesion characteristics in fundus picture and storage medium |
CN108389198A (en) * | 2018-02-27 | 2018-08-10 | 深思考人工智能机器人科技(北京)有限公司 | The recognition methods of atypia exception gland cell in a kind of Cervical smear |
CN108961296B (en) * | 2018-07-25 | 2020-04-14 | 腾讯医疗健康(深圳)有限公司 | Fundus image segmentation method, fundus image segmentation device, fundus image segmentation storage medium and computer equipment |
CN109190699B (en) * | 2018-08-29 | 2022-12-30 | 西安电子科技大学 | Multi-disease joint determination method based on multi-task learning |
CN109359689B (en) * | 2018-10-19 | 2021-06-04 | 科大讯飞股份有限公司 | Data identification method and device |
CN109902717A (en) * | 2019-01-23 | 2019-06-18 | 平安科技(深圳)有限公司 | Lesion automatic identifying method, device and computer readable storage medium |
CN110264443B (en) * | 2019-05-20 | 2024-04-16 | 平安科技(深圳)有限公司 | Fundus image lesion labeling method, device and medium based on feature visualization |
CN110210483B (en) * | 2019-06-13 | 2021-05-11 | 上海鹰瞳医疗科技有限公司 | Medical image lesion region segmentation method, model training method and device |
CN111383210B (en) * | 2020-03-03 | 2023-06-06 | 上海鹰瞳医疗科技有限公司 | Fundus image classification model training method and device |
CN111938569A (en) * | 2020-09-17 | 2020-11-17 | 南京航空航天大学 | Eye ground multi-disease classification detection method based on deep learning |
CN112734701A (en) * | 2020-12-28 | 2021-04-30 | 深圳市优必选科技股份有限公司 | Fundus focus detection method, fundus focus detection device and terminal equipment |
-
2020
- 2020-12-17 CN CN202011498021.2A patent/CN112541906B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108765371A (en) * | 2018-04-25 | 2018-11-06 | 浙江大学 | The dividing method of unconventional cell in a kind of pathological section |
CN110223281A (en) * | 2019-06-06 | 2019-09-10 | 东北大学 | A kind of Lung neoplasm image classification method when in data set containing uncertain data |
CN110786824A (en) * | 2019-12-02 | 2020-02-14 | 中山大学 | Coarse marking fundus oculi illumination bleeding lesion detection method and system based on bounding box correction network |
Non-Patent Citations (1)
Title |
---|
深度卷积神经网络的目标检测算法综述;黄健 等;《计算机工程与应用》;20200727;第56卷(第17期);12-23 * |
Also Published As
Publication number | Publication date |
---|---|
CN112541906A (en) | 2021-03-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
KR102641116B1 (en) | Method and device to recognize image and method and device to train recognition model based on data augmentation | |
US11521064B2 (en) | Training a neural network model | |
US10776698B2 (en) | Method for training an artificial neural network | |
EP3292512A1 (en) | Full reference image quality assessment based on convolutional neural network | |
US20190318822A1 (en) | Deep image classification of medical images | |
US20180174051A1 (en) | Method for training an artificial neural network | |
JP2015087903A (en) | Apparatus and method for information processing | |
KR102261473B1 (en) | Method for providing diagnosis system using semi-supervised machine learning and diagnosis system using the method | |
CN113330450A (en) | Method for identifying objects in an image | |
CN108629378A (en) | Image-recognizing method and equipment | |
KR102119057B1 (en) | Method for learning fracture diagnosis model and apparatus for the same | |
CN111914665A (en) | Face shielding detection method, device, equipment and storage medium | |
KR20220075273A (en) | Method of tracking multiple objects and apparatus for the same | |
CN112927209A (en) | CNN-based significance detection system and method | |
CN115100614A (en) | Evaluation method and device of vehicle perception system, vehicle and storage medium | |
CN113643302B (en) | Active contour model-based unsupervised medical image segmentation method and system | |
CN112541906B (en) | Data processing method and device, electronic equipment and storage medium | |
EP4168931A1 (en) | Domain aware medical image classifier interpretation by counterfactual impact analysis | |
CN114445679A (en) | Model training method, related device, equipment and storage medium | |
CN113076993B (en) | Information processing method and model training method for chest X-ray film recognition | |
EP4170677A1 (en) | Disease diagnosis method using neural network trained by using multi-phase biometric image, and disease diagnosis system performing same | |
CN114758188A (en) | Sample label smoothing method, device and equipment of multi-layer hierarchical classification neural network | |
KR102226897B1 (en) | Method for medical diagnosis using neural network trained by multi-phase biomedical image, medical diagnosis system performing the same | |
KR20180082680A (en) | Method for learning classifier and prediction classification apparatus using the same | |
US20220012531A1 (en) | Method for configuring an image evaluation device and also image evaluation method and image evaluation device |
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 |