WO2021174821A1 - 眼底彩照图像分级方法、装置、计算机设备及存储介质 - Google Patents

眼底彩照图像分级方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2021174821A1
WO2021174821A1 PCT/CN2020/118191 CN2020118191W WO2021174821A1 WO 2021174821 A1 WO2021174821 A1 WO 2021174821A1 CN 2020118191 W CN2020118191 W CN 2020118191W WO 2021174821 A1 WO2021174821 A1 WO 2021174821A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
grading
result
processed
processed image
Prior art date
Application number
PCT/CN2020/118191
Other languages
English (en)
French (fr)
Inventor
王关政
王立龙
Original Assignee
平安科技（深圳）有限公司
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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Priority to US17/613,454 priority Critical patent/US20230154142A1/en
Publication of WO2021174821A1 publication Critical patent/WO2021174821A1/zh

Links

Classifications

    • 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
    • G06V10/56—Extraction of image or video features relating to colour
    • 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
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformation in the plane of the image
    • G06T3/40—Scaling the whole image or part thereof
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/50—Image enhancement or restoration by the use of more than one image, e.g. averaging, subtraction
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • 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
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20212—Image combination
    • G06T2207/20221—Image fusion; Image merging
    • 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

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, and storage medium for grading fundus color photograph images.
  • Fundus images include various physiological structures such as blood vessels, optic discs, and macular regions. As we age, they are prone to pathological changes.
  • One of the most widely used and most important methods for fundus examinations is to take a color photo of the fundus, through which the fundus retina is taken. With structural color pictures, doctors can directly observe and analyze whether there is an abnormality in the fundus of the photographer.
  • the current reading and diagnosis of color fundus pictures is highly dependent on the doctor’s experience and the workload is relatively large.
  • the actual clinical fundus color photo pictures are not standard pictures, so the recognition accuracy is low.
  • fundus image automatic recognition and partition methods for specific Diseases such as the identification of diabetic retinopathy based on the classification of the sugar network, but the classification of common pathologies such as age-related macular degeneration cannot be considered. Therefore, a method for automatically screening out the color fundus images with the disease to classify the color images of the fundus is needed.
  • the purpose of this application is to provide a grading method, device, computer equipment and storage medium for fundus color photographs, which are used to solve the problem that the recognition of fundus color photographs in the prior art mostly relies on the experience of doctors, and only some diseases can be automatically identified based on the color fundus photographs.
  • the present application provides a method for grading fundus color photograph images, including:
  • the pre-trained grading model is used to process the first processed image and the second processed image to obtain a target grading result.
  • the present application also provides a device for grading fundus color photos, including:
  • the first processing module is configured to obtain an original image, perform enhancement processing on the original image, and obtain a target image;
  • the second processing module is configured to perform color processing on the original image and the target image to obtain a first processed image and a second processed image respectively;
  • the execution module is used to process the first processed image and the second processed image by using a pre-trained grading model to obtain a target grading result.
  • the present application also provides a computer device, the computer device including a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor executes the computer program when the computer program is executed.
  • the pre-trained grading model is used to process the first processed image and the second processed image to obtain a target grading result.
  • the present application also provides a computer-readable storage medium, which includes multiple storage media, each of which stores a computer program, and when the computer program stored in the multiple storage media is executed by a processor Jointly realize the following steps of the above-mentioned fundus color photo image classification method:
  • the pre-trained grading model is used to process the first processed image and the second processed image to obtain a target grading result.
  • the method, device, computer equipment, and readable storage medium for grading fundus color images obtained in this application obtain a target image after enhancing the original image, and then perform color space processing on the original image and the target image to obtain six channels respectively (RGBHSV) the first processed image and the second processed image, and finally the trained grading model is used to classify the first processed image and the second processed image to obtain the target grading result.
  • RGBHSV color space processing on the original image and the target image
  • the trained grading model is used to classify the first processed image and the second processed image to obtain the target grading result.
  • the first process of the multi-color space The processed image and the second processed image are used as model input, and the fusion features are predicted on the scale of the full image to classify and classify common fundus color photographs, which solves the problem that the recognition of fundus color photographs in the prior art is mostly dependent on doctors Experience shows that only some diseases can be automatically identified based on color fundus photos, and there is a problem that there is no automatic screening method for color fundus photos that is suitable for common diseases.
  • FIG. 1 is a flowchart of Embodiment 1 of the method for grading fundus color photos according to this application;
  • FIG. 2 is a flowchart of performing enhancement processing on the original image to obtain a target image in the first embodiment of the method for grading fundus color photos according to this application;
  • FIG. 3 is a flow chart of performing color processing on the original image and the target image in the first embodiment of the method for grading fundus color photograph images according to this application to obtain a first processed image and a second processed image respectively;
  • FIG. 4 is a flowchart of training the grading model by using training samples in the first embodiment of the method for grading fundus color photograph images according to this application;
  • FIG. 5 is a flow chart of the first convolutional network in the first embodiment of the method for grading fundus color images according to this application, after feature extraction is performed on the sample original image, and the grading result is predicted to obtain the first prediction result;
  • FIG. 6 is a flowchart of the second convolutional network in the first embodiment of the method for grading fundus color photos of the application to perform feature extraction on the sample-enhanced image and then predict the grading result to obtain the second prediction result;
  • FIG. 7 is a flow chart of using a pre-trained grading model to process the first processed image and the second processed image to obtain a target grading result in Embodiment 1 of the method for grading fundus color photograph images according to this application;
  • FIG. 8 is a flow chart of using the first convolutional network to perform feature extraction on the first processed image and predicting the classification result to obtain the first classification result in the first embodiment of the method for grading fundus color photograph images according to this application;
  • FIG. 9 is a flow chart of the second embodiment of the method for grading fundus color photograph images according to the application, using a second convolutional network to perform feature extraction on the second processed image and predicting a grading result to obtain a second grading result;
  • FIG. 10 is a schematic diagram of program modules of the second embodiment of the device for grading fundus color photos according to this application;
  • FIG. 11 is a schematic diagram of the hardware structure of the computer device in the third embodiment of the computer device of this application.
  • the grading method, device, computer equipment and readable storage medium of fundus color photographs provided in this application are applicable to the field of artificial intelligence technology and provide a method for grading fundus color photographs based on a first processing module, a second processing module, and an execution module .
  • This application uses the first processing module to perform enhancement processing such as cropping, scaling, normalization, and contrast algorithms on the original image, to obtain the enhanced original image as the target image, and then use the second processing module to color the original image and the target image separately Spatial processing, the RGB channel image is processed to obtain the RGB HSV six-channel image, the corresponding first processed image and second processed image are respectively obtained, and finally the first processed image and the second processed image are processed by the execution module using the trained hierarchical model Hierarchical prediction to obtain the target grading result.
  • enhancement processing such as cropping, scaling, normalization, and contrast algorithms
  • the execution module first uses the first convolutional network and the second convolutional network to perform hierarchical prediction on the first processed image and the second processed image, and then the second The first processed image and the second processed image are feature fused and then the classification prediction is performed. Finally, the classification results obtained by the above-mentioned classification predictions are combined to obtain the target classification result.
  • the first processed image and the second processed image in a multi-color space are used. The image is used as the model input, and the fusion features are predicted on the scale of the full image to classify and classify common fundus color photographs.
  • a method for grading images of fundus color photographs of this embodiment is applied to the server side, used in the medical field fundus color photograph recognition scene, and used to automatically screen out the color fundus photographs with disease to achieve the effect of pre-screening.
  • This application can be applied in smart medical scenarios to promote the construction of smart cities, including the following steps:
  • S100 Obtain an original image, and perform enhancement processing on the original image to obtain a target image;
  • the size of the original images collected by different fundus color photography equipment is different, and the color distribution is also quite different.
  • the original image needs to be preprocessed first, namely the following steps S110-S130
  • the enhancement processing adopts a relatively common processing method, and other image enhancement methods used for fundus color photo recognition in the prior art can also be applied to this.
  • the foregoing S100 obtains an original image, performs enhancement processing on the original image to obtain a target image, referring to FIG. 2, including the following steps:
  • the preset size can be set according to the application scenario.
  • the original picture is cropped into a square with a size of 512 ⁇ 512.
  • the normalization processing refers to the process of performing a series of standard processing transformations on the image to transform it into a fixed standard form. Specifically, in this solution, the cropped or scaled image is unified into the same average value. And variance distribution.
  • S130 Use a restricted contrast adaptive histogram equalization algorithm to process the processed image to obtain a target image.
  • Adopting limited contrast adaptive histogram equalization algorithm mainly includes image segmentation, that is, taking the block as the unit, first calculate the histogram, then trim the histogram, and equalize the linear interpolation between the blocks.
  • image segmentation that is, taking the block as the unit, first calculate the histogram, then trim the histogram, and equalize the linear interpolation between the blocks.
  • the contrast limit amplitude is obtained by evenly distributing the image into 8 ⁇ 8 rectangular blocks, and calculating the histogram of each block to obtain interpolation.
  • S200 Perform color processing on the original image and the target image to obtain a first processed image and a second processed image respectively;
  • the step S200 includes the following steps:
  • RGB channels are: R (red), G (green), B (blue), through the change of the three color channels of red (R), green (G), and blue (B) and They can be superimposed on each other to get a variety of colors.
  • S220 Perform image conversion on the first RGB channel image and the RGB channel image respectively to obtain corresponding first HSV channel images and second HSV channel images;
  • the OPENCV function is used to convert the RGB channel image to the HSV channel image; each color is composed of hue (Hue, Jane H), saturation (Saturation, Jan S) and hue (Value, Jan V) As indicated, the corresponding parameters of the HSV channel are: hue (H), saturation (S), and brightness (V).
  • S240 Use the fusion of the second RGB channel image and the second HSV channel image to obtain a second processed image.
  • the original image and the target image are converted into color space, and the original RGB three-channel image is merged into a six-channel (RGB, HSV) image.
  • multi-color space RGB, HSV six-channel image
  • RGB RGB, HSV six-channel image
  • S300 Use a pre-trained grading model to process the first processed image and the second processed image to obtain a target grading result.
  • the hierarchical model includes a first convolutional network, a second convolutional network, and a fusion network.
  • the first convolutional network is used to process the first processed image (ie, the original image)
  • the second convolutional network Used to process the second processed image (that is, the target image)
  • the first convolutional network and the second convolutional network have the same structure, the processing process is the same, and the processing process is executed synchronously.
  • the training includes using training samples to train the grading model. Referring to FIG. 4, the training includes the following steps:
  • S311 Acquire a plurality of training samples, each of the training samples includes a sample original image with a graded label, and a sample enhanced image with a graded label;
  • the grading labels include mild to moderate age-related macular degeneration, severe age-related macular degeneration, mild to moderate glyconet, severe glyconet, leopard-shaped fundus, and pathological myopia;
  • the above six types are used as classification labels.
  • the above six types are common disease types.
  • more types of classification labels can also be used for training to improve the applicability of the classification model and further increase The applicable scenarios of this classification method.
  • S312 Use the first convolutional network to perform feature extraction on the original image of the sample and then predict a classification result to obtain a first prediction result;
  • the first convolutional network is used to perform feature extraction on the sample original image and then predict the classification result to obtain the first prediction result.
  • FIG. 5 including the following:
  • S312-1 Use densenet121 network to process the sample original image to obtain first sample processing data
  • the densenet121 network is a classification network, which can achieve better results in the classification of large data sets through close connection.
  • the advantages of the densenet121 network include fewer parameters, which can significantly save bandwidth and reduce storage overhead; less calculation; it has better anti-overfitting performance and generalization performance, and can comprehensively utilize the features of shallow complexity and low complexity. It is easy to get a smooth decision function with better generalization performance.
  • densenet121 network is used for classification to realize the preliminary classification of the original image of the sample.
  • S312-2 Use the squeeze excitation layer to perform global average pooling processing on the first sample processed data to obtain a first prediction result.
  • the Squeeze_excitation layer can adjust the weight ratio of each channel from global information.
  • the specific implementation is to add a se-block layer after the densenet121 network to further improve the accuracy of the classification result.
  • S313 Use a second convolutional network to perform feature extraction on the sample enhanced image and predict a classification result to obtain a second prediction result;
  • the second convolutional network is used to perform feature extraction on the sample-enhanced image and then predict the classification result to obtain the second prediction result.
  • FIG. 6 including:
  • S313-1 Use densenet121 network to process the sample enhanced image to obtain second sample processing data
  • S313-2 Use the squeeze excitation layer to perform global average pooling processing on the second sample processed data to obtain a second prediction result.
  • the steps S313-1 and S313-2 of the second convolutional network processing are the same as the processing procedures of the first convolutional network in the above steps S312-1 and S312-2, and the steps S313-1 and S313-2 are the same as the steps S313-1 and S313-2.
  • S312-1 and S312-2 are synchronized to perform the classification of the sample original image and the sample enhanced image respectively.
  • S314 Perform feature fusion on the sample original image and the sample enhanced image after feature extraction by using a fusion network, and obtain a third prediction result based on the image prediction classification result after the feature fusion;
  • the fusion network is used to perform feature fusion and then the classification is performed. Specifically, the last convolutional layer of the above two networks (ie, the first convolutional network and the second convolutional network) is subjected to Concatenate (fusion) Operation and GlobalAveragePooling (global average pooling) operation, and finally output the prediction result.
  • Concatenate fusion
  • GlobalAveragePooling global average pooling
  • S315 Obtain the target prediction result by weighting based on the first prediction result, the second prediction result, and the third prediction result, and obtain the classification by weighting according to the loss function corresponding to the first convolutional network, the second convolutional network, and the fusion network.
  • the loss function corresponding to the model, the first prediction result, the second prediction result, the third prediction result, and the target prediction result are compared with the grading label, and each loss function is calculated to adjust the grading model until training Finish.
  • the first prediction result, the second prediction result, and the third prediction result are weighted and fused with weights of 0.2, 0.2, and 0.6 respectively, and the first convolutional network, the second convolutional network, and the fusion network are respectively combined.
  • Corresponding loss functions are also weighted and fused with weights of 0.2, 0.2, and 0.6 respectively.
  • the original image and the enhanced image are classified and predicted respectively, and the loss and prediction results are weighted and fused to improve model performance and prediction accuracy.
  • a pre-trained grading model is used to process the first processed image and the second processed image to obtain a target grading result, referring to FIG. 7, including:
  • S321 Use the first convolutional network to perform feature extraction on the first processed image and predict a classification result to obtain a first classification result
  • the first convolutional network is used to perform feature extraction on the first processed image and then predict the classification result to obtain the first classification result.
  • FIG. 8 including the following:
  • S321-1 Use the densenet121 network to process the first processed image to obtain first processed data
  • S321-2 Use the squeeze excitation layer to perform global average pooling processing on the first processed data to obtain a first classification result.
  • S322 Use a second convolutional network to perform feature extraction on the second processed image and predict a classification result to obtain a second classification result
  • the second convolutional network is used to perform feature extraction on the second processed image and then predict the classification result to obtain the second classification result.
  • FIG. 9 including:
  • S322-1 Use densenet121 network to process the second processed image to obtain second processed data
  • S322-2 Use the squeeze excitation layer to perform global average pooling processing on the second processed data to obtain a second classification result.
  • the above S321 and S322 are the same as the processing procedures of S312 and S313 in the above training process.
  • the first convolutional network and the second convolutional network are respectively used for the first processed image (original image after color transformation) and
  • the second processed image (the target image after color transformation) performs hierarchical prediction, and realizes the feature fusion added to the attention module through the squeeze excitation layer.
  • S323 Perform feature fusion on the first processed image and the second processed image after feature extraction by using a fusion network, and predict a classification result based on the feature fusion image to obtain a third classification result;
  • the above-mentioned fusion network S323 and the fusion network processing process in the training process are the same.
  • the features after the fusion are predicted at the scale of the whole image, which reduces the prediction errors caused by inaccurate segmentation and positioning, and further improves the classification results. Accuracy.
  • S324 Obtain a target prediction result by weighting based on the first grading result, the second grading result, and the third grading result.
  • the original image and the enhanced image are classified and predicted, and the first classification result, the second classification result, and the third classification result are weighted and fused with the weights of 0.2, 0.2, 0.6, and further Improve model performance and prediction accuracy.
  • the target prediction results include, but are not limited to, 6 common fundus color diseases, namely mild to moderate age-related macular degeneration, severe age-related macular degeneration, mild to moderate glyconet, and severe glyconet. , Leopard-shaped fundus, pathological myopia, automatic grading is achieved through the above grading model, which further reduces the workload and low efficiency caused by manual analysis. It can automatically screen out the color photos of the diseased fundus to achieve the effect of pre-screening. Improve work efficiency.
  • the original image, target image, and target grading result can also be uploaded to the blockchain, which can be used to obtain the above-mentioned data from the blockchain for reference or as a sample, thereby ensuring its safety and user protection. Fairness and transparency.
  • the user equipment can download the summary information from the blockchain to verify whether the priority list has been tampered with.
  • the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • a fundus color photograph image grading device 4 of this embodiment includes:
  • the first processing module 41 is configured to obtain an original image, perform enhancement processing on the original image, and obtain a target image;
  • the enhancement processing includes cropping or scaling the original image according to a preset size, normalizing the cropped original image, and using a contrast-limited adaptive histogram equalization algorithm for processing.
  • Other processing means can be The realization is also possible to enhance the original image.
  • the second processing module 42 is configured to perform color processing on the original image and the target image to obtain a first processed image and a second processed image respectively;
  • color processing converts the RGB channel image into the HSV channel image through the OPENCV function, and merges the RGB channel image with the HSV channel image, and finally obtains the six-channel original image and the six-channel target image as the first processed image and the first processed image.
  • color processing converts the RGB channel image into the HSV channel image through the OPENCV function, and merges the RGB channel image with the HSV channel image, and finally obtains the six-channel original image and the six-channel target image as the first processed image and the first processed image.
  • process the image converts the RGB channel image into the HSV channel image through the OPENCV function, and merges the RGB channel image with the HSV channel image, and finally obtains the six-channel original image and the six-channel target image as the first processed image and the first processed image.
  • the execution module 43 is configured to use a pre-trained grading model to process the first processed image and the second processed image to obtain a target grading result.
  • the execution module 43 further includes the following:
  • the training unit 431 is configured to use training samples to pre-train the grading model
  • the first unit 432 is configured to use a first convolutional network to perform feature extraction on the first processed image and then predict a classification result to obtain a first classification result;
  • the first convolutional network includes a densenet121 network, and a squeeze excitation layer is added after the output layer of the network, and a Global Average Pooling (global average pooling) operation is performed on the output.
  • a Global Average Pooling global average pooling
  • the second unit 433 uses a second convolutional network to perform feature extraction on the second processed image and then predicts a classification result to obtain a second classification result;
  • the second convolutional network has the same structure as the first convolutional network in the first processing unit 432, and the processing process is the same.
  • the fusion unit 434 is configured to use a fusion network to perform feature fusion on the first processed image and the second processed image after feature extraction, and to predict a classification result based on the feature fusion image to obtain a third classification result;
  • the result processing unit 435 obtains the target prediction result by weighting based on the first grading result, the second grading result, and the third grading result.
  • the result processing unit is used to weight the first grading result, the second grading result, and the third grading result with a weight of 0.2, 0.2, 0.6
  • the target prediction result includes, but is not limited to, 6 common fundus color diseases
  • the species are mild to moderate age-related macular degeneration, severe age-related macular degeneration, mild to moderate glyconet, severe glyconet, leopard-shaped fundus, and pathological myopia.
  • This technical solution is based on the predictive model in intelligent decision-making, through the first processing module to perform cropping, scaling, normalization, contrast algorithm and other enhancement processing on the original image, to obtain the enhanced original image as the target image, and then use the second processing module
  • the original image and the target image are processed in color space
  • the RGB channel image is processed to obtain an RGB HSV six-channel image
  • the corresponding first processed image and second processed image are obtained respectively.
  • the trained grading model is used to perform the module
  • the first processed image and the second processed image are classified and predicted to obtain the target classification result.
  • the first processed image and the second processed image in the multi-color space are used as the input of the classification model, and then the common fundus color photos are realized through the classification model processing.
  • the disease types are classified and graded, which can automatically screen out the color fundus photos with lesions, achieve the effect of pre-screening, improve work efficiency, and effectively solve the problem that the recognition of fundus color photos in the prior art mostly depends on the experience of doctors, and only some diseases can be based on the fundus color photos. For automatic recognition, there is a lack of an automatic screening method for fundus color photos suitable for common diseases.
  • the processing process of the execution module also includes the use of the first convolutional network of the first unit and the second convolutional network of the second unit to perform hierarchical prediction on the first processed image and the second processed image respectively, and then The first processed image and the second processed image are feature-fused by the fusion unit to perform hierarchical prediction, and finally based on the result processing unit, the grading results obtained by the above-mentioned classification predictions are fused and weighted to obtain the target grading result, which reduces the segmentation and positioning Inaccurate prediction errors are caused by classification and prediction of the original image and the enhanced image (target image), and the loss and prediction results are weighted and fused, which further improves the model performance and prediction accuracy.
  • the present application also provides a computer device 5, which may include multiple computer devices 5.
  • the components of the fundus color photograph image grading device 4 of the second embodiment can be dispersed in different computer devices 2.
  • Device 2 can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a rack server, a blade server, a tower server, or a rack server (including an independent server, or a server cluster composed of multiple servers) that executes the program )Wait.
  • the computer equipment of this embodiment at least includes, but is not limited to: a memory 51, a processor 52, a network interface 53, and a fundus color photograph image grading device 4 that can be communicably connected to each other through a system bus, as shown in FIG. 11. It should be pointed out that FIG. 11 only shows a computer device with components, but it should be understood that it is not required to implement all the illustrated components, and more or fewer components may be implemented instead.
  • the memory 51 includes at least one type of computer-readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory ( RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disks, optical disks, etc.
  • the memory 51 may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device.
  • the memory 51 may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), and a secure digital (Secure Digital, SD) equipped on the computer device. Card, Flash Card, etc.
  • the memory 51 may also include both the internal storage unit of the computer device and its external storage device.
  • the memory 51 is generally used to store an operating system and various application software installed in a computer device, such as the program code of the fundus color photograph image grading device 4 in the first embodiment, and so on.
  • the memory 51 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 52 may be a central processing unit (Central Processing Unit) in some embodiments. Processing Unit, CPU), controller, microcontroller, microprocessor, or other data processing chip.
  • the processor 52 is generally used to control the overall operation of the computer equipment.
  • the processor 52 is used to run the program code or process data stored in the memory 51, for example, to run a text sentiment analysis device, so as to implement the method for grading fundus color photograph images in the first embodiment.
  • the network interface 53 may include a wireless network interface or a wired network interface, and the network interface 53 is generally used to establish a communication connection between the computer device 5 and other computer devices 5.
  • the network interface 53 is used to connect the computer device 5 with an external terminal through a network, and establish a data transmission channel and a communication connection between the computer device 5 and the external terminal.
  • the network may be an intranet (Intranet), the Internet (Internet), a global system of mobile communication (GSM), a wideband code division multiple access (WCDMA), 4G network, 5G Network, Bluetooth (Bluetooth), Wi-Fi and other wireless or wired networks.
  • FIG. 11 only shows the computer device 5 with components 51-53, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.
  • the text sentiment analysis device 4 stored in the memory 51 may also be divided into one or more program modules.
  • the one or more program modules are stored in the memory 51 and are composed of one or more program modules.
  • a plurality of processors (the processor 52 in this embodiment) are executed to complete the application.
  • the present application also provides a computer-readable storage medium.
  • the computer-readable storage medium may be non-volatile or volatile, and includes multiple storage media, such as flash memory, hard disk, and multimedia.
  • the computer-readable storage medium of this embodiment is used to store a device for grading a color fundus photo image, and when executed by the processor 52, realizes the method for grading a color fundus photo image of the first embodiment.

Landscapes

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

Abstract

一种眼底彩照图像分级方法、装置、计算机设备及存储介质，涉及人工智能技术领域，涉及神经网络，所述方法包括：获取原始图像，对所述原始图像进行增强处理，获得目标图像(S100)；对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像(S200)；用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果(S300)。通过多色彩空间的第一处理图像和第二处理图像作为模型输入，并通过融合后的特征在全图的尺度下进行预测实现对常见的眼底彩照病种进行分类分级，可以将有病变的眼底彩照自动筛出，达到预筛选的效果，提高工作效率。

Description

眼底彩照图像分级方法、装置、计算机设备及存储介质
本申请要求于2020年8月7日提交中国专利局申请号为202010790150.2，名称为“眼底彩照图像分级方法、装置、计算机设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中
技术领域
本申请涉及人工智能技术领域，尤其涉及一种眼底彩照图像分级方法、装置、计算机设备及存储介质。
 背景技术
眼底图像包含血管、视盘、黄斑区域等多种生理结构，随着年龄的增长，极易发生病变，眼底检查时应用最广泛也是最重要的方式之一就是拍摄眼底彩照，通过其拍摄的眼底视网膜结构彩图，医生可以直接观察和分析拍摄者眼底是否存在异常，但是目前眼底彩照的阅读和诊断高度依赖于医生经验，工作量较大。
发明人发现现有的眼底彩照识别部分根据标准图片进行参照比较识别，然而现实临床中的眼底彩照图片并不是标准图片，因此识别准确性较低，还有一些眼底图像自动识别分区方法用于特定疾病，如基于糖网分级识别糖尿病视网膜病变，但无法考虑老年黄斑变性等常见病变的分级，因此需要一种将有病变的眼底彩照自动筛出眼底彩照图像分级方法。
 技术问题
本申请的目的是提供一种眼底彩照图像分级方法、装置、计算机设备及存储介质，用于解决现有技术中的眼底彩照识别大多依赖于医生经验，仅部分疾病能够根据眼底彩照进行自动识别，缺乏一种适用于常见病种的眼底彩照自动筛选方法的问题。
 技术解决方案
为实现上述目的，本申请提供一种眼底彩照图像分级方法，包括：
获取原始图像，对所述原始图像进行增强处理，获得目标图像；
对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
为实现上述目的，本申请还提供一种眼底彩照图像分级装置，包括：
第一处理模块，用于获取原始图像，对所述原始图像进行增强处理，获得目标图像；
第二处理模块，用于对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
执行模块，用于采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
为实现上述目的，本申请还提供一种计算机设备，所述计算机设备包括存储器、处理器以及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现所述眼底彩照图像分级方法的以下步骤：
获取原始图像，对所述原始图像进行增强处理，获得目标图像；
对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
为实现上述目的，本申请还提供一种计算机可读存储介质，其包括多个存储介质，各存储介质上存储有计算机程序，所述多个存储介质存储的所述计算机程序被处理器执行时共同实现上述眼底彩照图像分级方法的以下步骤：
获取原始图像，对所述原始图像进行增强处理，获得目标图像；
对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
 有益效果
本申请提供的眼底彩照图像分级方法、装置、计算机设备及可读存储介质，通过对原始图像进行增强处理后，获得目标图像，再对原始图像和目标图像分别进行色彩空间处理，分别获得六通道（RGBHSV）的第一处理图像和第二处理图像，最后采用训练好的分级模型对第一处理图像和第二处理图像进行分级预测，获得目标分级结果，本方案中通过多色彩空间的第一处理图像和第二处理图像作为模型输入，并通过融合后的特征在全图的尺度下进行预测实现对常见的眼底彩照病种进行分类分级，解决现有技术中的眼底彩照识别大多依赖于医生经验，仅部分疾病能够根据眼底彩照进行自动识别，缺乏一种适用于常见病种的眼底彩照自动筛选方法的问题。
 附图说明
图1为本申请所述眼底彩照图像分级方法实施例一的流程图；
图2为本申请所述眼底彩照图像分级方法实施例一中对所述原始图像进行增强处理，获得目标图像的流程图；
图3为本申请所述眼底彩照图像分级方法实施例一中对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像流程图；
图4为本申请所述眼底彩照图像分级方法实施例一中采用训练样本，对所述分级模型进行训练的流程图；
图5为本申请所述眼底彩照图像分级方法实施例一中采用第一卷积网络对所述样本原始图像进行特征提取后预测分级结果，获得第一预测结果的流程图；
图6为本申请所述眼底彩照图像分级方法实施例一中采用第二卷积网络对所述样本增强图像进行特征提取后预测分级结果，获得第二预测结果的流程图；
图7为本申请所述眼底彩照图像分级方法实施例一中采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果的流程图；
图8为本申请所述眼底彩照图像分级方法实施例一中采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果的流程图；
图9为本申请所述眼底彩照图像分级方法实施例一中采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果的流程图；
图10为本申请所述眼底彩照图像分级装置实施例二的程序模块示意图；
图11为本申请计算机设备实施例三中计算机设备的硬件结构示意图。
 本发明的实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请提供的眼底彩照图像分级方法、装置、计算机设备及可读存储介质，适用于人工智能技术领域，为提供一种基于第一处理模块、第二处理模块、执行模块的眼底彩照图像分级方法。本申请通过第一处理模块对原始图像进行裁剪、缩放、归一化、对比度算法等增强处理，获得增强后的原始图像作为目标图像，再使用第二处理模块对原始图像和目标图像分别进行色彩空间处理，将RGB通道图像处理获得RGB HSV六通道图像，分别获得对应的第一处理图像和第二处理图像，最后通过执行模块采用训练好的分级模型对第一处理图像和第二处理图像进行分级预测，获得目标分级结果，具体的，执行模块在处理过程中，先分别对第一处理图像和第二处理图像分别采用第一卷积网络和第二卷积网络进行分级预测，而后将第一处理图像和第二处理图像进行特征融合后进行分级预测，最后将上述各个分级预测获得的分级结果进行融合，获得目标分级结果，本方案中通过多色彩空间的第一处理图像和第二处理图像作为模型输入，并通过融合后的特征在全图的尺度下进行预测实现对常见的眼底彩照病种进行分类分级，解决现有技术中的眼底彩照识别大多依赖于医生经验，仅部分疾病能够根据眼底彩照进行自动识别，缺乏一种适用于常见病种的眼底彩照自动筛选方法的问题，通过上述分级模型可以将有病变的眼底彩照自动筛出，达到预筛选的效果，提高工作效率。
实施例一
请参阅图1，本实施例的一种眼底彩照图像分级方法，应用于服务器端，用于医疗领域眼底彩照识别场景中，用于将有病变的眼底彩照自动筛出，达到预筛选的效果，本申请可应用于智慧医疗场景中，从而推动智慧城市的建设，包括以下步骤：
S100：获取原始图像，对所述原始图像进行增强处理，获得目标图像；
在本实施方式中，不同眼底彩照设备采集得到的原始图像大小不一，颜色分布也有较大差异，为保证分级结果的准确性，首先需要对原始图像进行预处理，即下述步骤S110-S130的增强处理，在本方案中，所述增强处理采用较为常见的处理方式，现有技术中其他用于眼底彩照识别的图片增强方式也可适用于此。
具体的，上述S100获取原始图像，对所述原始图像进行增强处理，获得目标图像，参阅图2，包括以下步骤：
S110：根据预设尺寸对所述原始图像进行裁剪或缩放；
具体的，所述预设尺寸可以根据应用场景来设定，本方案中，对原始图片进行裁剪成正方形，大小为512×512。
S120：对裁剪或缩放后的原始图像进行归一化处理，获得处理图像；
所述归一化处理指对图像进行了一系列标准的处理变换，使之变换为一固定标准形式的过程，具体的，在本方案中，将裁剪或放缩后的图像同一化为同一均值和方差分布。
S130：采用限制对比度自适应直方图均衡算法对所述处理图像进行处理，获得目标图像。
采用限制对比度自适应直方图均衡算法（CLAHE）主要包括图像分块，即以块为单位，先计算直方图，然后修剪直方图，均衡块间线性插值，这里需要遍历、操作各个图像块，处理起来复杂一些，最后与原图做图层滤色混合操作；本实施例中，其对比度限定幅值通过将图像均匀分布为8×8个矩形块，计算每个块的直方图来插值获取。
S200：对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
在卷积神经网络中，图像通常采用RGB(红绿蓝)颜色空间进行输入，但RGB颜色空间与亮度密切相关，存在一定的局限性，所以在本方案中也采用HSV（色调，饱和度，亮度）颜色空间进行输入。具体的，参阅图3，所述步骤S200包括以下步骤：
S210：分别基于所述原始图像和所述目标图像获取第一RGB通道图像和第二RGB通道图像；
需要说明的是，上述RGB通道，分别为:R(red)，G(green)，B(blue) ，通过对红(R)、绿(G)、蓝(B)三个颜色通道的变化以及它们相互之间的叠加可以得到各式各样的颜色。
S220：分别对第一RGB通道图像和第RGB通道图像进行图像转换，获得对应的第一HSV通道图像和第二HSV通道图像；
在本方案中，采用OPENCV函数将RGB通道图像转换为HSV通道图像；每一种颜色都是由色相（Hue，简H），饱和度（Saturation，简S）和色明度（Value，简V）所表示的，HSV通道对应参数分别是：色调（H），饱和度（S），亮度（V）。
S230：采用所述第一RGB通道图像与所述第一HSV通道图像融合获得第一处理图像；
S240：采用所述第二RGB通道图像与所述第二HSV通道图像融合获得第二处理图像。
在本方案中，将原始图像和目标图像进行颜色空间转换，将原始RGB三通道图像融合为六通道（RGB、HSV）图像。
本方案中采用了多色彩空间（RGB、HSV六通道图像）作为分级模型的输入，减少了传统颜色空间对亮度敏感的局限性，提高了模型对于多设备图像的泛化性能。
S300：采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
在本方案中，所述分级模型包括第一卷积网络、第二卷积网络以及融合网络，所述第一卷积网络用于处理第一处理图像（即原始图像），第二卷积网络用于处理第二处理图像（即目标图像），第一卷积网络与第二卷积网络结构一致，处理过程一致，且同步执行处理过程。
在采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理前，包括采用训练样本，对所述分级模型进行训练，参阅图4，所述训练包括以下步骤：
S311:获取多个训练样本，每一所述训练样本包括带有分级标签的样本原始图像、带有分级标签的样本增强图像；
其中，所述分级标签包括轻中度老年黄斑变性、重度老年黄斑变性、轻中度糖网、重度糖网、豹纹状眼底、病理性近视；
在本方案中，采用上述六种作为分类标签，上述六种均为常见病种，在后续应用过程中，也可以采用更多种类的分级标签进行训练，以提高分级模型的适用性，进一步增加该分级方法的适用场景。
S312:采用第一卷积网络对所述样本原始图像进行特征提取后预测分级结果，获得第一预测结果；
更具体的，采用第一卷积网络对所述样本原始图像进行特征提取后预测分级结果，获得第一预测结果，参阅图5，包括以下：
S312-1：采用densenet121网络对所述样本原始图像进行处理，获得第一样本处理数据；
在本实施方式中，densenet121网络是一个分类网络，通过紧密连接，使其在各大数据集分类上取得更好的效果。densenet121网络优点包括参数较少，可以显著地节省带宽，降低存储开销；计算量较少；具有较好的抗过拟合性能和泛化性能，可以综合利用浅层复杂度低的特征，因而更容易得到一个光滑的具有更好泛化性能的决策函数，在本方案中采用densenet121网络进行分类实现对样本原始图像的初步分级。
S312-2：采用squeeze excitation层对所述第一样本处理数据进行全局平均池化处理，获得第一预测结果。
在本实施方式中，Squeeze_excitation层可以从全局信息来调整各个通道的权重比，具体的实现方式为在 densenet121网络后增加se-block层，用于进一步提高分类结果的准确度。
S313：采用第二卷积网络对所述样本增强图像进行特征提取后预测分级结果，获得第二预测结果；
具体的，采用第二卷积网络对所述样本增强图像进行特征提取后预测分级结果，获得第二预测结果，参阅图6，包括：
S313-1：采用densenet121网络对所述样本增强图像进行处理，获得第二样本处理数据；
S313-2：采用squeeze excitation层对所述第二样本处理数据进行全局平均池化处理，获得第二预测结果。
所述第二卷积网络处理的步骤S313-1、S313-2与上述步骤S312-1、S312-2中的第一卷积网络的处理过程一样，且步骤S313-1、S313-2与步骤S312-1、S312-2同步处理，分别执行对样本原始图像与样本增强图像的分类。
S314：采用融合网络对特征提取后的所述样本原始图像和所述样本增强图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三预测结果；
在本实施方式中，采用融合网络进行特征融合后进行分级，具体的通过将上述两个网络（即第一卷积网络和第二卷积网络）的最后一层卷积层进行Concatenate（融合）操作和GlobalAveragePooling（全局平均池化）操作，最后输出预测结果。
S315：基于所述第一预测结果、第二预测结果以及第三预测结果加权获得目标预测结果，根据第一卷积网络、第二卷积网络以及融合网络分别对应的损失函数加权获得所述分级模型对应的损失函数，将所述第一预测结果、第二预测结果、第三预测结果以及目标预测结果与所述分级标签进行比对，计算各个损失函数对所述分级模型进行调整，直至训练完成。
具体的，将所述第一预测结果、第二预测结果以及第三预测结果分别以0.2，0.2，0.6的权重进行加权融合，将所述一卷积网络、第二卷积网络以及融合网络分别对应的损失函数也分别以0.2，0.2，0.6的权重进行加权融合，通过分别对原始图像和增强后的图像进行分类预测，并将损失和预测结果进行加权融合，提高模型性能和预测精度。
具体的，采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果，参阅图7，包括：
S321：采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果；
具体的，采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果，参阅图8，包括以下：
S321-1：采用densenet121网络对所述第一处理图像进行处理，获得第一处理数据；
S321-2：采用squeeze excitation层对所述第一处理数据进行全局平均池化处理，获得第一分级结果。
S322：采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果；
具体的，采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果，参阅图9，包括：
S322-1：采用densenet121网络对所述第二处理图像进行处理，获得第二处理数据；
S322-2：采用squeeze excitation层对所述第二处理数据进行全局平均池化处理，获得第二分级结果。
在本方案中，上述S321与S322与上述训练过程中S312、S313的处理过程一致，分别采用第一卷积网络和第二卷积网络分别对第一处理图像（色彩变换后的原始图像）和第二处理图像（色彩变换后的目标图像）进行分级预测，通过squeeze excitation层实现加入注意力模块的特征融合。
S323：采用融合网络对特征提取后的所述第一处理图像和所述第二处理图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三分级结果；
本方案中，上述融合网络S323和训练过程中的融合网络处理过程一致，过融合后的特征在全图的尺度下进行预测，减少了因分割和定位不准造成的预测错误，进一步提高分级结果的准确性。
S324：基于所述第一分级结果、第二分级结果以及第三分级结果加权获得目标预测结果。
本方案中分别对原始图像和增强后的图像（即目标图像）进行分类预测，并将第一分级结果、第二分级结果以及第三分级结果以0.2，0.2,0.6的权重进行加权融合，进一步提高模型性能和预测精度，同时，所述目标预测结果包括但不限于6种常见的眼底彩照病种，分别为轻中度老年黄斑变性，重度老年黄斑变性，轻中度糖网，重度糖网，豹纹状眼底，病理性近视，通过上述分级模型实现自动分级，进一步减少人工分析造成的工作量大，效率低下的问题，可以将有病变的眼底彩照自动筛出，达到预筛选的效果，提高工作效率。
本方案中还可将原始图像、目标图像、目标分级结果上传至区块链，用于后续可从区块链中获取上述数据进行参考或作为样本，由此可保证其安全性和对用户的公正透明性。用户设备可以从区块链中下载得该摘要信息，以便查证优先级列表是否被篡改。本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
实施例二：
请参阅图10，本实施例的一种眼底彩照图像分级装置4，包括：
第一处理模块41，用于获取原始图像，对所述原始图像进行增强处理，获得目标图像；
具体的，所述增强处理包括根据预设尺寸对所述原始图像进行裁剪或缩放、对裁剪后的原始图像进行归一化处理以及采用限制对比度自适应直方图均衡算法进行处理，其他处理手段能够实现是对原始图像增强也可。
第二处理模块42，用于对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
具体的，色彩处理通过OPENCV函数将RGB通道图像转换为HSV通道图像，并将RGB通道图像与HSV通道图像融合，最终获得六通道的原始图像和六通道目标图像分别作为获得第一处理图像和第二处理图像。
执行模块43，用于采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
具体的，所述执行模块43还包括以下：
训练单元431，用于采用训练样本对所述分级模型进行预训练；
第一单元432，用于采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果；
具体的，所述第一卷积网络包括densenet121网络，并在该网络输出层后增加squeeze excitation层，对输出进行GlobalAveragePooling（全局平均池化）操作。
第二单元433，采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果；
具体的，所述第二卷积网络与上述第一处理单元432中的第一卷积网络结构一致，处理过程一致。
融合单元434，用于采用融合网络对特征提取后的所述第一处理图像和所述第二处理图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三分级结果；
具体的，将上述两个网络（即第一卷积网络和第二卷积网络）的最后一层卷积层进行Concatenate（融合）操作和GlobalAveragePooling（全局平均池化）操作，最后输出第三分级结果。
结果处理单元435，基于所述第一分级结果、第二分级结果以及第三分级结果加权获得目标预测结果。
具体的，采用结果处理单元将所述第一分级结果、第二分级结果以及第三分级结果以0.2，0.2,0.6的权重加权，所述目标预测结果包括但不限于6种常见的眼底彩照病种，分别为轻中度老年黄斑变性，重度老年黄斑变性，轻中度糖网，重度糖网，豹纹状眼底，病理性近视。
本技术方案基于智能决策中的预测模型，通过第一处理模块对原始图像进行裁剪、缩放、归一化、对比度算法等增强处理，获得增强后的原始图像作为目标图像，再使用第二处理模块对原始图像和目标图像分别进行色彩空间处理，将RGB通道图像处理获得RGB HSV六通道图像，分别获得对应的第一处理图像和第二处理图像，最后通过执行模块采用训练好的分级模型对第一处理图像和第二处理图像进行分级预测，获得目标分级结果，本方案中通过多色彩空间的第一处理图像和第二处理图像作为分级模型输入，再通过分级模型处理实现对常见的眼底彩照病种进行分类分级，可以将有病变的眼底彩照自动筛出，达到预筛选的效果，提高工作效率，有效解决现有技术中的眼底彩照识别大多依赖于医生经验，仅部分疾病能够根据眼底彩照进行自动识别，缺乏一种适用于常见病种的眼底彩照自动筛选方法的问题。
在本方案中，执行模块的处理过程中，还包括采用第一单元的第一卷积网络和第二单元的第二卷积网络分别对第一处理图像和第二处理图像进行分级预测，而后通过融合单元将第一处理图像和第二处理图像进行特征融合后进行分级预测，最后基于结果处理单元将上述各个分级预测获得的分级结果进行融合加权，获得目标分级结果，减少了因分割和定位不准造成的预测错误，同时分别对原始图像和增强后的图像（目标图像）进行分类预测，并将损失和预测结果进行加权融合，进一步提高了模型性能和预测精度。
实施例三：
为实现上述目的，本申请还提供一种计算机设备5，该计算机设备可包括多个计算机设备5，实施例二的眼底彩照图像分级装置4的组成部分可分散于不同的计算机设备2中，计算机设备2可以是执行程序的智能手机、平板电脑、笔记本电脑、台式计算机、机架式服务器、刀片式服务器、塔式服务器或机柜式服务器（包括独立的服务器，或者多个服务器所组成的服务器集群）等。本实施例的计算机设备至少包括但不限于：可通过系统总线相互通信连接的存储器51、处理器52、网络接口53以及眼底彩照图像分级装置4，如图11所示。需要指出的是，图11仅示出了具有组件-的计算机设备，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。
本实施例中，存储器51至少包括一种类型的计算机可读存储介质，所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器（例如，SD或DX存储器等）、随机访问存储器（RAM）、静态随机访问存储器（SRAM）、只读存储器（ROM）、电可擦除可编程只读存储器（EEPROM）、可编程只读存储器（PROM）、磁性存储器、磁盘、光盘等。在一些实施例中，存储器51可以是计算机设备的内部存储单元，例如该计算机设备的硬盘或内存。在另一些实施例中，存储器51也可以是计算机设备的外部存储设备，例如该计算机设备上配备的插接式硬盘，智能存储卡（Smart Media Card, SMC），安全数字（Secure Digital, SD）卡，闪存卡（Flash Card）等。当然，存储器51还可以既包括计算机设备的内部存储单元也包括其外部存储设备。本实施例中，存储器51通常用于存储安装于计算机设备的操作系统和各类应用软件，例如实施例一的眼底彩照图像分级装置4的程序代码等。此外，存储器51还可以用于暂时地存储已经输出或者将要输出的各类数据。
处理器52在一些实施例中可以是中央处理器（Central Processing Unit，CPU）、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器52通常用于控制计算机设备的总体操作。本实施例中，处理器52用于运行存储器51中存储的程序代码或者处理数据，例如运行文本情感分析装置，以实现实施例一的眼底彩照图像分级方法。
所述网络接口53可包括无线网络接口或有线网络接口，该网络接口53通常用于在所述计算机设备5与其他计算机设备5之间建立通信连接。例如，所述网络接口53用于通过网络将所述计算机设备5与外部终端相连，在所述计算机设备5与外部终端之间的建立数据传输通道和通信连接等。所述网络可以是企业内部网（Intranet）、互联网（Internet）、全球移动通讯系统（Global System of Mobile communication，GSM）、宽带码分多址（Wideband Code Division Multiple Access，WCDMA）、4G网络、5G网络、蓝牙（Bluetooth）、Wi-Fi等无线或有线网络。
需要指出的是，图11仅示出了具有部件51-53的计算机设备5，但是应理解的是，并不要求实施所有示出的部件，可以替代的实施更多或者更少的部件。
在本实施例中，存储于存储器51中的所述文本情感分析装置4还可以被分割为一个或者多个程序模块，所述一个或者多个程序模块被存储于存储器51中，并由一个或多个处理器（本实施例为处理器52）所执行，以完成本申请。
实施例四：
为实现上述目的，本申请还提供一种计算机可读存储介质，所述计算机可读存储介质可以是非易失性，也可以是易失性，其包括多个存储介质，如闪存、硬盘、多媒体卡、卡型存储器（例如，SD或DX存储器等）、随机访问存储器（RAM）、静态随机访问存储器（SRAM）、只读存储器（ROM）、电可擦除可编程只读存储器（EEPROM）、可编程只读存储器（PROM）、磁性存储器、磁盘、光盘、服务器、App应用商城等等，其上存储有计算机程序，程序被处理器52执行时实现相应功能。本实施例的计算机可读存储介质用于存储眼底彩照图像分级装置，被处理器52执行时实现实施例一的眼底彩照图像分级方法。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种眼底彩照图像分级方法，其中，包括：
    获取原始图像，对所述原始图像进行增强处理，获得目标图像；
    对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
    采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
  2. 根据权利要求1所述的眼底彩照图像分级方法，其中，对所述原始图像进行增强处理，获得目标图像，包括以下：
    根据预设尺寸对所述原始图像进行裁剪或缩放；
    对裁剪或缩放后的原始图像进行归一化处理，获得处理图像；
    采用限制对比度自适应直方图均衡算法对所述处理图像进行处理，获得目标图像。
  3. 根据权利要求1所述的眼底彩照图像分级方法，其中，对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像，包括以下：
    分别基于所述原始图像和所述目标图像获取第一RGB三通道图像和第二RGB三通道图像；
    分别对第一RGB三通道图像和第二RGB三通道图像进行图像转换，获得对应的第一HSV三通道图像和第二HSV三通道图像；
    采用所述第一RGB通道图像与所述第一HSV通道图像融合获得第一处理图像；
    采用所述第二RGB通道图像与所述第二HSV通道图像融合获得第二处理图像。
  4. 根据权利要求1所述的眼底彩照图像分级方法，其中，在采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理前，包括采用训练样本，对所述分级模型进行训练，所述训练包括以下：
    获取多个训练样本，每一所述训练样本包括带有分级标签的样本原始图像、带有分级标签的样本增强图像；
    其中，所述分级标签包括轻中度老年黄斑变性、重度老年黄斑变性、轻中度糖网、重度糖网、豹纹状眼底、病理性近视；
    采用第一卷积网络对所述样本原始图像进行特征提取后预测分级结果，获得第一预测结果；
    采用第二卷积网络对所述样本增强图像进行特征提取后预测分级结果，获得第二预测结果；
    采用融合网络对特征提取后的所述样本原始图像和所述样本增强图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三预测结果；
    基于所述第一预测结果、第二预测结果以及第三预测结果加权获得目标预测结果，根据第一卷积网络、第二卷积网络以及融合网络分别对应的损失函数加权获得所述分级模型对应的损失函数，将所述第一预测结果、第二预测结果、第三预测结果以及目标预测结果与所述分级标签进行比对，计算各个损失函数对所述分级模型进行调整，直至训练完成。
  5. 根据权利要求1所述的眼底彩照图像分级方法，其中，采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果，包括：
    采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果；
    采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果；
    采用融合网络对特征提取后的所述第一处理图像和所述第二处理图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三分级结果；
    基于所述第一分级结果、第二分级结果以及第三分级结果加权获得目标预测结果。
  6. 根据权利要求5所述的眼底彩照图像分级方法，其中，采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果，包括以下：
    采用densenet121网络对所述第一处理图像进行处理，获得第一处理数据；
    采用squeeze excitation层对所述第一处理数据进行全局平均池化处理，获得第一分级结果。
  7. 根据权利要求5所述的眼底彩照图像分级方法，其中，采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果，包括：
    采用densenet121网络对所述第二处理图像进行处理，获得第二处理数据；
    采用squeeze excitation层对所述第二处理数据进行全局平均池化处理，获得第二分级结果。
  8. 一种眼底彩照图像分级装置，其中，包括：
    第一处理模块，用于获取原始图像，对所述原始图像进行增强处理，获得目标图像；
    第二处理模块，用于对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
    执行模块，用于采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
  9. 根据权利要求8所述的眼底彩照图像分级装置，其中，所述第一处理模块用于根据预设尺寸对所述原始图像进行裁剪或缩放；对裁剪或缩放后的原始图像进行归一化处理，获得处理图像；采用限制对比度自适应直方图均衡算法对所述处理图像进行处理，获得目标图像。
  10. 根据权利要求8所述的眼底彩照图像分级装置，其中，所述第二处理模块用于分别基于所述原始图像和所述目标图像获取第一RGB三通道图像和第二RGB三通道图像；分别对第一RGB三通道图像和第二RGB三通道图像进行图像转换，获得对应的第一HSV三通道图像和第二HSV三通道图像；采用所述第一RGB通道图像与所述第一HSV通道图像融合获得第一处理图像；采用所述第二RGB通道图像与所述第二HSV通道图像融合获得第二处理图像。
  11. 根据权利要求8所述的眼底彩照图像分级装置，其中，所述执行模块用于采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果；采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果；采用融合网络对特征提取后的所述第一处理图像和所述第二处理图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三分级结果；基于所述第一分级结果、第二分级结果以及第三分级结果加权获得目标预测结果。
  12. 一种计算机设备，其中，所述计算机设备包括存储器、处理器以及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现所述眼底彩照图像分级方法的以下步骤：
    获取原始图像，对所述原始图像进行增强处理，获得目标图像；
    对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
    采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
  13. 根据权利要求12所述的计算机设备，其中，对所述原始图像进行增强处理，获得目标图像，包括以下：
    根据预设尺寸对所述原始图像进行裁剪或缩放；
    对裁剪或缩放后的原始图像进行归一化处理，获得处理图像；
    采用限制对比度自适应直方图均衡算法对所述处理图像进行处理，获得目标图像。
  14. 根据权利要求12所述的计算机设备，其中，对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像，包括以下：
    分别基于所述原始图像和所述目标图像获取第一RGB三通道图像和第二RGB三通道图像；
    分别对第一RGB三通道图像和第二RGB三通道图像进行图像转换，获得对应的第一HSV三通道图像和第二HSV三通道图像；
    采用所述第一RGB通道图像与所述第一HSV通道图像融合获得第一处理图像；
    采用所述第二RGB通道图像与所述第二HSV通道图像融合获得第二处理图像。
  15. 根据权利要求12所述的计算机设备，其中，采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果，包括：
    采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果；
    采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果；
    采用融合网络对特征提取后的所述第一处理图像和所述第二处理图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三分级结果；
    基于所述第一分级结果、第二分级结果以及第三分级结果加权获得目标预测结果。
  16. 一种计算机可读存储介质，其包括多个存储介质，各存储介质上存储有计算机程序，其中，所述多个存储介质存储的所述计算机程序被处理器执行时共同实现所述眼底彩照图像分级方法的以下步骤：
    获取原始图像，对所述原始图像进行增强处理，获得目标图像；
    对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像；
    采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果。
  17. 根据权利要求16所述的计算机可读存储介质，其中，对所述原始图像进行增强处理，获得目标图像，包括以下：
    根据预设尺寸对所述原始图像进行裁剪或缩放；
    对裁剪或缩放后的原始图像进行归一化处理，获得处理图像；
    采用限制对比度自适应直方图均衡算法对所述处理图像进行处理，获得目标图像。
  18. 根据权利要求16所述的计算机可读存储介质，其中，对所述原始图像和所述目标图像进行色彩处理，分别获得第一处理图像和第二处理图像，包括以下：
    分别基于所述原始图像和所述目标图像获取第一RGB三通道图像和第二RGB三通道图像；
    分别对第一RGB三通道图像和第二RGB三通道图像进行图像转换，获得对应的第一HSV三通道图像和第二HSV三通道图像；
    采用所述第一RGB通道图像与所述第一HSV通道图像融合获得第一处理图像；
    采用所述第二RGB通道图像与所述第二HSV通道图像融合获得第二处理图像。
  19. 根据权利要求16所述的计算机可读存储介质，其中，在采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理前，包括采用训练样本，对所述分级模型进行训练，所述训练包括以下：
    获取多个训练样本，每一所述训练样本包括带有分级标签的样本原始图像、带有分级标签的样本增强图像；
    其中，所述分级标签包括轻中度老年黄斑变性、重度老年黄斑变性、轻中度糖网、重度糖网、豹纹状眼底、病理性近视；
    采用第一卷积网络对所述样本原始图像进行特征提取后预测分级结果，获得第一预测结果；
    采用第二卷积网络对所述样本增强图像进行特征提取后预测分级结果，获得第二预测结果；
    采用融合网络对特征提取后的所述样本原始图像和所述样本增强图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三预测结果；
    基于所述第一预测结果、第二预测结果以及第三预测结果加权获得目标预测结果，根据第一卷积网络、第二卷积网络以及融合网络分别对应的损失函数加权获得所述分级模型对应的损失函数，将所述第一预测结果、第二预测结果、第三预测结果以及目标预测结果与所述分级标签进行比对，计算各个损失函数对所述分级模型进行调整，直至训练完成。
  20. 根据权利要求16所述的计算机可读存储介质，其中，采用预训练的分级模型对所述第一处理图像和所述第二处理图像进行处理，获得目标分级结果，包括：
    采用第一卷积网络对所述第一处理图像进行特征提取后预测分级结果，获得第一分级结果；
    采用第二卷积网络对所述第二处理图像进行特征提取后预测分级结果，获得第二分级结果；
    采用融合网络对特征提取后的所述第一处理图像和所述第二处理图像进行特征融合，并基于所述特征融合后的图像预测分级结果，获得第三分级结果；
    基于所述第一分级结果、第二分级结果以及第三分级结果加权获得目标预测结果。
PCT/CN2020/118191 2020-08-07 2020-09-27 眼底彩照图像分级方法、装置、计算机设备及存储介质 WO2021174821A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/613,454 US20230154142A1 (en) 2020-08-07 2020-09-27 Fundus color photo image grading method and apparatus, computer device, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010790150.2 2020-08-07
CN202010790150.2A CN111951933B (zh) 2020-08-07 2020-08-07 眼底彩照图像分级方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021174821A1 true WO2021174821A1 (zh) 2021-09-10

Family

ID=73331882

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/118191 WO2021174821A1 (zh) 2020-08-07 2020-09-27 眼底彩照图像分级方法、装置、计算机设备及存储介质

Country Status (3)

Country Link
US (1) US20230154142A1 (zh)
CN (1) CN111951933B (zh)
WO (1) WO2021174821A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113910269A (zh) * 2021-10-27 2022-01-11 因格(苏州)智能技术有限公司 一种机器人总控系统

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112991343B (zh) * 2021-04-30 2021-08-13 北京至真互联网技术有限公司 眼底图像黄斑区域的识别检测方法和装置及设备
CN116894804A (zh) * 2022-03-29 2023-10-17 山东第一医科大学附属省立医院(山东省立医院) 基于广角眼底图像的推理模型的生成方法及相关设备
CN114724001A (zh) * 2022-06-10 2022-07-08 科大讯飞股份有限公司 一种图像多属性识别方法、装置、设备及存储介质
CN115083005B (zh) * 2022-06-13 2023-07-04 广东省人民医院 一种基于深度学习的rop图像分类系统及方法
CN116188810B (zh) * 2023-04-25 2023-07-21 浙江一山智慧医疗研究有限公司 视神经盘特征提取方法、装置及应用
CN117576012B (zh) * 2023-11-10 2024-05-07 中国矿业大学 一种基于不平衡眼底图像数据的疾病预测方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108095683A (zh) * 2016-11-11 2018-06-01 北京羽医甘蓝信息技术有限公司 基于深度学习的处理眼底图像的方法和装置
CN109784148A (zh) * 2018-12-06 2019-05-21 北京飞搜科技有限公司 活体检测方法及装置
CN110796161A (zh) * 2019-09-18 2020-02-14 平安科技（深圳）有限公司 识别模型训练、眼底特征的识别方法、装置、设备及介质
US20200187775A1 (en) * 2018-12-12 2020-06-18 Samsung Life Public Welfare Foundation Eye disease diagnosis method and system using artificial intelligence
CN111481166A (zh) * 2017-05-04 2020-08-04 深圳硅基智能科技有限公司 基于眼底筛查的自动识别系统

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107704886A (zh) * 2017-10-20 2018-02-16 北京工业大学 一种基于深度卷积神经网络的医疗图像分级系统和方法
CN110348428B (zh) * 2017-11-01 2023-03-24 腾讯科技（深圳）有限公司 眼底图像分类方法、装置及计算机可读存储介质
CN108280462A (zh) * 2017-12-11 2018-07-13 北京三快在线科技有限公司 一种模型训练方法及装置，电子设备
CN110335269A (zh) * 2018-05-16 2019-10-15 腾讯医疗健康(深圳)有限公司 眼底图像的类别识别方法和装置
CN111460150B (zh) * 2020-03-27 2023-11-10 北京小米松果电子有限公司 一种分类模型的训练方法、分类方法、装置及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108095683A (zh) * 2016-11-11 2018-06-01 北京羽医甘蓝信息技术有限公司 基于深度学习的处理眼底图像的方法和装置
CN111481166A (zh) * 2017-05-04 2020-08-04 深圳硅基智能科技有限公司 基于眼底筛查的自动识别系统
CN109784148A (zh) * 2018-12-06 2019-05-21 北京飞搜科技有限公司 活体检测方法及装置
US20200187775A1 (en) * 2018-12-12 2020-06-18 Samsung Life Public Welfare Foundation Eye disease diagnosis method and system using artificial intelligence
CN110796161A (zh) * 2019-09-18 2020-02-14 平安科技（深圳）有限公司 识别模型训练、眼底特征的识别方法、装置、设备及介质

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113910269A (zh) * 2021-10-27 2022-01-11 因格(苏州)智能技术有限公司 一种机器人总控系统

Also Published As

Publication number Publication date
US20230154142A1 (en) 2023-05-18
CN111951933B (zh) 2023-01-17
CN111951933A (zh) 2020-11-17

Similar Documents

Publication Publication Date Title
WO2021174821A1 (zh) 眼底彩照图像分级方法、装置、计算机设备及存储介质
US11526983B2 (en) Image feature recognition method and apparatus, storage medium, and electronic apparatus
JP7058373B2 (ja) 医療画像に対する病変の検出及び位置決め方法、装置、デバイス、及び記憶媒体
Raj et al. Fundus image quality assessment: survey, challenges, and future scope
CN111291825B (zh) 病灶分类模型训练方法、装置、计算机设备和存储介质
CN108615236A (zh) 一种图像处理方法及电子设备
Kurmann et al. Expert-level automated biomarker identification in optical coherence tomography scans
TW202014984A (zh) 一種圖像處理方法、電子設備及存儲介質
JP7391267B2 (ja) 医用画像処理方法、装置、機器、記憶媒体及びコンピュータプログラム
WO2022242392A1 (zh) 血管图像分类处理方法、装置、设备及存储介质
WO2021159811A1 (zh) 青光眼辅助诊断装置、方法及存储介质
US20240112329A1 (en) Distinguishing a Disease State from a Non-Disease State in an Image
WO2022194152A1 (zh) 基于图像处理模型的图像处理方法、装置、电子设备、存储介质及计算机程序产品
CN112971718A (zh) 证候识别方法、装置、电子设备及存储介质
WO2021114818A1 (zh) 基于傅里叶变换的oct图像质量评估方法、系统及装置
Vodrahalli et al. TrueImage: a machine learning algorithm to improve the quality of telehealth photos
CN114330499A (zh) 分类模型的训练方法、装置、设备、存储介质及程序产品
CN113240655A (zh) 一种自动检测眼底图像类型的方法、存储介质及装置
Ullah et al. SSMD-UNet: Semi-supervised multi-task decoders network for diabetic retinopathy segmentation
WO2021159633A1 (zh) 图像识别模型训练方法及系统和图像识别方法
CN113723310B (zh) 基于神经网络的图像识别方法及相关装置
CN110428377B (zh) 数据扩充方法、装置、设备和介质
CN107038705A (zh) 视网膜图像出血区域分割方法、装置和计算设备
CN112966620A (zh) 眼底图像处理方法、模型训练方法及设备
WO2021097595A1 (zh) 图像的病变区域分割方法、装置及服务器

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20923447

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205N DATED 29/03/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20923447

Country of ref document: EP

Kind code of ref document: A1