A method and device for optimizing the resolution of optical coherence tomography

Abstract

The present invention relates to the field of image processing, and discloses a method and device for optimizing the resolution of optical coherence tomography. The method includes: obtaining first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device; inputting the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network, obtaining a training processing result corresponding to the convolutional neural network, where the training processing result includes spectral feature data corresponding to the first image data and a target neural network that has completed training; obtaining second image data to be processed, where the second image data includes narrowband spectral data corresponding to the retina collected by an OCT device; inputting the second image data into the target neural network to perform broadening processing on the second image through the target neural network, obtaining a broadening processing result corresponding to the second image data. It can be seen that implementing the present invention can improve the longitudinal resolution of OCT images.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method and device for optimizing the resolution of optical coherence tomography. Background Art

[0002] Optical Coherence Tomography (OCT): According to the principle of optical coherence, through its Michelson interferometer, it can selectively receive the backscattered light of different tissue depths, so as to obtain information such as the structure and blood flow of different strong tissues. As a non-contact and non-invasive ophthalmic imaging technology, OCT is widely used in retinal imaging.

[0003] In the field of retinal imaging, the OCT system uses a short-time domain coherent light source and uses the coherent gating technology for imaging. Therefore, its longitudinal resolution is strictly determined by the coherence length of the light source. Theoretical analysis shows that there is a relationship between the coherence length l of the light source and the coherence time τ: l = cτ, where c is the speed of light. The coherence time of the light source and the light source spectrum are in a Fourier transform relationship. In this way, by changing the shape of the light source spectrum, that is, changing the coherence time and coherence length of the light source, the longitudinal resolution of the OCT system is also changed, achieving the purpose of clearer imaging. However, in actual research, how to improve the longitudinal resolution of OCT images so as to achieve the technical effect of improving image clarity is still an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a method and device for optimizing the resolution of optical coherence tomography, which can improve the longitudinal resolution of OCT images.

[0005] In order to solve the above technical problems, a first aspect of the present invention discloses a method for optimizing the resolution of optical coherence tomography, the method comprising:

[0006] Obtain first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device;

[0007] Input the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network, and obtain a training processing result corresponding to the convolutional neural network, where the training processing result includes spectral feature data corresponding to the first image data and a target neural network that has completed training;

[0008] Obtain second image data to be processed, where the second image data includes narrowband spectral data corresponding to the retina collected by an OCT device;

[0009] Input the second image data into the target neural network to perform a widening process on the second image data through the target neural network, and obtain a widening process result corresponding to the second image data.

[0010] As an optional implementation manner, in the first aspect of the present invention, the pre-set convolutional neural network includes a first network and a second network;

[0011] The inputting the first image data into the pre-set convolutional neural network to perform a training process on the convolutional neural network and obtain a training process result corresponding to the convolutional neural network includes:

[0012] Input the first image data into the first network to perform a first preset processing operation on the first image data through the first network, and obtain a first output result corresponding to the first image data;

[0013] Input the first output result into the second network to perform a second preset processing operation on the first output result through the second network, and obtain a second output result corresponding to the first output result;

[0014] When it is determined that the second output result meets the set network training requirements, determine the second output result and the convolutional neural network that has completed training as the training process result;

[0015] Wherein, the first preset processing operation includes at least one of image convolution, image upsampling, image downsampling, pooling, rectified linear unit, and feature transformation;

[0016] The second preset processing operation includes at least one of feature extraction, region selection and generation, window classification, position refinement, NMS processing, detection box detection, and detection box recognition.

[0017] As an optional implementation manner, in the first aspect of the present invention, the first network includes a contracting path and an expanding path; both the contracting path and the expanding path are composed of a preset number of different convolutional layers repeatedly applied a preset number of times; and each convolutional layer is followed by a rectified linear unit;

[0018] The contracting path is used to perform downsampling processing on the image input to the contracting path; the expanding path is used to perform upsampling processing on the image input to the expanding path;

[0019] The inputting the first image data into the first network to perform a first preset processing operation on the training set through the first network and obtain a first output result corresponding to the first image data includes:

[0020] Input the first image data into the contraction path and the expansion path respectively to obtain a contraction processing result corresponding to the contraction path and an expansion processing result corresponding to the expansion path;

[0021] Perform feature mapping and feature transformation on the contraction processing result and the expansion processing result to obtain a first output result corresponding to the first image data.

[0022] As an optional implementation manner, in the first aspect of the present invention, the second network includes a region proposal network and a detection network;

[0023] Inputting the first output result into the second network to perform a second preset processing operation on the first output result through the second network to obtain a second output result corresponding to the first output result includes:

[0024] Input the first output result into the region proposal network, perform a feature extraction operation on the first output result through a preset image processing function, generate at least one candidate region according to the obtained feature extraction result, and then perform a preset adjustment operation on each candidate region to obtain a preset adjustment result corresponding to the candidate region. The preset adjustment operation includes window classification and position refinement;

[0025] According to a preset NMS algorithm, perform merging and deletion on all the preset adjustment results to obtain a plurality of proposal boxes corresponding to all the preset adjustment results;

[0026] According to the detection network, perform detection and recognition operations on the plurality of proposal boxes to obtain a detection and recognition result corresponding to the plurality of proposal boxes as the second output result.

[0027] As an optional implementation manner, in the first aspect of the present invention, the calculation formula corresponding to the image processing function is as follows:

[0028]

[0029] where p i is the probability that the candidate region is predicted as a target;

[0030] The definition label corresponding to the preset reference standard is:

[0031]

[0032] The vector t i ={t x , t y , t w , t h} represents the parameter coordinates of a pre-determined bounding box; is the coordinate vector of the reference standard bounding box corresponding to the positive label candidate area; the output corresponding to the cls layer is {p i}, and the output corresponding to the reg layer is {u i}; the is the logarithmic loss of two classes p i , .

[0033] As an alternative implementation, in the first aspect of the present invention, the first image data includes multiple sub-images, and there is a corresponding data annotation image for each sub-image;

[0034] Before inputting the first image data into a pre-set convolutional neural network, the method further includes:

[0035] Performing a partitioning process on all the sub-images to obtain a training set and a validation set; the training data in the training set is used to train the pre-set convolutional neural network; the validation data in the validation set is used to verify the training effect of the convolutional neural network;

[0036] The second image data is used as an actual test set to perform a test on the target neural network.

[0037] As an alternative implementation, in the first aspect of the present invention, the broadened processing result includes broadened spectral data corresponding to the second image data; the second image data includes the longitudinal resolution before the second image data is input into the target neural network, denoted as the first resolution; the broadened processing result further includes the longitudinal resolution after the second image data is input into the target neural network, denoted as the second resolution;

[0038] The method further includes:

[0039] Comparing the narrowband spectral data with the broadened spectral data to obtain spectral broadening information corresponding to the narrowband spectral data and the broadened spectral data;

[0040] Comparing the first resolution with the second resolution to obtain resolution optimization information corresponding to the first resolution and the second resolution;

[0041] Based on the spectral broadening information and the resolution optimization information, determining whether the target neural network meets a preset resolution improvement requirement; when the determination result is yes, determining that the target neural network is a qualified neural network that meets the resolution improvement requirement.

[0042] The second aspect of the present invention discloses a resolution optimization device for optical coherence tomography, and the device includes:

[0043] An acquisition module, configured to acquire first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device;

[0044] A training module, configured to input the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network, and obtain a training processing result corresponding to the convolutional neural network, where the training processing result includes spectral feature data corresponding to the first image data and a target neural network that has completed training;

[0045] The acquisition module is further configured to acquire second image data to be processed, where the second image data includes narrowband spectral data corresponding to the retina collected by an OCT device;

[0046] An expansion processing module, configured to input the second image data into the target neural network to perform expansion processing on the second image data through the target neural network, and obtain an expansion processing result corresponding to the second image data.

[0047] As an optional implementation manner, in the second aspect of the present invention, the pre-set convolutional neural network includes a first network and a second network;

[0048] The training module includes:

[0049] A first training sub-module, configured to input the first image data into the first network to perform a first preset processing operation on the first image data through the first network, and obtain a first output result corresponding to the first image data;

[0050] A second training sub-module, configured to input the first output result into the second network to perform a second preset processing operation on the first output result through the second network, and obtain a second output result corresponding to the first output result;

[0051] A determination sub-module, configured to determine the second output result and the convolutional neural network that has completed training as the training processing result when it is determined that the second output result meets the set network training requirements;

[0052] Wherein, the first preset processing operation includes at least one of image convolution, image upsampling, image downsampling, pooling, linear rectification, and feature transformation;

[0053] The second preset processing operation includes at least one of feature extraction, region selection and generation, window classification, position refinement, NMS processing, detection box detection, and detection box recognition.

[0054] As an alternative implementation, in the second aspect of the present invention, the first network includes a contracting path and an expanding path; both the contracting path and the expanding path are constituted by repeatedly applying a preset number of different convolutional layers a preset number of times; and each convolutional layer is followed by a rectified linear unit;

[0055] The contracting path is used to perform downsampling processing on the image input to the contracting path; the expanding path is used to perform upsampling processing on the image input to the expanding path;

[0056] The specific manner in which the first training sub-module inputs the first image data into the first network to perform a first preset processing operation on the training set through the first network to obtain a first output result corresponding to the first image data includes:

[0057] Input the first image data into the contracting path and the expanding path respectively to obtain a contracting processing result corresponding to the contracting path and an expanding processing result corresponding to the expanding path;

[0058] Perform feature mapping and feature transformation on the contracting processing result and the expanding processing result to obtain a first output result corresponding to the first image data.

[0059] As an alternative implementation, in the second aspect of the present invention, the second network includes a region proposal network and a detection network;

[0060] The specific manner in which the second training sub-module inputs the first output result into the second network to perform a second preset processing operation on the first output result through the second network to obtain a second output result corresponding to the first output result includes:

[0061] Input the first output result into the region proposal network, perform feature extraction operations on the first output result through a preset image processing function, generate at least one candidate region according to the obtained feature extraction result, and then perform a preset adjustment operation on each candidate region to obtain a preset adjustment result corresponding to the candidate region. The preset adjustment operation includes window classification and position refinement;

[0062] According to the preset NMS algorithm, perform merging and deletion on all the preset adjustment results to obtain a plurality of proposal boxes corresponding to all the preset adjustment results;

[0063] According to the detection network, perform detection and recognition operations on the plurality of proposal boxes to obtain a detection and recognition result corresponding to the plurality of proposal boxes as the second output result.

[0064] As an alternative implementation, in the second aspect of the present invention, the calculation formula corresponding to the image processing function is as follows:

[0065]

[0066] where p i is the probability that the candidate region is predicted as the target;

[0067] The definition label corresponding to the preset reference standard is:

[0068]

[0069] The vector t i ={t x , t y , t w , t h} represents the parameter coordinates of the pre-determined bounding box; is the coordinate vector of the reference standard bounding box corresponding to the positive label candidate region; the output of the cls layer is {p i}, and the output of the reg layer is {u i}; the is the logarithmic loss of the two classes p i , .

[0070] As an alternative implementation, in the second aspect of the present invention, the first image data includes multiple sub-images, and there is a corresponding data annotation image for each sub-image;

[0071] The device further includes:

[0072] A data preparation module, configured to perform a partitioning process on all the sub-images before the training module inputs the first image data into a preset convolutional neural network, to obtain a training set and a validation set; the training data in the training set is used to train the preset convolutional neural network; the validation data in the validation set is used to verify the training effect of the convolutional neural network;

[0073] The second image data is used as an actual test set to perform a test on the target neural network.

[0074] As an alternative implementation, in the second aspect of the present invention, the broadening processing result includes broadening spectral data corresponding to the second image data; the second image data includes the longitudinal resolution before the second image data is input into the target neural network, denoted as the first resolution; the broadening processing result further includes the longitudinal resolution after the second image data is input into the target neural network, denoted as the second resolution;

[0075] The device further includes:

[0076] A comparison module, configured to compare the narrowband spectral data with the broadened spectral data to obtain spectral broadening information corresponding to the narrowband spectral data and the broadened spectral data;

[0077] The comparison module is further configured to compare the first resolution with the second resolution to obtain resolution optimization information corresponding to the first resolution and the second resolution;

[0078] A judgment module, configured to judge whether the target neural network meets a preset resolution improvement requirement according to the spectral broadening information and the resolution optimization information; when the judgment result is yes, determine that the target neural network is a qualified neural network that meets the resolution improvement requirement.

[0079] A third aspect of the present invention discloses another resolution optimization device for optical coherence tomography, and the device includes:

[0080] A memory storing executable program code;

[0081] A processor coupled to the memory;

[0082] The processor calls the executable program code stored in the memory to execute the resolution optimization method for optical coherence tomography disclosed in the first aspect of the present invention.

[0083] A fourth aspect of the present invention discloses a computer storage medium, and the computer storage medium stores computer instructions, which are used to execute the resolution optimization method for optical coherence tomography disclosed in the first aspect of the present invention when called.

[0084] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0085] In an embodiment of the present invention, a method for optimizing the resolution of optical coherence tomography is provided. The method includes: obtaining first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device; inputting the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network, obtaining a training processing result corresponding to the convolutional neural network, where the training processing result includes spectral feature data corresponding to the first image data and a target neural network that has completed training; obtaining second image data to be processed, where the second image data includes narrowband spectral data corresponding to the retina collected by the OCT device; inputting the second image data into the target neural network to perform broadening processing on the second image through the target neural network, obtaining a broadening processing result corresponding to the second image data. It can be seen that implementing the present invention can combine an artificial intelligence algorithm (convolutional neural network) to achieve intelligent analysis and processing of OCT spectral images, and then, through the trained target neural network, perform broadening on the OCT spectral image data that needs to be broadened, thereby achieving optimization and improvement of the longitudinal resolution of the spectral image, and the improvement of this longitudinal resolution is beneficial to improving the clarity of fundus images. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for description in the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0087] Figure 1 is a schematic flowchart of a method for optimizing the resolution of optical coherence tomography disclosed in an embodiment of the present invention;

[0088] Figure 2 is a schematic flowchart of another method for optimizing the resolution of optical coherence tomography disclosed in an embodiment of the present invention;

[0089] Figure 3 is a schematic structural diagram of a device for optimizing the resolution of optical coherence tomography disclosed in an embodiment of the present invention;

[0090] Figure 4 is a schematic structural diagram of another device for optimizing the resolution of optical coherence tomography disclosed in an embodiment of the present invention;

[0091] Figure 5 is a schematic structural diagram of yet another device for optimizing the resolution of optical coherence tomography disclosed in an embodiment of the present invention;

[0092] Figure 6It is a comparison schematic diagram of narrowband spectral data and broadband spectral data disclosed in an embodiment of the present invention;

[0093] Figure 7 It is a comparison schematic diagram of a first resolution and a second resolution disclosed in an embodiment of the present invention. Detailed implementation manners

[0094] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0095] The terms "first", "second", etc. in the specification and claims of the present invention and the above accompanying drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or terminal including a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or terminals.

[0096] Referring to "embodiment" herein means that a specific feature, structure or characteristic described in connection with the embodiment may be included in at least one embodiment of the present invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein may be combined with other embodiments.

[0097] The present invention discloses a method and device for optimizing the resolution of optical coherence tomography, which can combine an artificial intelligence algorithm (convolutional neural network) to realize the intelligent analysis and processing of OCT spectral images, and then perform broadening on the OCT spectral image data to be broadened through the trained target neural network, thereby realizing the optimization and improvement of the longitudinal resolution of the spectral image, and the improvement of the longitudinal resolution is beneficial to improving the clarity of fundus images. The following will be described in detail respectively.

[0098] Embodiment 1

[0099] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of a method for optimizing the resolution of optical coherence tomography disclosed in an embodiment of the present invention. Among them, Figure 1The described resolution optimization method for optical coherence tomography can be applied to the resolution optimization device of optical coherence tomography, which is not limited in the embodiments of the present invention. As Figure 1 shown, the resolution optimization method for optical coherence tomography may include the following operations:

[0100] 101. Obtain the first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device.

[0101] In the embodiments of the present invention, through an OCT imaging device, a light source with a wavelength of 840 nm and a bandwidth length greater than 40 nm is used to obtain the OCT image and spectrum of the retina, and the first image data is obtained.

[0102] In the embodiments of the present invention, the first image data includes multiple sub-images, and there is a corresponding data annotation image for each sub-image. Among them, the data annotation image of each sub-image can be an image pre-annotated by professionals.

[0103] In the embodiments of the present invention, optionally, before performing step 102 of inputting the first image data into a pre-set convolutional neural network, the method may further include:

[0104] Perform a partitioning process on all sub-images to obtain a training set and a validation set; the training data in the training set is used to train the pre-set convolutional neural network; the validation data in the validation set is used to verify the training effect of the convolutional neural network;

[0105] The second image data is used as an actual test set to perform a test on the target neural network.

[0106] It can be seen that in the embodiments of the present invention, before inputting the first image data into the convolutional neural network to be trained, the data is pre-processed and partitioned into training data and validation data, which is beneficial to reducing the probability of the convolutional neural network training error to a certain extent. At the same time, the set validation data can verify the training effect of the convolutional neural network and improve the training process of the convolutional neural network.

[0107] 102. Input the first image data into a pre-set convolutional neural network to perform a training process on the convolutional neural network, and obtain a training process result corresponding to the convolutional neural network. The training process result includes spectral feature data corresponding to the first image data and a target neural network that has completed training.

[0108] In the embodiments of the present invention, after step 102 inputs the first image data into the convolutional neural network, since the first image data includes many sub-images, the convolutional neural network will be repeatedly trained, and the result finally output after the convolutional neural network completes training is selected as the training process result.

[0109] 103. Obtain the second image data to be processed, where the second image data includes the narrowband spectral data corresponding to the retina collected by an OCT device.

[0110] In the embodiments of the present invention, the difference between the collected second image data and the first image data is that the first image data is broadband spectral data and the second image data is narrowband spectral data. A difference is set in terms of the width of the bandwidth, and other parameters remain unchanged.

[0111] 104. Input the second image data into the target neural network to perform broadening processing on the second image data through the target neural network, and obtain a broadening processing result corresponding to the second image data.

[0112] It can be seen that Figure 1 The described method for optimizing the resolution of optical coherence tomography can combine an artificial intelligence algorithm (convolutional neural network) to achieve intelligent analysis and processing of OCT spectral images. Then, through the trained target neural network, the OCT spectral image data that needs to be broadened is broadened, thereby achieving the optimization and improvement of the longitudinal resolution of the spectral image, and the improvement of this longitudinal resolution is beneficial to improving the clarity of fundus images.

[0113] In an optional embodiment, the pre-set convolutional neural network includes a first network and a second network;

[0114] The specific manner of the above step 102 of inputting the first image data into the pre-set convolutional neural network to perform training processing on the convolutional neural network and obtain the training processing result corresponding to the convolutional neural network includes:

[0115] Input the first image data into the first network to perform a first preset processing operation on the first image data through the first network, and obtain a first output result corresponding to the first image data;

[0116] Input the first output result into the second network to perform a second preset processing operation on the first output result through the second network, and obtain a second output result corresponding to the first output result;

[0117] When it is determined that the second output result meets the set network training requirements, determine the second output result and the convolutional neural network that has completed training as the training processing result;

[0118] Among them, the first preset processing operation includes at least one of image convolution, image upsampling, image downsampling, pooling, linear correction, and feature transformation;

[0119] The second preset processing operation includes at least one of feature extraction, region selection and generation, window classification, position refinement, NMS processing, detection box detection, and detection box recognition.

[0120] It can be seen that in this optional embodiment, in terms of the function of processing images, the convolutional neural network can be divided into a first and a second network part. Through the first network and its set first preset processing operation, and the second network and its set second preset processing operation, the corresponding image data is separately processed at the functional level. At the same time, the processing data of the second network depends on the first output result fed back by the first network, realizing the linkage between the first and second networks and improving the image processing accuracy and reliability of the first image data input into the convolutional neural network.

[0121] In this optional embodiment, the first network includes a contracting path and an expanding path; both the contracting path and the expanding path are composed of a preset number of different convolutional layers repeatedly applied a preset number of times; and each convolutional layer is followed by a rectified linear unit;

[0122] The contracting path is used to perform downsampling processing on the image input into the contracting path; the expanding path is used to perform upsampling processing on the image input into the expanding path;

[0123] The above method of inputting the first image data into the first network to perform the first preset processing operation on the training set through the first network to obtain the first output result corresponding to the first image data specifically includes:

[0124] Input the first image data into the contracting path and the expanding path respectively to obtain a contracting processing result corresponding to the contracting path and an expanding processing result corresponding to the expanding path;

[0125] Perform feature mapping and feature transformation on the contracting processing result and the expanding processing result to obtain the first output result corresponding to the first image data.

[0126] In this optional embodiment, in actual settings, both the contracting path and the expanding path can be composed of 2 / 3 different convolutional layers repeatedly applied 2 times; and the convolutional layer can be a 3*3 convolutional layer; each convolutional layer is correspondingly followed by a rectified linear unit (ReLU). After the actual first image data is convolved by the convolutional layer in the contracting path, a downsampling process will be performed. This downsampling process can be completed by a 2*2 max pooling layer with a stride of 2, doubling the number of feature channels through this downsampling step; after the first image data is convolved by the convolutional layer in the expanding path, an upsampling process will be performed, halving the number of feature channels through this upsampling step. After the first image data passes through the contracting path and the expanding path, finally, a convolutional layer with a kernel size of 3*3 is set to convert the feature map into a result with a specific depth.

[0127] It can be seen that in this alternative embodiment, through the set expansion path and contraction path, accurate feature extraction of the input first image data is achieved, and then feature transformation is performed through the finally set convolutional layer, improving the processing accuracy and reliability of the first output result output by the previous first network.

[0128] Embodiment 2

[0129] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of another method for optimizing the resolution of optical coherence tomography disclosed in the embodiments of the present invention. Among them, Figure 2 the method for optimizing the resolution of optical coherence tomography described can be applied to the device for optimizing the resolution of optical coherence tomography, which is not limited in the embodiments of the present invention. As Figure 2 shown, the method for optimizing the resolution of optical coherence tomography may include the following operations:

[0130] 201. Obtain the first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device.

[0131] 202. Input the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network, and obtain the training processing result corresponding to the convolutional neural network. The training processing result includes spectral feature data corresponding to the first image data and the target neural network that has completed training.

[0132] 203. Obtain the second image data to be processed, where the second image data includes narrowband spectral data corresponding to the retina collected by an OCT device.

[0133] 204. Input the second image data into the target neural network to perform broadening processing on the second image data through the target neural network, and obtain the broadening processing result corresponding to the second image data.

[0134] In the embodiments of the present invention, for other descriptions of steps 201 - 204, please refer to the other specific descriptions of steps 101 - 104 in Embodiment 1, which will not be elaborated in the embodiments of the present invention.

[0135] In the embodiments of the present invention, the broadening processing result includes the broadened spectral data corresponding to the second image data; the second image data includes the longitudinal resolution before the second image data is input into the target neural network, denoted as the first resolution; the broadening processing result further includes the longitudinal resolution after the second image data is input into the target neural network, denoted as the second resolution.

[0136] 205. Compare the narrowband spectral data with the broadened spectral data to obtain the spectral broadening information corresponding to the narrowband spectral data and the broadened spectral data.

[0137] In the embodiments of the present invention, please refer to Figure 6 , Figure 6 which is a comparison schematic diagram of narrowband spectral data and broadband spectral data disclosed in the embodiments of the present invention. As Figure 6 shown, Figure 6 the left side in Figure 6 is the schematic diagram corresponding to the narrowband spectral data, and

[0138] the right side is the schematic diagram corresponding to the broadened spectral data.

[0139] In the embodiments of the present invention, please refer to Figure 7 , Figure 7 which is a comparison schematic diagram of a first resolution and a second resolution disclosed in the embodiments of the present invention. As Figure 7 shown, Figure 7 the left side in Figure 7 is the schematic diagram corresponding to the first resolution, and

[0140] 207. According to the spectral broadening information and the resolution optimization information, determine whether the target neural network meets the preset resolution improvement requirement; when the judgment result is yes, determine that the target neural network is a qualified neural network that meets the resolution improvement requirement.

[0141] In the embodiments of the present invention, the method further includes:

[0142] When the judgment result is no, according to the spectral broadening information and the resolution optimization information, and in combination with the resolution improvement requirement, calculate the difference between the spectral broadening information and the resolution optimization information and the resolution improvement requirement as the index to be optimized;

[0143] Re - execute the network training of the target neural network according to the index to be optimized until the new spectral broadening information and the new resolution optimization information indicate that the resolution improvement requirement is met.

[0144] It can be seen that implementing Figure 2The described method for optimizing the resolution of optical coherence tomography can, after the target neural network that has completed training performs broadening processing on the second image data to obtain the broadening processing result, set a data comparison mechanism for relevant data (narrowband spectral data and broadened spectral data, first resolution and second resolution), so as to further quantitatively compare the longitudinal resolution improvement effect of the target neural network at the data level, which is beneficial to improving the convenience and intuitiveness of viewing the processing effect of the target neural network performing broadening processing on the second image data; for the situation where the target neural network does not meet the preset resolution improvement requirements, a corresponding network retraining process is also set, which is beneficial to improving the accuracy and reliability of the obtained target neural network to a certain extent.

[0145] In an optional embodiment, the second network includes a region proposal network and a detection network;

[0146] The above-mentioned manner of inputting the first output result into the second network to perform a second preset processing operation on the first output result through the second network to obtain a second output result corresponding to the first output result specifically includes:

[0147] Input the first output result into the region proposal network, perform feature extraction operations on the first output result through a preset image processing function, generate at least one candidate region based on the obtained feature extraction result, and then perform a preset adjustment operation on each candidate region to obtain a preset adjustment result corresponding to the candidate region. The preset adjustment operation includes window classification and position refinement;

[0148] According to the preset NMS algorithm, perform merging and deletion on all preset adjustment results to obtain multiple proposal boxes corresponding to all preset adjustment results;

[0149] According to the detection network, perform detection and recognition operations on the multiple proposal boxes to obtain a detection and recognition result corresponding to the multiple proposal boxes as the second output result.

[0150] In this optional embodiment, the overall network of the second network can adopt the Faster R-CNN algorithm. Further, the detection network in the second network can adopt the Fast R-CNN detection network.

[0151] In this optional embodiment, before inputting data into the second network, the second network needs to be trained, and data preparation for the training of Faster R-CNN needs to be carried out first. The input image of the Faster R-CNN network comes from the result map obtained by the above convolutional neural network.

[0152] Among them, a positive label is defined as: a candidate region whose intersection over union (IoU) with any ground truth bounding box is greater than x (IoU > x);

[0153] A negative label is defined as: a candidate region whose intersection over union with all ground truth bounding boxes is less than (1 - x).

[0154] Meanwhile, the column pixel coordinates of the broadband spectrogram corresponding to the first image data are used as the ground truth of the detection box of the Faster R-CNN network.

[0155] In this alternative embodiment, the region proposal network is used to extract the detection region, which can share the convolutional features of the entire image with the entire detection network, so that the region proposal takes almost no time. The region proposal network first performs feature extraction, then generates candidate regions, and finally performs window classification and position refinement. The feature extraction is performed using a common deep learning network on ImageNet, and the deep learning network ResNet101 is adopted. To train the region proposal network, it is necessary to assign class labels {object, non-object} to each candidate region. For positive and negative labels, object is equal to the positive label, and non-object is equal to the negative label; it is used to describe the multi-task loss in Faster R-CNN.

[0156] In this alternative embodiment, before inputting the first output result into the region proposal network, the method further includes:

[0157] Determining the column pixel coordinates in the first image data as the ground truth of the region proposal network;

[0158] According to the ground truth, a candidate region whose intersection over union with any ground truth bounding box is greater than a first set value is defined as a positive label; a candidate region whose intersection over union with any ground truth bounding box is less than a second set value is defined as a negative label; the sum of the first set value and the second set value is 1.

[0159] In this alternative embodiment, the calculation formula corresponding to the image processing function is as follows:

[0160]

[0161] Among them, p i is the probability that the candidate region is predicted as an object;

[0162] The defined label corresponding to the ground truth is:

[0163]

[0164] The vector t i = {t x, t y , t w , t h} represents the parametric coordinates of a pre-determined bounding box; is the coordinate vector of the reference standard bounding box corresponding to the positive label candidate region; the output corresponding to the cls layer is {p i}, and the output corresponding to the reg layer is {u i};

[0165] is the log loss between two classes p i , , and the corresponding calculation formula is:

[0166]

[0167] where, is the regression loss, and the corresponding calculation formula is:

[0168]

[0169] where there is a regression loss only when ; R is the smooth L1 loss function, and the calculation formula corresponding to R is:

[0170]

[0171] In the bounding box regression calculation of the second network:

[0172] t x =(x - x a ) / w a , t y =(y - y a ) / h a

[0173] t w = logW / w a , t h = logh / h a

[0174]

[0175]

[0176] where x, y, w, and h are the center coordinates, width, and height of the box respectively; x is the predicted box, x a is the anchor box, and x * is the ground truth.

[0177] In this alternative embodiment, since the feature extraction network extracts a large number of candidate boxes and there are many overlapping regions between the candidate boxes, the non-maximum suppression (NMS) method needs to be used to merge and delete the detection boxes.

[0178] Among them, the non-maximum suppression method is as follows: sort the scores of all candidate boxes, select the highest score and its corresponding box. Delete all candidate boxes whose overlap with the candidate box with the highest score is greater than 0.7 (IoU>0.7), and the remaining candidate box with the highest score is left. The non-maximum suppression method does not affect the final detection accuracy, but greatly reduces the number of proposed boxes. After the region proposal network (RPN) extracts the candidate regions, the Fast R-CNN module is used to achieve the final detection and recognition. The region proposal network and Fast R-CNN share the convolutional layer of ResNet101 through training.

[0179] It can be seen that in this alternative embodiment, through the set second network, operations such as region proposal, image detection and recognition can be accurately performed on the first output result output from the first network, improving the applicability and integrity of the overall convolutional neural network, and at the same time improving the accuracy and reliability of the second output result obtained; thus, it is beneficial to improve the reliability of the finally obtained target neural network.

[0180] Embodiment Three

[0181] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of a resolution optimization device for optical coherence tomography disclosed in an embodiment of the present invention. Among them, the resolution optimization device for optical coherence tomography can be a resolution optimization terminal, device, system or server for optical coherence tomography. The server can be a local server, a remote server, or a cloud server (also known as a cloud server). When the server is a non-cloud server, the non-cloud server can communicate with the cloud server. The embodiments of the present invention do not make any limitations. As Figure 3 shown, the resolution optimization device for optical coherence tomography can include an acquisition module 301, a training module 302, and a broadening processing module 303, where:

[0182] The acquisition module 301 is used to acquire first image data to be processed, and the first image data includes broadband spectral data corresponding to the retina collected by an OCT device.

[0183] A training module 302 is configured to input first image data into a pre-set convolutional neural network to perform a training process on the convolutional neural network, and obtain a training result corresponding to the convolutional neural network. The training result includes spectral feature data corresponding to the first image data and a target neural network that has completed training.

[0184] An acquisition module 301 is further configured to acquire second image data to be processed. The second image data includes narrowband spectral data corresponding to the retina collected by an OCT device.

[0185] A broadening processing module 303 is configured to input the second image data into the target neural network to perform a broadening process on the second image data through the target neural network, and obtain a broadening result corresponding to the second image data.

[0186] It can be seen that Figure 3 the resolution optimization device for optical coherence tomography described above can combine an artificial intelligence algorithm (convolutional neural network) to achieve intelligent analysis and processing of OCT spectral images. Subsequently, through the trained target neural network, the OCT spectral image data that needs to be broadened is broadened, thereby achieving the optimization and improvement of the longitudinal resolution of the spectral image, and the improvement of the longitudinal resolution is beneficial to improving the clarity of fundus images.

[0187] In an alternative embodiment, please refer to Figure 4 , Figure 4 which is a schematic structural diagram of a resolution optimization device for optical coherence tomography disclosed in an embodiment of the present invention. Optionally, the pre-set convolutional neural network includes a first network and a second network.

[0188] As Figure 4 shown, the training module 302 may include a first training sub-module 3021, a second training sub-module 3022, and a determination sub-module 3023, where:

[0189] The first training sub-module 3021 is configured to input the first image data into the first network to perform a first preset processing operation on the first image data through the first network, and obtain a first output result corresponding to the first image data.

[0190] The second training sub-module 3022 is configured to input the first output result into the second network to perform a second preset processing operation on the first output result through the second network, and obtain a second output result corresponding to the first output result.

[0191] The determination sub-module 3023 is configured to, when determining that the second output result meets the set network training requirements, determine the second output result and the convolutional neural network that has completed training as the training result.

[0192] Among them, the first preset processing operation includes at least one of image convolution, image upsampling, image downsampling, pooling, rectified linear unit, and feature transformation;

[0193] The second preset processing operation includes at least one of feature extraction, region selection and generation, window classification, position refinement, NMS processing, detection box detection, and detection box recognition.

[0194] It can be seen that in this optional embodiment, in terms of the function of processing images, this convolutional neural network can be divided into the first and second network parts. Through the first network and its set first preset processing operation, and the second network and its set second preset processing operation, the corresponding image data can be processed separately at the functional level. At the same time, the processing data of the second network depends on the first output result fed back by the first network, realizing the linkage between the first and second networks, and improving the image processing accuracy and reliability of the first image data input into this convolutional neural network.

[0195] In this optional embodiment, the first network includes a contracting path and an expanding path; both the contracting path and the expanding path are composed of a preset number of different convolutional layers repeatedly applied a preset number of times; and each convolutional layer is followed by a rectified linear unit;

[0196] The contracting path is used to perform downsampling processing on the image input into this contracting path; the expanding path is used to perform upsampling processing on the image input into this expanding path;

[0197] Further, the specific manner in which the above-mentioned first training sub-module 3021 inputs the first image data into the first network to perform the first preset processing operation on the training set through the first network to obtain the first output result corresponding to the first image data is as follows:

[0198] Input the first image data into the contracting path and the expanding path respectively to obtain the contracting processing result corresponding to the contracting path and the expanding processing result corresponding to the expanding path;

[0199] Perform feature mapping and feature transformation on the contracting processing result and the expanding processing result to obtain the first output result corresponding to the first image data.

[0200] It can be seen that in this optional embodiment, through the set expanding path and contracting path, precise feature extraction of the input first image data is realized, and then feature transformation is performed through the finally set convolutional layer, improving the processing accuracy and reliability of the first output result output by the pre-set first network.

[0201] In another optional embodiment, the second network includes a region proposal network and a detection network;

[0202] Further, the above-mentioned second training sub-module 3022 inputs the first output result into the second network to perform a second preset processing operation on the first output result through the second network, and the specific manner of obtaining the second output result corresponding to the first output result includes:

[0203] Input the first output result into the region proposal network, perform feature extraction operations on the first output result through a preset image processing function, generate at least one candidate region according to the obtained feature extraction result, and then perform a preset adjustment operation on each candidate region to obtain a preset adjustment result corresponding to the candidate region. The preset adjustment operation includes window classification and position refinement;

[0204] According to the preset NMS algorithm, perform merging and deletion on all preset adjustment results to obtain multiple proposal boxes corresponding to all preset adjustment results;

[0205] According to the detection network, perform detection and recognition operations on the multiple proposal boxes to obtain a detection and recognition result corresponding to the multiple proposal boxes as the second output result.

[0206] In this optional embodiment, the calculation formula corresponding to the above-mentioned image processing function is as follows:

[0207]

[0208] Among them, p i is the probability that the candidate region is predicted as the target;

[0209] The definition label corresponding to the reference standard is:

[0210]

[0211] The vector t i ={t x , t y , t w , t h} represents the parametric coordinates of the pre-determined bounding box; is the coordinate vector of the reference standard bounding box corresponding to the positive label candidate region; the output corresponding to the cls layer is {p i}, and the output corresponding to the reg layer is {u i}; is the logarithmic loss of two categories p i , .

[0212] It can be seen that in this alternative embodiment, through the second network provided, operations such as region proposal, image detection, and recognition can be accurately performed on the first output result output from the first network, improving the applicability and integrity of the overall convolutional neural network, and at the same time improving the accuracy and reliability of the second output result obtained; thus facilitating the improvement of the reliability of the finally obtained target neural network.

[0213] In yet another alternative embodiment, the first image data includes multiple sub-images, and there is a corresponding data annotation image for each sub-image;

[0214] As Figure 4 shown, the device further includes a data preparation module 304, where:

[0215] The data preparation module 304 is configured to perform a partitioning process on all sub-images before the training module 302 inputs the first image data into a pre-set convolutional neural network, to obtain a training set and a validation set; the training data in the training set is used to train the pre-set convolutional neural network; the validation data in the validation set is used to verify the training effect of the convolutional neural network;

[0216] The second image data is used as an actual test set to perform a test on the target neural network.

[0217] It can be seen that in this alternative embodiment, before inputting the first image data into the convolutional neural network to be trained, the data is pre-processed and partitioned into training data and validation data, which to a certain extent helps to reduce the probability of the convolutional neural network training error, and at the same time the set validation data can verify the training effect of the convolutional neural network, improving the training process of the convolutional neural network.

[0218] In another alternative embodiment, the broadening processing result includes broadening spectral data corresponding to the second image data; the second image data includes the longitudinal resolution before the second image data is input into the target neural network, denoted as the first resolution; the broadening processing result further includes the longitudinal resolution after the second image data is input into the target neural network, denoted as the second resolution;

[0219] As Figure 4 shown, the device further includes a comparison module 305 and a judgment module 306, where:

[0220] The comparison module 305 is configured to compare the narrowband spectral data with the broadening spectral data to obtain spectral broadening information corresponding to the narrowband spectral data and the broadening spectral data;

[0221] The comparison module 305 is further configured to compare the first resolution with the second resolution to obtain resolution optimization information corresponding to the first resolution and the second resolution;

[0222] A judgment module 306, configured to determine whether a target neural network meets a preset resolution improvement requirement according to the spectral broadening information and the resolution optimization information; when the judgment result is yes, determine that the target neural network is a qualified neural network that meets the resolution improvement requirement.

[0223] It can be seen that in this optional embodiment, after the target neural network obtained through training performs broadening processing on the second image data to obtain a broadening processing result, a data comparison mechanism for relevant data (narrowband spectral data and broadened spectral data, first resolution and second resolution) can be set, so as to further quantitatively compare the longitudinal resolution improvement effect of the target neural network at the data level, which is beneficial to improving the convenience and intuitiveness of viewing the processing effect of the target neural network performing broadening processing on the second image data; for the case where the target neural network does not meet the preset resolution improvement requirement, a corresponding network retraining process is also set, which is beneficial to improving the accuracy and reliability of the obtained target neural network to a certain extent.

[0224] Embodiment 4

[0225] Please refer to Figure 5 , Figure 5 which is a structural schematic diagram of another resolution optimization device for optical coherence tomography disclosed in the embodiments of the present invention. As Figure 5 shown, the resolution optimization device for optical coherence tomography may include:

[0226] A memory 401 storing executable program code;

[0227] A processor 402 coupled to the memory 401;

[0228] The processor 402 calls the executable program code stored in the memory 401 and executes the steps in the resolution optimization method for optical coherence tomography described in Embodiment 1 or Embodiment 2 of the present invention.

[0229] Embodiment 5

[0230] The embodiments of the present invention disclose a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute the steps in the resolution optimization method for optical coherence tomography described in Embodiment 1 or Embodiment 2 of the present invention.

[0231] Embodiment 6

[0232] The embodiments of the present invention disclose a computer program product, which includes a non-transitory computer storage medium storing a computer program, and the computer program is operable to cause a computer to execute the steps in the resolution optimization method for optical coherence tomography described in Embodiment 1 or Embodiment 2.

[0233] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0234] Through the specific descriptions of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer storage medium. The storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disk memories, tape memories, or any other computer-readable medium that can be used to carry or store data.

[0235] Finally, it should be noted that: what is disclosed in an optical coherence tomography resolution optimization method and device disclosed in an embodiment of the present invention is only a preferred embodiment of the present invention, only for explaining the technical solution of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the resolution of optical coherence tomography, characterized in that, The method includes: Obtaining first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device; Inputting the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network, obtaining a training processing result corresponding to the convolutional neural network, where the training processing result includes spectral feature data corresponding to the first image data and a target neural network that has completed training; Obtaining second image data to be processed, where the second image data includes narrowband spectral data corresponding to the retina collected by an OCT device; Inputting the second image data into the target neural network to perform broadening processing on the second image data through the target neural network, obtaining a broadening processing result corresponding to the second image data; The pre-set convolutional neural network includes a first network and a second network; The step of inputting the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network and obtaining a training processing result corresponding to the convolutional neural network includes: Inputting the first image data into the first network to perform a first preset processing operation on the first image data through the first network, obtaining a first output result corresponding to the first image data; Inputting the first output result into the second network to perform a second preset processing operation on the first output result through the second network, obtaining a second output result corresponding to the first output result; When it is determined that the second output result meets the set network training requirements, determining the second output result and the convolutional neural network that has completed training as the training processing result; Wherein, the first preset processing operation includes at least one of image convolution, image upsampling, image downsampling, pooling, rectified linear unit, and feature transformation; The second preset processing operation includes at least one of feature extraction, region selection and generation, window classification, position refinement, NMS processing, detection box detection, and detection box recognition; The first network includes a contracting path and an expanding path; both the contracting path and the expanding path are composed of a preset number of different convolutional layers repeatedly applied a preset number of times; and each convolutional layer is followed by a rectified linear unit; The contracting path is used to perform downsampling processing on the image input into the contracting path; the expanding path is used to perform upsampling processing on the image input into the expanding path; The step of inputting the first image data into the first network to perform a first preset processing operation on the first image data through the first network and obtaining a first output result corresponding to the first image data includes: Inputting the first image data into the contracting path and the expanding path respectively, obtaining a contracting processing result corresponding to the contracting path and an expanding processing result corresponding to the expanding path; Performing feature mapping and feature transformation on the contracting processing result and the expanding processing result to obtain a first output result corresponding to the first image data.

2. The resolution optimization method of optical coherence tomography according to claim 1, characterized in that The second network includes a region proposal network and a detection network; The step of inputting the first output result into the second network to perform a second preset processing operation on the first output result through the second network to obtain a second output result corresponding to the first output result includes: Inputting the first output result into the region proposal network, performing a feature extraction operation on the first output result through a preset image processing function, generating at least one candidate region according to the obtained feature extraction result, and then performing a preset adjustment operation on each candidate region to obtain a preset adjustment result corresponding to the candidate region, where the preset adjustment operation includes window classification and position refinement; Performing merging and deletion on all the preset adjustment results according to a preset NMS algorithm to obtain a plurality of proposal boxes corresponding to all the preset adjustment results; Performing detection and recognition operations on the plurality of proposal boxes according to the detection network to obtain a detection and recognition result corresponding to the plurality of proposal boxes as the second output result.

3. The resolution optimization method of optical coherence tomography according to claim 2, wherein The calculation formula corresponding to the image processing function is as follows: where p i is the probability that the candidate region is predicted as the target; The defined label corresponding to the preset reference standard is: vector t i ={t x , t y , t w , t h} represents the parametric coordinates of a pre-determined bounding box; is the coordinate vector of the reference standard bounding box corresponding to the positive label candidate region; the output corresponding to the cls layer is {p i}, and the output corresponding to the reg layer is {u i}; the is the log loss of two classes p i , .

4. The resolution optimization method of optical coherence tomography according to claim 1, characterized in that The first image data includes multiple sub-images, and there is a corresponding data annotation image for each sub-image; Before inputting the first image data into a preset convolutional neural network, the method further includes: Performing a partitioning process on all the sub-images to obtain a training set and a validation set; the training data in the training set is used to train the preset convolutional neural network; The validation data in the validation set is used to verify the training effect of the convolutional neural network; The second image data is used as an actual test set to perform a test on the target neural network.

5. The resolution optimization method of optical coherence tomography according to claim 1 or 3, characterized in that The widened processing result includes widened spectral data corresponding to the second image data; the second image data includes the longitudinal resolution before the second image data is input into the target neural network, denoted as the first resolution; The widened processing result further includes the longitudinal resolution after the second image data is input into the target neural network, denoted as the second resolution; The method further includes: Comparing the narrowband spectral data with the widened spectral data to obtain spectral widening information corresponding to the narrowband spectral data and the widened spectral data; Comparing the first resolution with the second resolution to obtain resolution optimization information corresponding to the first resolution and the second resolution; Judging whether the target neural network meets a preset resolution improvement requirement according to the spectral widening information and the resolution optimization information; when the judgment result is yes, determining that the target neural network is a qualified neural network that meets the resolution improvement requirement.

6. An apparatus for optimizing the resolution of optical coherence tomography, characterized in that, The device includes: An acquisition module, configured to acquire first image data to be processed, where the first image data includes broadband spectral data corresponding to the retina collected by an OCT device; A training module, configured to input the first image data into a pre-set convolutional neural network to perform training processing on the convolutional neural network, and obtain a training processing result corresponding to the convolutional neural network, where the training processing result includes spectral feature data corresponding to the first image data and a target neural network that has completed training; The obtaining module is further configured to obtain second image data to be processed, where the second image data includes narrowband spectral data corresponding to the retina collected by an OCT device; A broadening processing module, configured to input the second image data into the target neural network to perform broadening processing on the second image data through the target neural network, and obtain a broadening processing result corresponding to the second image data; The pre-set convolutional neural network includes a first network and a second network; The training module includes: A first training sub-module, configured to input the first image data into the first network to perform a first preset processing operation on the first image data through the first network, and obtain a first output result corresponding to the first image data; A second training sub-module, configured to input the first output result into the second network to perform a second preset processing operation on the first output result through the second network, and obtain a second output result corresponding to the first output result; A determining sub-module, configured to determine the second output result and the convolutional neural network that has completed training as the training processing result when it is determined that the second output result meets the set network training requirements; Wherein, the first preset processing operation includes at least one of image convolution, image upsampling, image downsampling, pooling, rectified linear unit, and feature transformation; The second preset processing operation includes at least one of feature extraction, region selection and generation, window classification, position refinement, NMS processing, detection box detection, and detection box recognition; The first network includes a contracting path and an expanding path; both the contracting path and the expanding path are composed of a preset number of different convolutional layers repeatedly applied a preset number of times; and each convolutional layer is followed by a rectified linear unit; The contracting path is configured to perform downsampling processing on the image input to the contracting path; the expanding path is configured to perform upsampling processing on the image input to the expanding path; The manner in which the first training sub-module inputs the first image data into the first network to perform a first preset processing operation on the first image data through the first network and obtain a first output result corresponding to the first image data specifically includes: Inputting the first image data into the contracting path and the expanding path respectively to obtain a contracting processing result corresponding to the contracting path and an expanding processing result corresponding to the expanding path; Performing feature mapping and feature transformation on the contracting processing result and the expanding processing result to obtain a first output result corresponding to the first image data.

7. An apparatus for optimizing the resolution of optical coherence tomography, characterized in that The device includes: A memory storing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory and executes the method for optimizing the resolution of optical coherence tomography according to any one of claims 1-5.

8. A computer storage medium, characterized in that, The computer storage medium stores computer instructions which, when called, are used to execute the method for optimizing the resolution of optical coherence tomography according to any one of claims 1-5.

Images