CN115018874A - Fundus blood vessel segmentation domain generalization method based on frequency domain analysis - Google Patents
Fundus blood vessel segmentation domain generalization method based on frequency domain analysis Download PDFInfo
- Publication number
- CN115018874A CN115018874A CN202210786220.6A CN202210786220A CN115018874A CN 115018874 A CN115018874 A CN 115018874A CN 202210786220 A CN202210786220 A CN 202210786220A CN 115018874 A CN115018874 A CN 115018874A
- Authority
- CN
- China
- Prior art keywords
- fundus
- frequency domain
- blood vessel
- domain
- method based
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/168—Segmentation; Edge detection involving transform domain methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20052—Discrete cosine transform [DCT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Pathology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Eye Examination Apparatus (AREA)
Abstract
A fundus blood vessel segmentation domain generalization method based on frequency domain analysis comprises the following steps: 1) preprocessing the fundus image, including numerical processing and size adjustment; 2) performing feature extraction on the preprocessed fundus images, including the construction and training of a neural network model; 3) the fundus blood vessels and retinal background were segmented using a training model. The method adopts a feature normalization algorithm based on frequency domain analysis, enhances the feature expression capability and generalization capability of the model by learning a uniform semantic frequency distribution space, effectively solves the problem of data dependence of a supervised learning method and the problem of accuracy reduction of a segmentation algorithm when the segmentation algorithm is applied to data of different domains, and further improves the generalization capability of the fundus blood vessel segmentation algorithm.
Description
Technical Field
The invention belongs to the field of medical image analysis, relates to the field of domain generalization in fundus blood vessel segmentation, uses a frequency domain analysis method based on Discrete Cosine Transform (DCT), is one of deep learning technologies, and belongs to a labeled data learning method.
Background
The fundus is distributed with abundant capillaries, which can moisten the eyeball. Modern imaging technology can help doctors to directly observe the vascular structure of the retina through fundus images so as to judge the state of illness of patients. For example, capillary hemangioma, oozing blood and small bleeding spots are often seen in fundus images of diabetic patients, and arteriosclerosis phenomenon is observed in fundus blood vessels of hypertensive patients. The lesions of the ocular vessels reflect the general vascular conditions to a certain extent, and doctors can analyze and judge the types and severity of diseases through fundus examination. Fundus examination is therefore an important means of diagnosing disease in the fundus, and fundus vessel segmentation can provide an aid in quantification tools for disease analysis and diagnosis.
The fundus image has abundant details, and if the manual analysis of a doctor is relied on, the fundus image has higher requirements on the observation capability and the medical experience of the doctor. Computer-aided diagnosis techniques can help physicians improve the efficiency and accuracy of diagnosis. Conventional computer-aided diagnosis techniques extract some manually preset features from an algorithm, and the doctor makes a judgment based on the features. The rapid development of deep learning in recent years enables the visual recognition accuracy of computers to exceed that of human beings in certain specific fields. The trained deep neural network learns semantic information in the image by extracting features layer by layer, and automatically outputs the segmentation result of the fundus blood vessel at the network output end, so that the workload of doctors is remarkably reduced, and the screening and diagnosis efficiency is improved.
In an actual application scenario, samples of a deep learning training set and a testing set have differences in aspects such as illumination, contrast, sampling equipment and the like, that is, a data source domain in a training process and a data target domain in a testing process generate a domain offset problem, so that generalization performance of a model is affected, and a deep neural network often cannot obtain a satisfactory blood vessel segmentation result in a testing stage. Meanwhile, a model with good generalization capability needs a large amount of abundant labeled data for training, however, the collection and labeling of the fundus blood vessel segmentation data set are difficult, and the disclosed data set often has the problems of small sample size and insufficient diversity.
Disclosure of Invention
In order to overcome the problems of data dependence of a supervised learning method and reduction of precision of a target domain data segmentation algorithm, and further improve the generalization capability of the fundus blood vessel segmentation algorithm, the invention provides a fundus blood vessel segmentation domain generalization method based on frequency domain analysis.
The technical scheme adopted by the invention for solving the problems is as follows:
a fundus blood vessel segmentation domain generalization method based on frequency domain analysis, the method comprising the steps of:
1) preprocessing the fundus image;
2) performing characteristic extraction on the preprocessed fundus image;
3) and segmenting blood vessels and the retina background according to the features extracted by the threshold value.
Further, the step 1) comprises the following steps:
11) enhancing an original picture by using random rotation, horizontal random inversion, vertical random inversion, random brightness and saturation, and enhancing the contrast effect of the image by using Gamma nonlinear coding;
12) mapping the pixel value from 0-255 to 0-1 by using the most value normalization;
13) redundant pixels in the fundus image are cropped out, only the retinal area in the image is reserved, and then, in order to ensure the input scale invariance, the cropped fundus image is unified to 512 × 512 pixels by a bilinear interpolation method. In order to fully utilize the image information and introduce randomness, random position cutting is carried out for 64 times on each fundus image with 512 × 512 pixels, and the size of the cut image block is 64 × 64.
Still further, the step 2) comprises the following steps:
21) constructing a neural network model, and taking the preprocessed data as the input of model training;
22) training neural network, updating model parameters by algorithm
23) The predicted probabilities output by the model are used as features extracted from the input data.
Further, in the step 21), the process of network construction is as follows:
211) constructing a frequency domain-space domain normalization layer;
212) setting an encoder, wherein each layer of module consists of a residual block, a frequency domain-space domain normalization layer and a down-sampling layer;
213) a decoder is arranged, and each layer of module consists of a residual error block, a frequency domain-space domain normalization layer and an upper sampling layer;
214) the encoder and decoder are connected in a cross-layer hopping fashion.
Further, in the step 22), the training process of the network is as follows:
221) setting a loss function;
222) setting model parameters for initialization;
223) and setting an optimizer.
The process of the step 3) is as follows: and segmenting the blood vessels and the retinal background according to the extracted features, setting 0.5 as a threshold, marking feature pixels larger than 0.5 as blood vessel features, and marking feature pixel values smaller than 0.5 as fundus background features.
Compared with the prior art, the invention has the beneficial effects that: the fundus data set is preprocessed, a model fusing frequency domain information and spatial domain information is established, and features of different domains are normalized to a uniform hidden space by introducing an attention mechanism. The method can improve the robustness of the blood vessel segmentation model to data distribution disturbance, and further realize the generalization capability of fundus images from different sources.
Drawings
Fig. 1 is a flowchart of a fundus blood vessel segmentation domain generalization method based on frequency domain analysis.
Fig. 2 is a schematic diagram of preprocessing of fundus images.
Fig. 3 is a schematic diagram of a frequency domain normalization module.
FIG. 4 is a schematic illustration of spatial domain normalization.
Fig. 5 is a schematic view of a fundus blood vessel segmentation model.
Fig. 6 shows the fundus blood vessel segmentation results, where (a) is the original image, (b) is the expert label, and (c) is the segmentation result.
Detailed Description
The present invention is further described below with reference to the flow chart.
Referring to fig. 1 to 5, a fundus blood vessel segmentation domain generalization method based on frequency domain analysis includes the following steps:
1) referring to fig. 2, the fundus image is preprocessed, and the image preprocessing includes the following steps: 11) enhancing an original picture by using random rotation, horizontal random inversion, vertical random inversion, random brightness and saturation, and enhancing the contrast effect of the image by using Gamma nonlinear coding; 12) mapping the pixel value from 0-255 to 0-1 by using the most value normalization; 13) redundant pixels in the fundus image are cropped out, only the retinal area in the image is reserved, and then, in order to ensure the input scale invariance, the cropped fundus image is unified to 512 × 512 pixels by a bilinear interpolation method. In order to fully utilize image information and introduce randomness, random position cutting is carried out on each fundus image with 512 x 512 pixels for 64 times, and the size of a cut image block is 64 x 64;
2) referring to fig. 3, 4 and 5, the blood vessels and the retinal background are segmented according to the extracted features, wherein the feature extraction comprises the following steps: 21) constructing a neural network model, and taking the preprocessed data as the input of model training; 22) training a neural network, updating model parameters by an algorithm 23) the predicted probability of the model output as a feature extracted from the input data;
the process of network construction is as follows:
211) with reference to fig. 2 and 3, a frequency domain-spatial domain normalization layer is constructed, which is composed of a frequency domain normalization module and a spatial domain normalization module, wherein the frequency domain normalization module is decomposed by DCT features and summedIntroducing a learnable weight to obtain a normalized frequency domain hidden space; and the space domain normalization module dynamically generates a scaling coefficient gamma from the input image s And a translation coefficient beta s And obtaining normalized space domain characteristics.
212) With reference to fig. 5, an encoder is provided, the purpose of which is to obtain hidden spatial features, each layer of modules being composed of a residual block, a frequency domain-spatial domain normalization layer and a downsampling layer;
213) the decoder is arranged with reference to fig. 5, the purpose of which is to reconstruct a segmented image from the eigenvectors z, each layer of blocks being composed of a residual block, a frequency-domain-spatial normalization layer and an upsampling layer;
214) the encoder and decoder are connected in a cross-layer hopping fashion, the purpose of this step being to connect features of different levels and resolutions.
Further, in the step 22), the training process of the network is as follows:
221) setting a loss function, and calculating a binary cross entropy between the network output characteristics and the segmentation labels;
222) setting model parameters for initialization, and initializing parameters of the fundus blood vessel segmentation network randomly by adopting Kaiming;
223) and setting an optimizer, adopting an Adam optimization algorithm, setting the momentum to be 0.99, setting the default learning rate to be 0.002, carrying out network cycle training, and updating the parameters for 4000 rounds.
3) And (3) segmenting blood vessels and the retinal background according to the features extracted by the threshold, setting 0.5 as the threshold, marking the feature pixels larger than 0.5 as the blood vessel features, and marking the feature pixel values smaller than 0.5 as the fundus background features.
The above description is intended to be illustrative of the preferred embodiment of the present invention and should not be taken as limiting the invention, but rather, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.
Claims (6)
1. A fundus blood vessel segmentation domain generalization method based on frequency domain analysis is characterized by comprising the following steps:
1) preprocessing the fundus image;
2) performing characteristic extraction on the preprocessed fundus image;
3) segmenting the blood vessels and the retinal background according to the extracted features.
2. A fundus blood vessel segmentation domain generalization method based on frequency domain analysis according to claim 1, wherein said step 1) comprises the steps of:
11) enhancing an original picture by using random rotation, horizontal random inversion, vertical random inversion, random brightness and saturation, and enhancing the contrast effect of the image by using Gamma nonlinear coding;
12) mapping the pixel value from 0-255 to 0-1 by using the most value normalization;
13) redundant pixels in the fundus image are cut out, only the retinal area in the image is reserved, and then, in order to ensure the invariance of the input scale, bilinear interpolation is carried out on the cut fundus image.
3. A fundus blood vessel segmentation domain generalization method based on frequency domain analysis according to claim 1 or 2, wherein said step 2) comprises the steps of:
21) constructing a neural network model, and taking the preprocessed data as the input of model training;
22) training a neural network, and updating model parameters through an algorithm;
23) the predicted probabilities output by the model are used as features extracted from the input data.
4. A fundus blood vessel segmentation domain generalization method based on frequency domain analysis according to claim 3 wherein in said step 21), the process of network construction is as follows:
211) constructing a frequency domain-space domain normalization layer;
212) setting an encoder, wherein each layer of module consists of a residual block, a frequency domain-space domain normalization layer and a down-sampling layer;
213) a decoder is arranged, and each layer of module consists of a residual block, a frequency domain-space domain normalization layer and a down-sampling layer;
214) the encoder and decoder are connected in a cross-layer hopping fashion.
5. A fundus blood vessel segmentation domain generalization method based on frequency domain analysis according to claim 3 wherein in said step 22), the training process of the network is as follows:
221) setting a loss function;
222) setting model parameters for initialization;
223) and setting an optimizer.
6. The fundus blood vessel segmentation domain generalization method based on frequency domain analysis according to claim 1 or 2, wherein the process of step 3) is: and segmenting blood vessels and a retinal background according to the extracted features, setting 0.5 as a threshold, marking the feature pixels more than 0.5 as blood vessel features, and marking the feature pixels less than 0.5 as fundus background features.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210786220.6A CN115018874A (en) | 2022-07-04 | 2022-07-04 | Fundus blood vessel segmentation domain generalization method based on frequency domain analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210786220.6A CN115018874A (en) | 2022-07-04 | 2022-07-04 | Fundus blood vessel segmentation domain generalization method based on frequency domain analysis |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115018874A true CN115018874A (en) | 2022-09-06 |
Family
ID=83079396
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210786220.6A Pending CN115018874A (en) | 2022-07-04 | 2022-07-04 | Fundus blood vessel segmentation domain generalization method based on frequency domain analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115018874A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116071268A (en) * | 2023-03-01 | 2023-05-05 | 中国民用航空飞行学院 | Image illumination removal model based on contrast learning and training method thereof |
-
2022
- 2022-07-04 CN CN202210786220.6A patent/CN115018874A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116071268A (en) * | 2023-03-01 | 2023-05-05 | 中国民用航空飞行学院 | Image illumination removal model based on contrast learning and training method thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109544518B (en) | Method and system applied to bone maturity assessment | |
CN109614991A (en) | A kind of segmentation and classification method of the multiple dimensioned dilatancy cardiac muscle based on Attention | |
CN116071292B (en) | Ophthalmoscope retina image blood vessel identification method based on contrast generation learning | |
CN113724228A (en) | Tongue color and coating color identification method and device, computer equipment and storage medium | |
KR20210005206A (en) | Image processing methods, electronic devices and storage media | |
CN111951288A (en) | Skin cancer lesion segmentation method based on deep learning | |
CN113936011A (en) | CT image lung lobe image segmentation system based on attention mechanism | |
CN114512228A (en) | Traditional Chinese medicine disease auxiliary diagnosis system, equipment and storage medium | |
CN115359066B (en) | Focus detection method and device for endoscope, electronic device and storage medium | |
CN117115045A (en) | Method for improving medical image data quality based on Internet generation type artificial intelligence | |
CN113539402A (en) | Multi-mode image automatic sketching model migration method | |
CN112785581A (en) | Training method and device for extracting and training large blood vessel CTA (computed tomography angiography) imaging based on deep learning | |
CN115018874A (en) | Fundus blood vessel segmentation domain generalization method based on frequency domain analysis | |
Zhao et al. | Attention residual convolution neural network based on U-net (AttentionResU-Net) for retina vessel segmentation | |
CN114140437A (en) | Fundus hard exudate segmentation method based on deep learning | |
CN114519705A (en) | Ultrasonic standard data processing method and system for medical selection and identification | |
CN112489053B (en) | Tongue image segmentation method and device and storage medium | |
CN116740041B (en) | CTA scanning image analysis system and method based on machine vision | |
CN113706442A (en) | Medical image processing method and device based on artificial intelligence and electronic equipment | |
CN117474876A (en) | Deep learning-based kidney cancer subtype auxiliary diagnosis and uncertainty evaluation method | |
CN116721289A (en) | Cervical OCT image classification method and system based on self-supervision cluster contrast learning | |
CN116883341A (en) | Liver tumor CT image automatic segmentation method based on deep learning | |
CN115762721A (en) | Medical image quality control method and system based on computer vision technology | |
CN112967269A (en) | Pulmonary nodule identification method based on CT image | |
CN112967295A (en) | Image processing method and system based on residual error network and attention mechanism |
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 |