CN113012093A - Training method and training system for glaucoma image feature extraction - Google Patents

Abstract

The present disclosure describes a training method for glaucoma image feature extraction based on an artificial neural network, which includes: preparing a fundus image and an annotation image, wherein the annotation image comprises a disc annotation image for marking out a disc region and a cup annotation image for marking out a cup region; preprocessing the fundus image to obtain a preprocessed fundus image, and detecting blood vessels to obtain a blood vessel image; carrying out mixed weighting on the optic disc labeling image and the blood vessel image to generate a mixed weight distribution map; and training the artificial neural network by adopting a mixed weight distribution map based on the preprocessed fundus image and the labeled image. Therefore, the accuracy of the artificial neural network for identifying the features of the glaucoma image can be improved.

Description

Training method and training system for glaucoma image feature extraction

Technical Field

The disclosure relates to a training method and a training system for glaucoma image feature extraction.

Background

Glaucoma has now become the second blinding eye disease worldwide. Patients with primary glaucoma have exceeded millions of people worldwide, with more than one patient possibly developing double-blind eyes. Early glaucoma screening is of great importance because glaucoma can develop irreversible blindness if it is not diagnosed early.

The main pathological process of glaucoma is the defect of nerve fibers at the edge of the optic disc due to the death of retinal ganglion cells and the loss of axons, thereby causing the morphological change of the optic disc, such as the enlargement of the optic disc pit, the deepening of the optic disc pit and the like. Clinical medical research shows that cup-to-disc ratio (CDR) of the fundus image is a reliable index for measuring optic disc depression, so that glaucoma can be identified through the cup-to-disc ratio of the fundus image. In clinical medicine, existing identification methods include identifying discs or cups by processing features in fundus images through artificial intelligence techniques to identify lesions in the fundus.

However, since the optic disc region generally has only 4-6 pairs of grade 1 or grade 2 retinal artery and vein, when the fundus image is recognized using the artificial intelligence technique, it is easy to ignore the arteriovenous information and to cause the small blood vessel running to be not accurately learned, resulting in the inability to accurately recognize the optic disc or cup.

Disclosure of Invention

The present disclosure has been made in view of the above-described state of the art, and an object of the present disclosure is to provide a training method and a training system for glaucoma image feature extraction using an artificial neural network, which can improve the accuracy of glaucoma image feature extraction by the artificial neural network.

To this end, the present disclosure provides, in a first aspect, a training method for glaucoma image feature extraction based on an artificial neural network, including: preparing a fundus image and an annotation image, wherein the annotation image comprises a disc annotation image for marking out a disc region and a cup annotation image for marking out a cup region; preprocessing the fundus image to obtain a preprocessed fundus image, and generating a blood vessel image containing a blood vessel region according to a blood vessel detection result; performing mixed weighting on the optic disc labeling image and the blood vessel image to generate a mixed weight distribution map; and training an artificial neural network based on the preprocessed fundus image, the annotation image and the mixed weight distribution map, wherein the weight of the optic disc region is greater than that of a non-optic disc region, and the weight of the blood vessel region is greater than that of the non-blood vessel region when the mixed weighting is performed. According to the method, the artificial neural network is trained on the basis of the preprocessed fundus image, the labeled image and the mixed weight distribution map, so that the blood vessel region and the optic disc region can be considered in the training of the artificial neural network, the learning of small blood vessel walking is optimized, and the imbalance of positive and negative samples is restrained, so that the accuracy of the artificial neural network in extracting the glaucoma image features can be improved.

In addition, in the training method for artificial neural network-based glaucoma image feature extraction according to the first aspect of the present disclosure, optionally, the optic disc region in the optic disc labeling image is dilated to form an optic disc dilated image; expanding the vessel region in the vessel image to form a vessel expanded image. In this case, a disc expansion image containing a region near the disc and a blood vessel expansion image containing a blood vessel boundary are obtained by the expansion processing.

In addition, in the training method for artificial neural network-based glaucoma image feature extraction according to the first aspect of the present disclosure, optionally, the optic disc expansion image and the blood vessel expansion image are mixed and weighted to generate the mixed weight distribution map. Therefore, the accuracy of the artificial neural network for extracting the glaucoma image features can be further improved.

In addition, in the training method for glaucoma image feature extraction based on an artificial neural network according to the first aspect of the present disclosure, optionally, in the training process, coefficients of a loss function in the artificial neural network are obtained based on the mixed weight distribution map, and the artificial neural network is trained based on the loss function. In this case, the artificial neural network is optimized based on the loss function coefficient obtained from the mixed weight distribution map, and imbalance between the positive and negative samples can be suppressed, whereby the accuracy of the artificial neural network in extracting the glaucoma image features can be further improved.

In addition, in the training method for artificial neural network-based glaucoma image feature extraction according to the first aspect of the present disclosure, optionally, the mixed weight distribution map includes a fundus region and a background region, and the weight of the background region is set to zero. Therefore, the interference of the background area on the recognition of the glaucoma image characteristics by the artificial neural network can be reduced.

In addition, in the training method for artificial neural network-based glaucoma image feature extraction according to the first aspect of the present disclosure, optionally, the preprocessing includes performing cropping and normalization processing on the fundus image. In this case, the cropping process enables the fundus image to be converted into an image of a fixed standard format, and the normalization process enables the difference between different fundus images to be overcome, thereby enabling the artificial neural network to more conveniently extract the glaucoma image features.

In addition, in the training method for artificial neural network-based glaucoma image feature extraction according to the first aspect of the present disclosure, optionally, the mixed weight distribution map includes an intra-optic disc blood vessel region, an intra-optic disc non-blood vessel region, an extra-optic disc blood vessel region, and an extra-optic disc non-blood vessel region. Thus, the artificial neural network can more accurately extract the glaucoma image features of each region in the fundus image.

In addition, in the training method for artificial neural network-based glaucoma image feature extraction according to the first aspect of the present disclosure, optionally, the weight of the optic disc region is a first weight, the weight of the non-optic disc region is a second weight, the weight of the blood vessel region is a third weight, and the weight of the non-blood vessel region is a fourth weight, the weight of the intra-optic disc blood vessel region is the first weight multiplied by the third weight, the weight of the intra-optic disc non-blood vessel region is the first weight multiplied by the fourth weight, the weight of the extra-optic disc blood vessel region is the second weight multiplied by the third weight, and the weight of the extra-optic disc non-blood vessel region is the second weight multiplied by the fourth weight. Therefore, the weight values of the intra-optic disc blood vessel region, the intra-optic disc non-blood vessel region, the extra-optic disc blood vessel region and the extra-optic disc non-blood vessel region can be respectively obtained according to the weights of the optic disc region, the non-optic disc region, the blood vessel region and the non-blood vessel region.

In addition, in the method for training glaucoma image features based on an artificial neural network according to the first aspect of the present disclosure, optionally, the vessel region detection is performed based on frani filtering to form the vessel image. Therefore, the blood vessel region can be automatically identified, and the subsequent artificial neural network can conveniently identify and process the blood vessel region.

A second aspect of the present disclosure provides a training system for glaucoma image feature extraction based on an artificial neural network, which includes: the acquisition module is used for acquiring a fundus image and an annotation image, wherein the annotation image comprises a disc annotation image for marking out a disc area and a cup annotation image for marking out a cup area; the image preprocessing module is used for preprocessing the fundus image to obtain a preprocessed fundus image; a blood vessel region detection module that performs blood vessel region detection on the preprocessed fundus image to form a blood vessel image; a mixed weight generation module, which performs mixed weighting on the optic disc labeling image and the blood vessel image to generate a mixed weight distribution map; and the model training module is used for training an artificial neural network based on the preprocessed fundus image, the labeled image and the mixed weight distribution map, wherein the weight of the optic disc region is greater than that of a non-optic disc region when the mixed weight is carried out, and the weight of the blood vessel region is greater than that of the non-blood vessel region. According to the method, the artificial neural network is trained on the basis of the preprocessed fundus image, the labeled image and the mixed weight distribution map, so that the blood vessel region and the optic disc region can be considered in the training of the artificial neural network, the learning of small blood vessel walking is optimized, and the imbalance of positive and negative samples is restrained, so that the accuracy of the artificial neural network in extracting the glaucoma image features can be improved.

In addition, in the training system for artificial neural network-based glaucoma image feature extraction according to the second aspect of the present disclosure, optionally, the optic disc region in the optic disc labeling image is dilated to form an optic disc dilated image; expanding the vessel region in the vessel image to form a vessel expanded image. In this case, a disc expansion image containing a region near the disc and a blood vessel expansion image containing a blood vessel boundary are obtained by the expansion processing.

In addition, in the training system for artificial neural network-based glaucoma image feature extraction according to the second aspect of the present disclosure, the optic disc expansion image and the blood vessel expansion image are optionally mixed and weighted to generate the mixed weight distribution map. Therefore, the accuracy of the artificial neural network for extracting the glaucoma image features can be further improved.

In addition, in the training system for glaucoma image feature extraction based on an artificial neural network according to the second aspect of the present disclosure, optionally, in the training process, coefficients of a loss function in the artificial neural network are obtained based on the mixed weight distribution map, and the artificial neural network is trained based on the loss function. In this case, the artificial neural network is optimized based on the loss function coefficient obtained from the mixed weight distribution map, and imbalance between the positive and negative samples can be suppressed, whereby the accuracy of the artificial neural network in extracting the glaucoma image features can be further improved.

In addition, in the training system for artificial neural network-based glaucoma image feature extraction according to the second aspect of the present disclosure, optionally, the mixed weight distribution map includes a fundus region and a background region, and the weight of the background region is set to zero. Therefore, the interference of the background area on the recognition of the glaucoma image characteristics by the artificial neural network can be reduced.

In addition, in the training system for glaucoma image feature extraction based on an artificial neural network according to the second aspect of the present disclosure, optionally, the vessel region detection is performed based on Frangi filtering to form the vessel image. Therefore, the blood vessel region can be automatically identified, and the subsequent artificial neural network can conveniently identify and process the blood vessel region.

In addition, in the training system for artificial neural network-based glaucoma image feature extraction according to the second aspect of the present disclosure, optionally, the mixed weight distribution map includes an intra-optic disc blood vessel region, an intra-optic disc non-blood vessel region, an extra-optic disc blood vessel region, and an extra-optic disc non-blood vessel region. Thus, the artificial neural network can more accurately extract the glaucoma image features of each region in the fundus image.

In the training system for artificial neural network-based glaucoma image feature extraction according to the second aspect of the present disclosure, the weight of the optic disc region may be a first weight, the weight of the non-optic disc region may be a second weight, the weight of the blood vessel region may be a third weight, and the weight of the non-blood vessel region may be a fourth weight, so that the weight of the intra-optic disc blood vessel region may be the first weight multiplied by the third weight, the weight of the intra-optic disc non-blood vessel region may be the first weight multiplied by the fourth weight, the weight of the extra-optic disc blood vessel region may be the second weight multiplied by the third weight, and the weight of the extra-optic disc non-blood vessel region may be the second weight multiplied by the fourth weight. Therefore, the weight values of the intra-optic disc blood vessel region, the intra-optic disc non-blood vessel region, the extra-optic disc blood vessel region and the extra-optic disc non-blood vessel region can be respectively obtained according to the weights of the optic disc region, the non-optic disc region, the blood vessel region and the non-blood vessel region.

According to the disclosure, a training method and a training system for glaucoma image feature extraction based on an artificial neural network are provided, which can improve the accuracy of glaucoma image feature extraction by the artificial neural network.

Drawings

Embodiments of the present disclosure will now be explained in further detail, by way of example only, with reference to the accompanying drawings, in which:

fig. 1 shows a schematic diagram of an electronic device according to an embodiment of the present disclosure.

Fig. 2 shows a block diagram of a training system for artificial neural network-based glaucoma image feature extraction according to an embodiment of the present disclosure.

Fig. 3 is a schematic diagram illustrating an annotation image formed by performing annotation on a fundus image according to an embodiment of the present disclosure.

Fig. 4 is a schematic diagram showing a blood vessel image formed by performing blood vessel region detection on a preprocessed fundus image according to an embodiment of the present disclosure.

Fig. 5 shows a schematic diagram of the formation of a first mixing weight distribution map according to an embodiment of the present disclosure.

FIG. 6 illustrates a block diagram of a hybrid weight generation module of a training system in accordance with an embodiment of the present disclosure.

Fig. 7 shows a schematic diagram of a disc inflation image according to an embodiment of the present disclosure.

Fig. 8 shows a schematic diagram of a vessel expansion image according to an embodiment of the present disclosure.

Fig. 9 shows a schematic diagram of the formation of a second mixing weight distribution map according to an embodiment of the present disclosure.

Fig. 10 shows a flowchart of a training method for artificial neural network-based glaucoma image feature extraction according to an embodiment of the present disclosure.

Description of the symbols:

1 … electronic device, 1a … host, 1b … display device, 1c … input device, 10 … training system, 100 … acquisition module, 200 … image preprocessing module, 300 … blood vessel region detection module, 400 … hybrid weight generation module, 410 … optic disc expansion module, 420 … vasodilation module, 500 … model training module, P1 … fundus image, P20 … optic disc labeling image, P30 … optic cup labeling image, P10 … preprocessing fundus image, P40 … blood vessel image, P21 … weighted optic disc labeling image, P41 … weighted blood vessel image, P3 … first hybrid weight distribution map, P23 … optic disc expansion image, P43 … blood vessel expansion image, P4 … second hybrid weight distribution map, a1 … optic disc region, a1'… non-disc region, a2 … optic cup region, A3 … optic disc expansion region, A3' … non-blood vessel region, 4 … a blood vessel region, a4'… non-vascular region, a5 … vascular dilation region, a5' … non-vascular dilation region, a30 … vascular region within the first disc, a30'… non-vascular region within the first disc, a31 … vascular region outside the first disc, a31' … non-vascular region outside the first disc, a40 … vascular region within the second disc, a40'… non-vascular region within the second disc, a41 … vascular region outside the second disc, a41' … non-vascular region outside the second disc.

Detailed Description

Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the following description, the same components are denoted by the same reference numerals, and redundant description thereof is omitted. In addition, the drawings are only schematic, and the ratio of the sizes of the components to each other, the shapes of the components, and the like may be different from actual ones.

Fig. 1 shows a schematic diagram of an electronic device according to an embodiment of the present disclosure. Fig. 2 shows a block diagram of a training system for artificial neural network-based glaucoma image feature extraction according to an embodiment of the present disclosure.

In some examples, referring to fig. 1 and 2, a training system (which may be referred to simply as a "training system") 10 for artificial neural network-based glaucoma image feature extraction according to the present disclosure may be implemented by means of an electronic device 1 (such as a computer). In some examples, as shown in fig. 1, the electronic device 1 may include a host 1a, a display device 1b, and an input device 1c (e.g., mouse, keyboard). The host 1a may comprise, among other things, one or more processors, memory, and a computer program stored in the memory, in which case the training system 10 may be stored as a computer program in the memory.

In some examples, the glaucoma image feature may be feature information related to glaucoma. The main pathological process of glaucoma is due to death of retinal ganglion cells and loss of axons, which leads to defect of nerve fibers at the edge of the optic disc, thereby causing morphological change of the optic disc, such as enlargement of optic disc depression, deepening of optic disc depression and the like. Clinical medical research shows that the cup-to-disc ratio (CDR) of the fundus image is a reliable index for measuring the optic disc depression. Therefore, the glaucoma image characteristic can be an optic cup and an optic disc. In other examples, the glaucoma image feature may be a cup-to-disc ratio (CDR).

In some examples, the one or more processors may include a central processing unit, an image processing unit, and any other electronic components capable of processing input data. For example, a processor may execute instructions and programs stored on a memory.

As described above, the training system 10 may be implemented by program instructions and algorithms encoded in a computer program. Additionally, training system 10 may also be stored in memory of the cloud server. In some examples, the cloud server may be leased. This can reduce the maintenance cost of the server. In other examples, the cloud server may be self-built. In this case, the memory can be arranged in a self-built server, so that the confidentiality of data is ensured, and the data leakage of a client or a patient is prevented.

In some examples, the training system 10 may use one or more artificial neural networks to extract and learn glaucoma image features in the fundus image. In some examples, the artificial neural network may be implemented by one or more processors (e.g., microprocessors, integrated circuits, field programmable gate arrays, etc.). The training system 10 may be used to receive multiple fundus images and train on the fundus images. Wherein the artificial neural network parameter values may be determined via the artificial neural network by successive iterations of the training data set. Wherein the training data set may be composed of a plurality of fundus images. In some examples, when training an artificial neural network, the pre-processed fundus image and the annotation image are used as inputs, and the output of the artificial neural network is continuously optimized by artificially setting the weight of the loss function of each region (i.e., each pixel point) of the pre-processed fundus image.

In this embodiment, the training system 10 may include an acquisition module 100, an image preprocessing module 200, a blood vessel region detection module 300, a hybrid weight generation module 400, and a model training module 500 (see fig. 2).

In some examples, the acquisition module 100 may be used to acquire fundus images and annotation images. The image preprocessing module 200 may be configured to preprocess the fundus image to obtain a preprocessed fundus image. The blood vessel region detection module 300 may be configured to perform blood vessel region detection on the preprocessed fundus image to form a blood vessel image. The blending weight generation module 400 may be configured to blend-weight the disc label image and the vessel image to generate a blending weight distribution map. The model training module 500 may train the artificial neural network based on the pre-processed fundus image, the annotated image, and the mixed weight profile. In the above example, the training system 10 may derive the hybrid weight distribution map based on the disc label image and the vessel image. In this case, the training system 10 can train the artificial neural network based on the preprocessed fundus image, the annotation image, and the mixed weight distribution map, and can allow the artificial neural network to consider both the blood vessel region and the optic disc region during training, optimize learning of small blood vessel travel, and suppress imbalance of positive and negative samples, thereby improving accuracy of the artificial neural network in extracting features of the glaucoma image.

Fig. 3 is a schematic diagram showing an annotation image formed by performing annotation on a fundus image according to an embodiment of the present disclosure, in which fig. 3(a) shows a fundus image P1, fig. 3(b) shows a disc annotation image P20, and fig. 3(c) shows a cup annotation image P30.

In some examples, as described above, training system 10 may include acquisition module 100 (see fig. 2). The acquisition module 100 may be used to acquire fundus images and annotation images. The labeling image may be an image obtained by labeling the fundus image.

In some examples, the acquisition module 100 may be used to acquire fundus images. The fundus image may be an image taken by a fundus camera or other fundus camera about the fundus of the eye. As an example of the fundus image, fig. 3(a) shows a fundus image P1 taken by a fundus camera, for example. In some examples, the fundus image may include regions of the optic disc and optic cup, but the present embodiment is not limited thereto, and in some examples, the fundus image may include only the optic disc region.

In some examples, the plurality of fundus images may constitute a training data set. The training data set may include a training set and a test set. For example, 5-20 ten thousand fundus images from a cooperative hospital with patient information removed may be selected as a training set (training set), and 5000-.

In some examples, the fundus image may be a color fundus image. The colorful fundus images can clearly present rich fundus information such as optic discs, optic cups, yellow spots, blood vessels and the like. In addition, the fundus image may be an image in an RGB mode, a CMYK mode, an Lab mode, a grayscale mode, or the like.

In some examples, the acquisition module 100 can be used to acquire an annotation image. The annotation images can include a disc annotation image and a cup annotation image. Medically, the optic disc and optic cup have a well-defined anatomical definition, i.e., the optic disc is defined as the edge of the posterior scleral foramen, bounded by the inner edge of the scleral ring; the visual cup is defined as the range from the scleral sieve plate to the retinal plane, and the small blood vessel running is taken as an important basis for identifying the visual cup area.

In other examples, the annotation image obtained by the acquisition module 100 may include only a disc annotation image.

In some examples, the optic disc annotation image or the optic cup annotation image may be used as a true value for the training of the artificial neural network. In other examples, the disc annotation image and the cup annotation image may be combined into one annotation image as a true value for artificial neural network training.

In some examples, as described above, the annotation image may be an image obtained after annotation of the fundus image. In this case, the disc-labeling image may be an image obtained after labeling the disc in the fundus image. The cup labeling image may be an image obtained by labeling a cup in the fundus image.

Specifically, the disc region in the fundus image P1 may be artificially labeled, thereby obtaining a disc-labeled image P20 (see fig. 3 (b)). In some examples, manual labeling may be performed by experienced physicians, thereby improving the accuracy of optic disc region labeling. The cup region in the fundus image P1 may be artificially annotated, thereby obtaining a cup annotation image P30 (see fig. 3 (c)). In some examples, manual labeling may be performed by experienced physicians, thereby improving the accuracy of cup region labeling.

In this embodiment, the image preprocessing module 200 may be configured to preprocess the fundus image to obtain a preprocessed fundus image. Specifically, the image preprocessing module 200 may acquire the fundus image output by the acquisition module 100, and preprocess the fundus image to obtain a preprocessed fundus image.

In some examples, the image pre-processing module 200 may crop the fundus image. In general, since fundus images acquired by the acquisition module 100 may have problems of different image formats, sizes, and the like, it is necessary to crop the fundus images so that the fundus images are converted into images of a fixed standard form. Fixed standard form may mean that the images are of the same format and consistent size. For example, in some examples, the size of the fundus images after being preprocessed may be unified into fundus images of 512 × 512 or 1024 × 1024 pixels.

In some examples, the image pre-processing module 200 may perform normalization processing on the fundus image. In some examples, the normalization process may include operations such as coordinate centering, scaling normalization, etc. of the fundus image. Therefore, the difference of different fundus images can be overcome, and the performance of the artificial neural network is improved.

In addition, in some examples, the image pre-processing module 200 may include noise reduction, graying processing, and the like on the fundus image. This can highlight the features of the cyan-eye image.

Additionally, in some examples, the image pre-processing module 200 may include scaling, flipping, translating, etc. the fundus image. In this case, the data amount of the artificial neural network training can be increased, and thus the generalization ability of the artificial neural network can be improved.

In some examples, the fundus image may also be used directly for artificial neural network training without image pre-processing.

In other examples, the fundus image may be pre-processed and then the pre-processed fundus image may be annotated.

Additionally, in some examples, the image pre-processing module 200 can obtain the annotation image output by the acquisition module 100. The pre-processing of the annotation image can be included at the same time as the pre-processing of the fundus image. Therefore, the size of the marked image and the size of the preprocessed fundus image can be kept consistent all the time, and the artificial neural network training is further facilitated.

Fig. 4 is a schematic diagram showing a blood vessel image formed by performing blood vessel region detection on a preprocessed fundus image according to an embodiment of the present disclosure, in which fig. 4(a) shows a preprocessed fundus image P10, and fig. 4(b) shows a blood vessel image P40.

In this embodiment, the training system 10 may include a vessel region detection module 300, as described above. The blood vessel region detection module 300 may be configured to perform blood vessel region detection on the preprocessed fundus image to form a blood vessel image.

As an example of the blood vessel image formation, as shown in fig. 4, blood vessel region detection is performed on the preprocessed fundus image P10 to form a blood vessel image P40 including a blood vessel region a 4.

In some examples, vessel region detection may be performed on the preprocessed fundus image P10 based on Frangi (multi-scale linear) filtering to form a vessel image P40. Specifically, the Frangi filtering is an edge detection enhanced filtering algorithm constructed based on a Hessian matrix (Hessian matrix).

In the Frangi filtering, first, the preprocessed fundus image P10 is converted into a grayscale image. Image noise reduction is performed on the preprocessed fundus image P10 using gaussian filtering. Next, a Hessian matrix (Hessian matrix) is calculated. The Hessian matrix is a squared matrix of the second partial derivative of a scalar function, which describes the local curvature of a multivariate function, and the Hessian matrix is basically of the form shown in equation (1) below:

wherein, the second partial differential in the x direction:

second partial differential in y direction:

mixed partial differential in x, y direction:

wherein f is_xy＝f_yxH is a real symmetric matrix, and two eigenvalues λ can be used₁、λ₂To construct the enhancement filter. In two-dimensional preprocessing of fundus images, the characteristic value λ₁、λ₂Can be calculated by the following formula:

since the second partial derivative is sensitive to noise, gaussian smoothing is performed when solving the Hessian matrix. The pre-processed fundus image P10 pixel point P blood vessel region response function is V (σ, P):

wherein, the sigma is a scale factor, and the scale factor sigma is a standard deviation of Gaussian smoothing when the Hessian matrix is solved. Beta may be set to 0.5 for distinguishing between lines and blocks.

Is a parameter for controlling the overall smoothness of the wire. R_BAnd s is represented by the eigenvalue λ₁、λ₂And (4) defining.

The output of the filter is maximum when the scale factor sigma is closest to the actual width of the vessel. The maximum response of each pixel point P in the preprocessed fundus image P10 with different scale factors can be taken as the final blood vessel response. In this case, the scale factor σ is closest to the actual width of the vessel, and the final vessel response is shown as the following equation (9):

wherein σ_minIs the minimum value of the scale factor σ, σ_maxIs the maximum value of the scale factor sigma.

Finally, a threshold T may be set, and the position where the vascular response is greater than T is the detected vascular region a4 (see fig. 4 (b)).

In the present disclosure, a blood vessel image including a blood vessel region is obtained by automatically detecting the blood vessel region in the preprocessed fundus image using the Frangi algorithm. In this case, the blood vessel region in the blood vessel image is conspicuous compared with the blood vessel region in the preprocessed fundus image. Therefore, the identification and processing of the blood vessel region in the later period can be facilitated.

However, examples of the present disclosure are not limited thereto, and in other examples, the blood vessel region detection for preprocessing the fundus image may be implemented using a matched filter algorithm, an adaptive contrast enhancement algorithm, a two-dimensional Gabor filter algorithm, or using other types of artificial neural networks, or the like. Therefore, the detection of the blood vessel region can be realized by selecting a proper algorithm or an artificial neural network according to different requirements.

Fig. 5 shows a schematic diagram of the formation of a first mixing weight distribution map according to an embodiment of the present disclosure. Fig. 5(a) shows a schematic diagram of the weighted disc-label image P21, fig. 5(b) shows a schematic diagram of the weighted blood vessel image P41, and fig. 5(c) shows a schematic diagram of the first mixed weight distribution map P3 generated based on the weighted disc-label image P21 and the weighted blood vessel image P41.

In this embodiment, as described above, the training system 10 may include the hybrid weight generation module 400. The blending weight generation module 400 may be configured to blend-weight the disc label image and the vessel image to generate a blending weight distribution map.

In some examples, as shown in fig. 5, the blending weight profile may be a first blending weight profile P3. The first mixed weight distribution map P3 may be obtained based on the disc-labeling image P20 (see fig. 3(b)) and the blood vessel image P40 (see fig. 4 (b)). For example, the disk annotation image P20 may be weighted to generate a weighted disk annotation image P21, the blood vessel image P40 may be weighted to generate a weighted blood vessel image P41, and the first mixed weight distribution map P3 may be generated based on the weighted disk annotation image P21 and the weighted blood vessel image P41. In this case, the artificial neural network is trained using the mixed weight distribution map, so that the artificial neural network can give consideration to both the blood vessel region and the optic disk region during training, thereby optimizing learning of small blood vessel travel and suppressing imbalance of positive and negative samples.

In some examples, the hybrid weighting process includes weighting the optic disc region by a first weight, weighting the non-optic disc region by a second weight, weighting the vessel region by a third weight, weighting the non-vessel region by a fourth weight, and weighting the intra-optic disc vessel region by the first weight multiplied by the third weight, weighting the intra-optic disc non-vessel region by the first weight multiplied by the fourth weight, weighting the extra-optic disc vessel region by the second weight multiplied by the third weight, and weighting the extra-optic disc non-vessel region by the second weight multiplied by the fourth weight. The hybrid weighting process is described in detail below with reference to the drawings.

In some examples, as shown in fig. 5, the weight of the disc area a1 is made the first weight w at the time of the hybrid weighting process₁Let the weight of the non-optic disc area A1' be the second weight w₂(see FIG. 5 (a)). The vascular region A4 is weighted by a third weight v₁The non-blood vessel region A4' has a fourth weight v₂(see FIG. 5 (b)).

In some examples, the disc region may be weighted more heavily than the non-disc region, and the vessel region may be weighted more heavily than the non-vessel region when performing the hybrid weighting process. For example, as shown in fig. 5, when performing hybrid weighting, the weight of the optic disc region a1 may be made greater than the weight of the non-optic disc region a1', and the weight of the blood vessel region a4 may be made greater than the weight of the non-blood vessel region a 4'. That is, can let w₁>w₂，v₁>v₂. In the present disclosure, the optic disc region may also be referred to as an intra-disc region, and the non-optic disc region may also be referred to as an extra-disc region.

In some examples, the first mixed weight distribution map P3 obtained based on the disc-labeling image P20 and the blood vessel image P40 includes four mixed regions of a blood vessel region a30 within the first disc, a non-blood vessel region a30 'within the first disc, a blood vessel region a31 outside the first disc, and a non-blood vessel region a31' outside the first disc (see fig. 5 (c)). In this case, for example, the weight of blood vessel region a30 within the first optic disc may be obtained by the product of the weight of region a1 within the optic disc and the weight of blood vessel region a 4. This enables more accurate recognition of the glaucoma image feature in each region in the fundus image.

In some examples, as described above, the weight of the disc area a1 is made the first weight w when the hybrid weighting process is performed₁Let the weight of the non-optic disc area A1' be the second weight w₂The weight of the blood vessel region A4 is a third weight v₁The non-blood vessel region A4' has a fourth weight v₂. The weight of the blood vessel region a30 in the first disc is the first weight multiplied by the third weight, i.e. w₁v₁. The non-vascular region A30' within the first disc is weighted by the first weight multiplied by the fourth weight, i.e. w₁v₂. The vascular region A31 outside the first optic disc has a weight which is the second weight multiplied by a third weight, i.e. w₂v₁. The non-vascular region A31' outside the first optic disc is weighted by the second weight multiplied by the fourth weight, i.e. w₂v₂(see FIG. 5 (c)). Therefore, the weight values of the intra-optic disc blood vessel region, the intra-optic disc non-blood vessel region, the extra-optic disc blood vessel region and the extra-optic disc non-blood vessel region can be respectively obtained according to the weights of the optic disc region, the non-optic disc region, the blood vessel region and the non-blood vessel region.

In some examples, the blending weight distribution map may be divided based on the eyeball contour when the blending weighting process is performed. In this case, the blending weight profile may include a fundus region and a background region. Wherein the fundus region may be a region within the contour of the eyeball. The fundus region may include four blended regions of a first intradisc vessel region, a first intradisc non-vessel region, a first extradisc vessel region, and a first extradisc non-vessel region of the first blended weight profile. The background region may be a region outside the contour of the eyeball. The background region may be a partial region of the non-vascular region outside the first optic disc. When training the artificial neural network, the weight of the background region may be made zero. Therefore, the interference of the background area on the extraction of the glaucoma image features by the artificial neural network can be reduced in the training process.

In some examples, as described above, the optic disc region of the hybrid weight profile has a weight, and as the optic disc region contains an optic cup region, the optic cup region of the hybrid weight profile thus has a weight.

In other examples, the blending weight generation module 400 may weight based on only the disc label images.

In other examples, the hybrid weight generation module 400 may weight based on the vessel image alone.

FIG. 6 illustrates a block diagram of a hybrid weight generation module of a training system in accordance with an embodiment of the present disclosure. Fig. 7 is a schematic diagram showing a disc-dilated image according to an embodiment of the present disclosure, in which fig. 7(a) shows a disc-annotation image P20, and fig. 7(b) shows a disc-dilated image P23.

In some examples, the hybrid weight generation module 400 may include a disk dilation module 410 (see fig. 6).

In some examples, in the disc dilation module 410, a disc region in the disc annotation image may be dilated to form a disc dilation image. The disc expansion image includes a disc expansion region. The disc expansion region may include a disc region and a disc vicinity region.

For example, as shown in fig. 7(a) and 7(b), the disc region in the disc annotation image P20 is dilated to form a disc dilated image P23. The disc-dilated image P23 includes a disc dilated area A3. The disc expansion region A3 of the disc expansion image P23 corresponds to the disc region a1 of the disc annotation image P20. In fig. 7(b), a3' is a non-optic disc expansion region. The disc expansion region may include a disc region and a disc vicinity region. In this case, since the area near the optic disc affects the segmentation of the optic cup or optic disc, and thus the extraction of the glaucoma image features, the optic disc expansion image is acquired by the expansion process so as to perform the correlation process based on the optic disc expansion image, thereby improving the accuracy of the extraction of the glaucoma image features.

Fig. 8 is a schematic diagram showing a blood vessel expansion image according to an embodiment of the present disclosure, in which fig. 8(a) shows a blood vessel image P40, and fig. 8(b) shows a blood vessel expansion image P43.

In some examples, the hybrid weight generation module 400 may include a vessel dilation module 420 (see fig. 6).

In some examples, in the vessel expansion module 420, a vessel region in the vessel image may be expanded to form a vessel expansion image. The vessel expansion image includes a vessel expansion region. The vessel expansion region may include a vessel region and a vessel vicinity region. For example, as shown in fig. 8(a) and 8(b), the blood vessel region a4 in the blood vessel image P40 may be dilated to form a blood vessel dilated image P43. The blood vessel expansion image P43 includes a blood vessel expansion region a 5. The blood vessel expansion region a5 of the blood vessel expansion image P43 corresponds to the blood vessel region a4 of the blood vessel image P40. In fig. 8(b), a5' is a non-vascular expansion region. In this case, an error in detecting the boundary of the blood vessel based on the blood vessel detection algorithm can be reduced by the dilation process.

Fig. 9 is a schematic diagram showing formation of a second mixed weight distribution map according to the embodiment of the present disclosure, in which fig. 9(a) shows a schematic diagram of weighting the disk-dilated image P23, fig. 9(b) shows a schematic diagram of weighting the blood vessel dilated image P43, and fig. 9(c) shows a schematic diagram of generating the second mixed weight distribution map P4 based on the weighted disk-dilated image P23 and the weighted blood vessel dilated image P43.

In other examples, the blending weight generation module 400 may be configured to blend-weight the disc-expansion image and the vessel-expansion image to generate a blending weight distribution map. That is, the hybrid weight distribution map may be generated by hybrid weighting of the disk-expansion image and the blood-vessel expansion image. In this case, the artificial neural network is trained based on the mixed weight distribution map generated by the dilated optic disc image and the dilated blood vessel image, so that the error of the blood vessel detection algorithm on the blood vessel boundary can be reduced.

For example, in some examples, as shown in fig. 9, the blending weight profile may be a second blending weight profile P4. Specifically, the disk inflation image P23 and the blood vessel inflation image P43 may be weighted separately (see fig. 9(a) and 9(b)), and the weighted disk inflation image P23 and the weighted blood vessel inflation image P43 may be mixed and weighted to generate the second mixed weight distribution map P4.

In some examples, as shown in fig. 9, based on the four basic regions of the expanded optic disc expansion region A3, non-optic disc expansion region A3', vessel expansion region a5, and non-vessel expansion region a5', the second hybrid weight distribution map P4 may include a vessel region a40 inside the second optic disc, a non-vessel region a40 'inside the second optic disc, a vessel region a41 outside the second optic disc, and a non-vessel region a41' outside the second optic disc.

In some examples, the disc expansion area a3 may be weighted by a first weight w₁'let the weight of the non-optic disc expansion area A3' be the second weight w₂', the expanded vascular zone A5 is weighted by a third weight v₁', the non-vascular expansion region A5' is weighted by a fourth weight v₂'. The weight of the blood vessel region a40 in the second disc is the first weight multiplied by the third weight, i.e. w₁'v₁'. The weight of the non-vascular region A40 'in the secondary video disc is the first weight multiplied by the fourth weight, i.e. w'₁v₂'. The vascular region A41 outside the second optic disc has a weight which is the second weight multiplied by a third weight, i.e. w₂'v₁'. The non-vascular region A41' outside the second optic disc is weighted by the second weight multiplied by a fourth weight, i.e. w₂'v₂' (see FIG. 9 (c)).

In the present disclosure, the difference between the second hybrid weight profile P4 and the first hybrid weight profile P3 is: the first mixed weight distribution map P3 is a mixed weight distribution map generated based on the optic disc labeling image and the blood vessel image, and the second mixed weight distribution map P4 is a mixed weight distribution map generated based on the optic disc expansion image and the blood vessel expansion image. Therefore, the mixing and weighting process of the second mixing and weighting profile P3 can be referred to the mixing and weighting process of the first mixing and weighting profile P4, and will not be described in detail.

Examples of the present disclosure are not limited thereto, and for example, a mixed weight distribution map may be generated based on the optic disc labeling image and the blood vessel expansion image, and may also be generated based on the optic disc expansion image and the blood vessel image.

In this embodiment, the training system 10 may include a model training module 500, as described above. The model training module 500 may include an artificial neural network. The model training module 500 may train the artificial neural network based on the pre-processed fundus image, the annotated image, and the mixed weight profile. Among other things, the pre-processed fundus image may be generated by the image pre-processing module 200. The annotation image can be generated by the acquisition module 100. The blending weight profile may be generated by the blending weight generation module 400.

Specifically, the optic disc annotation image and/or the optic cup annotation image included in the annotation image can be used as a true value, each pixel point of the preprocessed fundus image is predicted, and the loss function weight (i.e., the coefficient of the loss function) of each pixel point in the preprocessed fundus image is distributed by means of the mixed weight distribution map. Training the artificial neural network based on the loss function, and optimizing the output of the artificial neural network to obtain an optimal model of the artificial neural network. In this case, the artificial neural network has good segmentation accuracy and generalization capability and can automatically extract the glaucoma image features.

In some examples, as described above, the glaucoma image feature may be a cup-to-disc ratio (CDR), in which case the cup-to-disc ratio of the fundus image may be predicted based on the optimal model, and a possible glaucoma lesion in the fundus image may be accurately identified.

But examples of the present disclosure are not limited thereto and the artificial neural network may be replaced with other image feature extraction models. Preferably, other image feature extraction models may employ UNet or its modified type as an artificial neural network for glaucoma image feature extraction.

In the present disclosure, the loss function may be used to calculate the loss, measure the goodness of the model prediction. The difference between the predicted value and the true value of the model based on the artificial neural network with respect to a single sample can be referred to as loss. The smaller the loss, the better the model. A single sample in the present invention may refer to each pixel point in the pre-processed fundus image.

In some examples, the loss function may use a predefined loss function, which may be a cross-entropy loss function, a Dice loss function, or the like, in some examples. The cross entropy loss function is a function for measuring the difference between the real distribution and the predicted distribution, and the Dice loss function is a set similarity measurement function.

Specifically, taking the cross entropy loss function as an example, the loss function of each pixel point in the preprocessed fundus image is:

wherein c represents the prediction category of each pixel point of the preprocessed fundus image, and the prediction category comprises two categories of a cup type or a optic disc type. (i, j) represents coordinates of pixel points in the preprocessed fundus image.

The value of the pixel point with the coordinate (i, j) in the optic cup labeling image or the optic disc labeling image is represented as the real value of the pixel point with the coordinate (i, j) in the preprocessed fundus image,

and (3) representing the predicted value of the pixel point with the coordinate (i, j) in the preprocessed fundus image. w is a_cAre the weights of the classes.

In some examples, a mixed weight profile may be used to assign weights to the loss functions for various pixel points in the pre-processed fundus image. As described above, the mixed weight profile may include a fundus region and a background region. Wherein the fundus region may be a region within the contour of the eyeball. The fundus region may include four blended regions of a first intradisc vessel region, a first intradisc non-vessel region, a first extradisc vessel region, and a first extradisc non-vessel region of the first blended weight profile. The background region may be a region outside the contour of the eyeball. The background region may be a partial region of the non-vascular region outside the first optic disc. In some examples, the weight of the optic disc region may be a first weight, the weight of the non-optic disc region may be a second weight, the weight of the blood vessel region may be a third weight, the weight of the non-blood vessel region may be a fourth weight, and the weight of the background region may be zero (i.e., the weight of the portion of the non-blood vessel region outside the first optic disc that belongs to the background region may be zero). The value of each pixel point in the mixed weight distribution map (i.e., the weight of the loss function of each pixel point in the preprocessed fundus image) can be represented by the following formula (11):

in the formula, w_i,jIs the weight, p, of a pixel point whose coordinate is (i, j)_i,jIs pixel point (i, j), w₁Is the weight (first weight), w, of a pixel point in the video disc₂Is the weight (second weight), v, of the pixel outside the disc₁Is the weight (third weight) of the pixel point of the vessel region, v₂The weight (fourth weight) of the pixel point of the non-vessel region, R₁Set of pixel points, R, for a vessel region within the optic disc₂Set of pixels, R, being non-vascular regions within the optic disc₃Set of pixel points, R, for the vascular region outside the optic disc₄Set of pixels, R, being non-vascular regions outside the optic disc₅Is a set of pixels in the background area.

In some examples, the loss function L of the artificial neural network may be obtained based on the weight of the loss function of each pixel:

L＝∑_i,j(w_i,j*loss_i,j) … … type (12)

Wherein, w_i,jIs the weight, loss, of the pixel point with coordinates (i, j)_i,jIs a loss function with coordinates of (i, j) pixel points. Therefore, the artificial neural network can be trained on the basis of the loss function so as to optimize the output of the artificial neural network and further obtain the optimal model.

In some examples, the parameters of the artificial neural network may be optimized by using a minimum gradient descent method, and the adjustment is performed according to the direction in which the loss function descends most quickly. Thus, the training system 10 can be optimized by means of the coefficients in the loss function. In other examples, parameter optimization may be performed using a random gradient descent method.

In some examples, the model training module 500 may extract glaucoma image features in the fundus image using the optimal model to predict lesions that may be present in the fundus image. In some examples, the trained artificial neural network may be used to identify fundus images in the test set, resulting in an average identification accuracy of, for example, up to 90% or more. As can be seen from this, the training system 10 according to the present embodiment can obtain an improved accuracy of determining glaucoma lesion while taking into account fundus clinical situations.

The training method for glaucoma image feature extraction based on the artificial neural network of the present disclosure is described in detail below with reference to fig. 10. The training method for glaucoma image feature extraction based on the artificial neural network can be simply referred to as a training method. The training method according to the present disclosure is applied to the training system 10 described above. Fig. 10 shows a flowchart of a training method for artificial neural network-based glaucoma image feature extraction according to an embodiment of the present disclosure.

In this embodiment, the training method for glaucoma image feature extraction based on the artificial neural network may include the following steps: preparing a fundus image and an annotation image (step S100), preprocessing the fundus image to form a preprocessed fundus image (step S200), performing blood vessel region detection on the preprocessed fundus image to form a blood vessel image (step S300), forming a mixed weight distribution map based on the annotation image and the blood vessel image (step S400), and training an artificial neural network based on the preprocessed fundus image, the annotation image, and the mixed weight distribution map (step S500). In this case, the artificial neural network is trained based on the preprocessed fundus image, the annotation image, and the mixed weight distribution map, so that the artificial neural network can give consideration to both the blood vessel region and the optic disc region in the training, the learning of small blood vessel walking is optimized, and the imbalance of positive and negative samples is suppressed. Therefore, the accuracy of the artificial neural network for extracting the glaucoma image features can be improved.

In step S100, a fundus image may be prepared. The fundus image may be an image of the fundus captured by a fundus camera or other fundus photographing apparatus, may be a color fundus image, and may also be an image of an RGB mode, a CMYK mode, an Lab mode, a gray scale mode, or the like. For a detailed description, reference may be made to the obtaining module 100, which is not described herein again.

In step S100, an annotation image may be prepared. In some examples, the annotation image can include a disc annotation image and a cup annotation image. The labeling image can be manually labeled by experienced doctors to the optic disc area and the optic cup area in the fundus image, so that the optic disc labeling image and the optic cup labeling image are obtained. Therefore, the accuracy of labeling the optic disc region and the optic cup region can be improved. For a detailed description, reference may be made to the obtaining module 100, which is not described herein again.

In step S200, the fundus image may be preprocessed to obtain a preprocessed fundus image. In some examples, during pre-processing, the fundus image may be cropped, normalized, etc. Therefore, the fundus images can be converted into images in a fixed standard form, the difference of different fundus images can be overcome, and the performance of the artificial neural network is improved. In some examples, in the preprocessing, the fundus image may be subjected to noise reduction, graying processing. This can highlight the features of the cyan-eye image. In some examples, during the preprocessing, the fundus image may be scaled, flipped, translated, and so on, thereby increasing the data amount of the artificial neural network training and improving the generalization capability of the artificial neural network. For a detailed description, reference may be made to the image preprocessing module 200, which is not described herein again.

In step S300, blood vessel region detection may be performed on the preprocessed fundus image to form a blood vessel image including a blood vessel region. In some examples, vessel region detection may be performed based on Frangi filtering to form a vessel image. In this case, the blood vessel region in the blood vessel image is conspicuous compared with the blood vessel region in the preprocessed fundus image. Therefore, the identification and processing of the blood vessel region in the later period can be facilitated. In other examples, the filtering may be implemented using a matched filtering algorithm, an adaptive contrast enhancement algorithm, a two-dimensional Gabor filtering algorithm, or using other types of artificial neural networks. Therefore, the detection of the blood vessel region can be realized by selecting a proper algorithm or an artificial neural network according to different requirements. For details, reference may be made to the blood vessel region detection module 300, which is not described herein again.

In step S400, the disc annotation image obtained in step S100 and the blood vessel image obtained in step S300 may be subjected to hybrid weighting to generate a hybrid weight distribution map. In this case, the artificial neural network is trained using the mixed weight distribution map, so that the artificial neural network can give consideration to both the blood vessel region and the optic disk region during training, thereby optimizing learning of small blood vessel travel and suppressing imbalance of positive and negative samples.

In step S400, the blending weight distribution map may include four blending regions of an intra-optic disc blood vessel region, an intra-optic disc non-blood vessel region, an extra-optic disc blood vessel region, and an extra-optic disc non-blood vessel region. In some examples, the mixed weight distribution map may include a fundus region and a background region, and the weight of the background region may be set to zero, thereby reducing the interference of the background region on the extraction of the glaucoma image features by the artificial neural network during the training process. For details, reference may be made to the hybrid weight generation module 400, which is not described herein again.

In step S400, in some examples, the weight of the optic disc region may be a first weight, the weight of the non-optic disc region may be a second weight, the weight of the blood vessel region may be a third weight, and the weight of the non-blood vessel region may be a fourth weight. The weight of the vessel region within the optic disc is the first weight multiplied by the third weight. The weight of the non-vascular region within the optic disc is the first weight multiplied by the fourth weight. The weighting of the vascular region outside the optic disc is the second weight multiplied by the third weight. The non-vascular region outside the optic disc is weighted by the second weight multiplied by the fourth weight. Therefore, the weight values of the intra-optic disc blood vessel region, the intra-optic disc non-blood vessel region, the extra-optic disc blood vessel region and the extra-optic disc non-blood vessel region can be respectively obtained according to the weights of the optic disc region, the non-optic disc region, the blood vessel region and the non-blood vessel region. In some examples, the weight of the optic disc region may be made greater than the weight of the non-optic disc region, and the weight of the vessel region may be made greater than the weight of the non-vessel region. For details, reference may be made to the hybrid weight generation module 400, which is not described herein again.

In step S400, in other examples, the disc area in the disc label image may be expanded to form a disc expanded image including the disc area. Because the area near the optic disc affects the segmentation of the optic cup or the optic disc and further affects the extraction of the glaucoma image characteristics, the optic disc expansion image is obtained through expansion processing, so that the subsequent related processing is carried out based on the optic disc expansion image, and the accuracy of the extraction of the glaucoma image characteristics is improved. For details, reference may be made to the disc expansion module 410, which is not described herein again.

In step S400, in other examples, a blood vessel region in the blood vessel image may be expanded to form a blood vessel expanded image including the blood vessel region. In this case, an error in detecting the boundary of the blood vessel based on the blood vessel detection algorithm can be reduced by the dilation process. For details, reference may be made to the vessel expansion module 420, which is not described in detail herein.

In step S400, in other examples, the above-mentioned optic disc expansion image and blood vessel expansion image may be mixed and weighted to generate a mixed weight distribution map. In this case, the artificial neural network is trained by using the mixed weight distribution map generated based on the dilated optic disc image and the dilated blood vessel image, so that the error of the blood vessel detection algorithm on the blood vessel boundary can be reduced. For details, reference may be made to the hybrid weight generation module 400, which is not described herein again.

In step S500, an artificial neural network may be trained based on the preprocessed fundus image, the annotation image, and the blending weight profile. The pre-processed fundus image may be generated by step S200. The annotation image can be generated by step S100. The mixing weight profile may be generated by step S400. In some examples, the loss function weights, i.e., coefficients of the loss functions, of the respective pixel points in the preprocessed fundus image may be assigned by means of a mixed weight profile. Training the artificial neural network based on the loss function, and optimizing the output of the artificial neural network to obtain an optimal model of the artificial neural network. In this case, the artificial neural network has good segmentation accuracy and generalization capability and can automatically extract the glaucoma image features. For details, see model training module 500, which is not described herein.

While the present disclosure has been described in detail above with reference to the drawings and the embodiments, it should be understood that the above description does not limit the present disclosure in any way. Those skilled in the art can make modifications and variations to the present disclosure as needed without departing from the true spirit and scope of the disclosure, which fall within the scope of the disclosure.

Claims (10)

1. A glaucoma image feature extraction training method based on an artificial neural network is characterized in that,

the method comprises the following steps:

a fundus image and an annotation image are prepared,

the labeling image comprises a video disc labeling image for labeling a video disc area and a video cup labeling image for labeling a video cup area;

preprocessing the fundus image to obtain a preprocessed fundus image, and generating a blood vessel image containing a blood vessel region according to a blood vessel detection result;

performing mixed weighting on the optic disc labeling image and the blood vessel image to generate a mixed weight distribution map; and is

Training an artificial neural network based on the pre-processed fundus image, the annotation image, and the mixed weight profile,

wherein, when the mixed weighting is carried out, the weight of the optic disc area is greater than that of the non-optic disc area, and the weight of the blood vessel area is greater than that of the non-blood vessel area.

2. Training method according to claim 1,

expanding the optic disc region in the optic disc labeling image to form an optic disc expanded image;

expanding the vessel region in the vessel image to form a vessel expanded image;

blending the disc-expanded image and the vessel-expanded image to generate the blended weight profile.

3. Training method according to claim 1,

in the training process, obtaining the coefficient of a loss function in the artificial neural network based on the mixed weight distribution diagram, and training the artificial neural network based on the loss function.

4. Training method according to claim 1,

the mixed weight distribution map comprises an intra-optic disc blood vessel region, an intra-optic disc non-blood vessel region, an extra-optic disc blood vessel region and an extra-optic disc non-blood vessel region;

and setting the weight of the optic disc region as a first weight, the weight of the non-optic disc region as a second weight, the weight of the blood vessel region as a third weight, and the weight of the non-blood vessel region as a fourth weight, wherein the weight of the blood vessel region in the optic disc is the first weight multiplied by the third weight, the weight of the non-blood vessel region in the optic disc is the first weight multiplied by the fourth weight, the weight of the blood vessel region outside the optic disc is the second weight multiplied by the third weight, and the weight of the non-blood vessel region outside the optic disc is the second weight multiplied by the fourth weight.

5. Training method according to claim 1,

vessel region detection is performed based on Frangi filtering to form the vessel image.

6. A glaucoma image feature extraction training system based on an artificial neural network is characterized in that,

the method comprises the following steps:

the acquisition module is used for acquiring a fundus image and an annotation image, wherein the annotation image comprises a disc annotation image for marking out a disc area and a cup annotation image for marking out a cup area;

the image preprocessing module is used for preprocessing the fundus image to obtain a preprocessed fundus image;

a blood vessel region detection module that performs blood vessel region detection on the preprocessed fundus image to form a blood vessel image;

a mixed weight generation module, which performs mixed weighting on the optic disc labeling image and the blood vessel image to generate a mixed weight distribution map; and

a model training module for training an artificial neural network based on the preprocessed fundus image, the labeled image and the mixed weight distribution map,

wherein, when the mixed weighting is carried out, the weight of the optic disc area is greater than that of the non-optic disc area, and the weight of the blood vessel area is greater than that of the non-blood vessel area.

7. Training system according to claim 6,

expanding the optic disc region in the optic disc labeling image to form an optic disc expanded image;

expanding the vessel region in the vessel image to form a vessel expanded image;

blending the disc-expanded image and the vessel-expanded image to generate the blended weight profile.

8. Training system according to claim 6,

in the training process, obtaining the coefficient of a loss function in the artificial neural network based on the mixed weight distribution diagram, and training the artificial neural network based on the loss function.

9. Training system according to claim 6,

vessel region detection is performed based on Frangi filtering to form the vessel image.

10. Training system according to claim 6,

the mixed weight distribution map comprises an intra-optic disc blood vessel region, an intra-optic disc non-blood vessel region, an extra-optic disc blood vessel region and an extra-optic disc non-blood vessel region;

and setting the weight of the optic disc region as a first weight, the weight of the non-optic disc region as a second weight, the weight of the blood vessel region as a third weight, and the weight of the non-blood vessel region as a fourth weight, wherein the weight of the blood vessel region in the optic disc is the first weight multiplied by the third weight, the weight of the non-blood vessel region in the optic disc is the first weight multiplied by the fourth weight, the weight of the blood vessel region outside the optic disc is the second weight multiplied by the third weight, and the weight of the non-blood vessel region outside the optic disc is the second weight multiplied by the fourth weight.

Images