Method and system for acquiring three-dimensional morphology of sweat gland of human skin and electronic equipment

Abstract

The invention provides a human skin sweat gland three-dimensional shape obtaining method and system and electronic equipment, and relates to the technical field of optical coherence tomography and medical image segmentation, and the method comprises the steps: obtaining an initial to-be-detected region skin OCT image through a spectral domain optical coherence tomography system; preprocessing the skin OCT image of the initial to-be-detected area, wherein preprocessing comprises but is not limited to one or more of cutting, overturning, size adjustment and contrast adjustment; inputting the preprocessed skin OCT image of the initial to-be-detected area into a three-dimensional segmented skin sweat gland network model trained in advance based on deep learning, and performing three-dimensional segmentation on the preprocessed skin OCT image of the initial to-be-detected area by the deep learning model to obtain a three-dimensional segmentation result of sweat glands; according to the three-dimensional shape obtaining method, the three-dimensional shape information of the sweat gland of the human skin can be obtained in a real-time and non-invasive mode, the two-dimensional, in-vitro and destructive limitation of a traditional skin biopsy technology is overcome, and the accuracy and efficiency of sweat gland shape analysis are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical coherence tomography and the field of medical image segmentation, in particular, relates to a human skin optical coherence tomography (OCT) image processing method, system and electronic equipment, which is used to realize three-dimensional morphology acquisition of human skin sweat glands. BACKGROUND

[0002] Sweat glands, as an important skin accessory organ, are directly involved in body temperature regulation. Their morphological characteristics vary significantly among different individuals and under pathological conditions, reflecting adaptive functional changes under different physiological and pathological conditions. Studying the morphology of sweat glands is of great significance for a better understanding of their physiological functions and the diagnosis of related diseases. Traditional sweat gland research usually relies on skin biopsy techniques, but this method is two-dimensional, destructive and lagging, making it difficult to provide real-time, non-invasive and quantifiable sweat gland morphology information.

[0003] Optical coherence tomography (OCT) as a high-resolution, non-invasive imaging technology can perform tomographic imaging of biological tissues and has become an important tool in sweat gland research. However, existing neural network methods often ignore the three-dimensional spatial characteristics of OCT images when segmenting skin sweat glands, limiting their performance in segmenting sweat glands with spiral structures. SUMMARY

[0004] To solve the problems existing in the prior art, the present application provides a method for acquiring three-dimensional morphology of human skin sweat glands. This method uses a Fourier domain optical coherence tomography (FD-OCT) system to acquire high-resolution three-dimensional skin tissue images and a Transformer neural network model to extract a three-dimensional skin sweat gland segmentation method with high precision, reliability and robustness, which can automatically and accurately extract three-dimensional skin sweat glands.

[0005] To achieve the above-mentioned purposes, the technical solutions adopted by the present application are as follows:

[0006] In a first aspect, the present application provides a method for acquiring three-dimensional morphology of human skin sweat glands, comprising:

[0007] An initial skin OCT image of a region to be inspected is obtained from an optical coherence tomography system; the initial skin OCT image of the region to be inspected has a sweat gland region. In the present application, a Fourier domain optical coherence tomography (FD-OCT) system is used to collect skin images. The system can be based on spectral domain OCT (SD-OCT) or swept source OCT (SS-OCT) and obtain high-resolution three-dimensional structural information of skin tissue through optical interference principles. In the method of the present application, SD-OCT is used for data acquisition, and a wideband light source and spectral detection are used to obtain high-resolution three-dimensional skin tissue images, ensuring accurate presentation of the sweat gland region in the OCT image.

[0008] The initial skin OCT image of the region to be inspected is input into a pre-trained three-dimensional skin sweat gland segmentation deep learning model, and the initial skin OCT image of the region to be inspected is three-dimensionally segmented by the three-dimensional segmentation deep learning model to obtain a three-dimensional segmentation result of the corresponding skin sweat gland region.

[0009] Specifically, the three-dimensional skin sweat gland segmentation deep learning model is a U-shaped structure composed of a five-layer SwinTransformer encoder, an Efficient Channel Attention (ECA) attention module at the end of each encoder, and a full convolutional decoder connected in sequence. The three-dimensional skin sweat gland segmentation network model includes an encoding module, an ECA attention module, a decoding module, and a saliency layer; the encoding module includes an Embedding layer and a plurality of encoding layers connected in sequence, the decoding module includes a plurality of decoding layers connected in sequence, the number of encoding layers and the number of decoding layers are the same, the ECA attention module is connected after the Embedding layer and each encoding layer, the output of the ECA attention module is skip-connection to the corresponding decoding layer in the decoding module; the saliency layer includes a 3*3*3 convolution layer and a 1*1*1 convolution layer, the image to be segmented is input into the encoding module and the 3*3*3 convolution layer in the saliency layer, respectively, the output of the 3*3*3 convolution layer is input into the 1*1*1 convolution layer after ConCate operation with the output of the decoding module, to obtain the segmented image.

[0010] As a preferred embodiment, the encoding module comprises an Embedding layer and four encoding layers, namely a first encoding layer, a second encoding layer, a third encoding layer and a fourth encoding layer, each encoding layer has the same structure and comprises two Swin Transformer modules and one Patch Merging layer connected in sequence; the ECA attention module connected after the Embedding layer is a first ECA attention module; the output of the Patch Merging layer of each encoding layer is connected to the ECA attention module of the corresponding layer, which are a second ECA attention module, a third ECA attention module, a fourth ECA attention module and a fifth ECA attention module, respectively.

[0011] As a preferred embodiment, the Embedding layer and the ECA attention module connected after each encoding layer are a first ECA attention module, a second ECA attention module, a third ECA attention module, a fourth ECA attention module and a fifth ECA attention module, respectively; the decoding module comprises five decoding layers connected in sequence, from bottom to top, which are a first decoding layer, a second decoding layer, a third decoding layer, a fourth decoding layer and a fifth decoding layer; the input of the first decoding layer is the output of the fifth ECA attention module corresponding to the encoding layer, which is input into the second decoding layer after up-sampling operation together with the output of the fourth ECA attention module, and the second decoding layer comprises a DFF module and an up-sampling operation; the third decoding layer, the fourth decoding layer and the fifth decoding layer have the same structure and each comprises a ConCate operation and an up-sampling operation, the ConCate operation is used to concatenate the outputs of the previous decoding layer and the ECA attention module of the corresponding layer in the channel, and the up-sampling operation is used to up-sample the output after channel concatenation, and the output after up-sampling is the output of the decoding layer.

[0012] As a preferred embodiment, the first ECA attention module, the second ECA attention module and the third ECA attention module are further connected with a residual module; the third decoding layer, the fourth decoding layer and the fifth decoding layer are further provided with a residual module between the Concate operation and the up-sampling operation.

[0013] As a preferred embodiment, the fourth ECA attention module and the fifth ECA attention module are further connected with a DLK module; the input of the first decoding layer is the output of the DLK module after the fifth ECA attention module, which is input into the second decoding layer after up-sampling operation of the first decoding layer together with the output of the DLK module after the fourth ECA attention module; the second decoding layer is further provided with a DLK module between the DFF module and the up-sampling operation.

[0014] As a preferred embodiment, the hybrid loss function of the three-dimensional segmented skin sweat gland network model is L SweatGland , the expression of which is as follows:

[0015] L SweatGland = aL BCE + bL Dice

[0016] wherein a and b are weight coefficients, L BCE is a binary cross-entropy loss function, and L Dice is a Dice loss function, the expressions of the two loss functions being as follows:

[0017]

[0018]

[0019] wherein N is the number of samples, Pi is the predicted probability of the i-th sample, and yi is the true label of the i-th sample.

[0020] Specifically, the DLK module comprises, in sequence, a first LayerNorm layer, a DLK unit, a second LayerNorm layer and an MLP unit, the input of the first LayerNorm layer being connected to the output of the DLK unit in a skip connection manner, and the output of the DLK unit and the output of the second LayerNorm layer being jointly connected to the output of the MLP unit in a skip connection manner; the DLK unit comprises, in sequence, a first 1*1*1 convolutional layer, a GELU layer, a DLK layer and a second 1*1*1 convolutional layer, the input of the first 1*1*1 convolutional layer being connected to the output of the second 1*1*1 convolutional layer in a skip connection manner.

[0021] Optionally, the steps of the present application further comprise: pre-processing the initial skin OCT image of the region to be inspected, the pre-processing including but not limited to one or more of cropping, flipping, adjusting the size and adjusting the contrast.

[0022] Specifically, before training, the three-dimensional segmented skin sweat gland network model further comprises:

[0023] An OCT image dataset of skin sweat glands is constructed, the sample dataset comprising a plurality of image pairs, each image pair comprising an original skin OCT image and a true value image in which a sweat gland region is artificially labeled. The human skin OCT image dataset is pre-processed; the pre-processing includes but is not limited to cropping, flipping, adjusting the size, adjusting the contrast and the like, and finally the pre-processed dataset is divided into a training set, a test set and a validation set in a ratio of 8:1:1.

[0024] In a second aspect, a skin sweat gland three-dimensional segmentation system based on deep learning comprises:

[0025] An image acquisition module is configured to acquire an initial skin OCT image of a region to be inspected output by an optical coherence tomography system;

[0026] A preprocessing module is configured to preprocess the initial skin OCT image of the region to be inspected, and the preprocessing includes one or more of cropping, flipping, adjusting size, and contrast adjustment;

[0027] A segmentation module is configured to input the preprocessed initial skin OCT image of the region to be inspected into a three-dimensional skin sweat gland segmentation network model based on deep learning that is trained in advance, and perform three-dimensional segmentation on the preprocessed initial skin OCT image of the region to be inspected by the deep learning model to obtain a three-dimensional segmentation result of the sweat gland.

[0028] The three-dimensional skin sweat gland segmentation network model includes an encoding module, an ECA attention module, a decoding module, and a saliency layer; the encoding module includes an Embedding layer and a plurality of encoding layers connected in sequence, the decoding module includes a plurality of decoding layers connected in sequence, the number of layers of the encoding module is the same as the number of layers of the decoding module, the ECA attention module is connected after the Embedding layer and each encoding layer, and the output of the ECA attention module is connected to the corresponding decoding layer in the decoding module in a skip layer manner; the saliency layer includes a 3*3*3 convolution layer and a 1*1*1 convolution layer, the image to be segmented is input into the encoding module and the 3*3*3 convolution layer in the saliency layer respectively, the output of the 3*3*3 convolution layer is input into the 1*1*1 convolution layer after a ConCate operation with the output of the decoding module, and a segmented image is obtained.

[0029] In a third aspect, the embodiments of the present application further provide an electronic device, including a processor, a storage medium, and a bus, the storage medium stores program instructions executable by the processor, when the electronic device is running, the processor and the storage are in communication through the bus, and the machine readable instructions are executed by the processor to perform the three-dimensional skin sweat gland segmentation method based on deep learning as described above.

[0030] The present application has the following beneficial effects:

[0031] The present application provides a human skin sweat gland three-dimensional morphology acquisition method and an electronic device storage medium, which solves the blank of the prior art, lays a foundation for further deepening the understanding of sweat gland function, studying the behavior of sweat glands under various physiological and pathological conditions, promoting dermatological research and clinical application, and has good development prospects in the medical field. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those of ordinary skill in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0033] Figure 1 The flowchart of the method for obtaining the three-dimensional appearance of human skin sweat glands according to the present application;

[0034] Figure 2 The application scenario structure diagram of the method for obtaining the three-dimensional appearance of human skin sweat glands according to the present application;

[0035] Figure 3 The deep learning network structure diagram of the method for obtaining the three-dimensional appearance of human skin sweat glands according to the present application.

[0036] Figure 4 The flowchart of the method for constructing the OCT image data set of human skin sweat glands provided by the embodiments of the present application.

[0037] Figure 5 The structure diagram of the spectral domain optical coherence tomography system provided by the embodiments of the present application.

[0038] Figure 6 The comparison diagram of the validation set segmentation sample of the method for obtaining the three-dimensional appearance of human skin sweat glands according to the present application.

[0039] Figure 7 The single sweat gland recognition result diagram of the multiple preferred embodiments of the method for obtaining the three-dimensional appearance of human skin sweat glands in clinical practice according to the present application.

[0040] Figure 8 The system structure diagram of the method for obtaining the three-dimensional appearance of human skin sweat glands provided by the embodiments of the present application.

[0041] Figure 9 The structure block diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of illustration and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowcharts can be implemented out of sequence, and steps without logical context can be reversed or implemented simultaneously. In addition, those skilled in the art, under the guidance of the contents of this application, can add one or more other operations to the flowchart, or remove one or more operations from the flowchart.

[0043] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the accompanying drawings can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work fall within the scope of protection of the present application.

[0044] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0045] Figure 2 This is a schematic diagram of the application scenario of a method for acquiring the three-dimensional morphology of sweat glands on human skin according to the present invention. Figure 2 As shown, the present application is applicable to the scenario of obtaining an OCT image of any sample. The sample can be, for example, a biological sample. This scenario involves an optical coherence tomography (OCT) system and an electronic device. First, any sample is scanned by the OCT system to output an initial skin OCT image of the area to be tested, and then the initial skin OCT image of the area to be tested is input into the above-mentioned electronic device, which uses the trained segmentation deep learning network model to perform three-dimensional segmentation on the skin OCT image, and finally obtains a three-dimensional sweat gland segmentation result.

[0046] Figure 2 FIG. 1 is a flow chart of a method for acquiring the three-dimensional morphology of sweat glands on human skin according to the present invention. Figure 2As shown, the method is applied to the above-mentioned electronic device, which can be a desktop computer, a notebook computer, or other terminal device with computing processing capability and display function, or can also be a server. The method comprises:

[0047] S101, obtaining an initial skin OCT image of a to-be-inspected region output by a spectral domain optical coherence tomography system.

[0048] Optionally, the SD-OCT system can include a light source, a coupler, a reference mirror, a sample arm, a reference arm, and a signal detector. Specifically, the system can use a low-coherence light source as the light source, for example, a broadband light source with a center wavelength of 850 nm and a full-width-at-half-maximum of 165 nm. The light emitted by the low-coherence light source passes through the coupler and is irradiated to the sample and the reference mirror through the sample arm and the reference arm, respectively. The reflected light in the two light paths converges in the coupler under certain conditions to generate an interference signal, which is output to the signal detector. The signal detector transmits the detected interference signal to the computer, which performs gray-scale imaging and outputs an initial skin OCT image of a to-be-inspected region. The initial skin OCT image of the to-be-inspected region contains sweat glands.

[0049] S102, preprocessing the initial skin OCT image of the to-be-inspected region, which includes but is not limited to one or more of cropping, flipping, adjusting size, and contrast adjustment. The preprocessed initial skin OCT image of the to-be-inspected region is input into a pre-trained deep learning network model for three-dimensional segmentation of sweat glands, which performs three-dimensional segmentation on the initial skin OCT image of the to-be-inspected region to obtain a three-dimensional sweat gland image.

[0050] The deep learning network model for three-dimensional segmentation of sweat glands is trained according to the method provided in the present application. The trained deep learning network model for three-dimensional segmentation of skin sweat glands can be combined with the OCT system in step S01, and the initial skin OCT image of the to-be-inspected region obtained in step S01 can be input into the trained deep learning network model for three-dimensional segmentation of sweat glands, which performs three-dimensional segmentation on the initial skin OCT image of the to-be-inspected region to obtain a three-dimensional sweat gland segmentation result.

[0051] The present embodiment obtains an initial skin optical coherence tomography image output by a spectral domain optical coherence tomography system; inputs the initial skin optical coherence tomography image into a pre-trained deep learning network model for three-dimensional segmentation of sweat glands, which performs three-dimensional segmentation on the initial skin optical coherence tomography image to obtain a three-dimensional sweat gland segmentation result.

[0052] The deep learning network model for three-dimensional segmentation of sweat glands is improved based on a Transformer neural network model, and sequentially connected are a five-layer Swin Transformer-based encoder, an ECA (Efficient Channel Attention) attention module at the end of each layer of the encoder, and a full convolutional decoder.

[0053] The specific structure of the model is shown in Figure 3 The model includes an encoding module, an ECA attention module, a decoding module, and a saliency layer, and the main structure of the entire model is U-shaped; the encoding module includes an Embedding layer and a plurality of encoding layers connected in sequence, the decoding module includes a plurality of decoding layers connected in sequence, the number of layers of the encoding module and the number of layers of the decoding module are the same, and the ECA attention module is connected after the Embedding layer and each encoding layer, and the output of the ECA attention module is connected to the corresponding decoding layer in the decoding module in a skip layer manner; the saliency layer includes a 3*3*3 convolution layer and a 1*1*1 convolution layer, the image to be segmented is input into the encoding module and the 3*3*3 convolution layer in the saliency layer, respectively, the output of the 3*3*3 convolution layer is input into the 1*1*1 convolution layer after ConCate operation with the output of the decoding module, and the segmented image is obtained.

[0054] The encoding module includes an Embedding layer and four encoding layers, namely a first encoding layer, a second encoding layer, a third encoding layer, and a fourth encoding layer, each encoding layer has the same structure and includes two Swin Transformer modules and one Patch Merging layer connected in sequence; the ECA attention module connected after the Embedding layer is a first ECA attention module; the output of the Patch Merging layer of each encoding layer is connected to the ECA attention module of the corresponding layer, which are a second ECA attention module, a third ECA attention module, a fourth ECA attention module, and a fifth ECA attention module.

[0055] The decoding module includes five decoding layers connected in sequence, from bottom to top, a first decoding layer, a second decoding layer, a third decoding layer, a fourth decoding layer, and a fifth decoding layer; the input of the first decoding layer is the output of the fifth ECA attention module corresponding to the encoding layer, which is input into the second decoding layer after up-sampling operation and together with the output of the fourth ECA attention module; the second decoding layer includes a DFF (Dynamic Feature Fusion) module and an up-sampling operation; the third decoding layer, the fourth decoding layer, and the fifth decoding layer have the same structure, each including a ConCate operation and an up-sampling operation, the ConCate operation being used for channel splicing the output of the previous decoding layer and the output of the ECA attention module of the corresponding level, and the up-sampling operation being used for up-sampling the output after channel splicing, and the output after up-sampling being the output of the decoding layer.

[0056] The first ECA attention module, the second ECA attention module, and the third ECA attention module are further connected with a residual module; the third decoding layer, the fourth decoding layer, and the fifth decoding layer are further provided with a residual module between the Concate operation and the up-sampling operation. The fourth ECA attention module and the fifth ECA attention are further connected with a DLK (Dynamic Large Kernel) module; the input of the first decoding layer is the output of the DLK module after the fifth ECA attention module, which is input into the second decoding layer together with the output of the DLK module after the fourth ECA attention module after up-sampling operation of the first decoding layer; the DFF module and the up-sampling operation of the second decoding layer are further provided with a DLK module.

[0057] The DLK module includes a first LayerNorm layer, a DLK unit, a second LayerNorm layer, and an MLP unit connected in sequence, the input of the first LayerNorm layer is connected to the output of the DLK unit in a skip layer manner, and the output of the DLK unit and the output of the MLP unit are connected together in a skip layer manner; the DLK unit includes a first 1*1*1 convolution layer, a GELU layer, a DLK layer, and a second 1*1*1 convolution layer connected in sequence, and the input of the first 1*1*1 convolution layer is connected to the output of the second 1*1*1 convolution layer in a skip layer manner.

[0058] Specifically, the ECA attention module, the DFF module, and the DLK module are all specific structures of the prior art, Figure 3 which have been described in detail and will not be described here.

[0059] In the above model, the loss information determined according to the output segmentation result and the corresponding true value image includes:

[0060] The mixed loss function of the three-dimensional segmentation skin sweat gland network model is LSweatGland , the expression of which is as follows:

[0061] L SweatGland = aL BCE + bL Dice

[0062] wherein a and b are weight coefficients, which are respectively set to 0.9 and 0.1 in the present application, L BCE is a binary cross entropy loss function, and L Dice is a Dice loss function, the expressions of the two loss functions being as follows:

[0063]

[0064] wherein N is the number of samples, p i is the predicted probability of the i-th sample, and y i is the true label (0 or 1) of the i-th sample. In the present embodiment, the number of samples is specifically the number of voxel points. The binary cross entropy loss function can control the balance between false positives, true negatives, true positives and false negatives, improve the segmentation accuracy, and reduce outliers; and the Dice loss function can refine the segmentation boundary, thereby further improving the segmentation effect.

[0065] The initial skin OCT image containing sweat glands in the region to be examined is input into the encoding input coding module of the three-dimensional segmentation skin sweat gland deep learning network model, and is processed through the Embedding layer and four encoding layers based on Swin Transformer. The output features of each layer are processed through a series of processes, and finally the comprehensive high-level semantic features are obtained. Then, the feature map is processed through the ECA (Efficient Channel Attention) attention module, forms a skip connection, and is spliced with the feature map after upsampling in the decoding module. This operation is repeated four times until the output size is restored to the original input size, and the output features are spliced with the original input image through the saliency layer, fully integrating low-level spatial information and high-level semantic information, and finally obtaining the segmentation prediction result. In particular, in the fourth and fifth layers of the semantic feature abstract encoder (the first and second layers of the decoder), the DFF (Dynamic Feature Fusion) and DLK (Dynamic Large Kernel) modules are used to replace the Concate operation and the residual module, respectively, so as to expand the receptive field of the deep network and enhance the perception ability of the model to abstract features.

[0066] Figure 4 A flowchart of a method for constructing an OCT image dataset of human skin is provided in the embodiments of the present application. The method specifically includes:

[0067] S201, using a modified optical coherence tomography system to scan the human skin to obtain a plurality of initial images corresponding to the sample object.

[0068] S202, respectively, the initial skin OCT image of each initial detection area as the initial skin OCT image of the image pair (referred to as the initial image), according to the selected sweat gland target segmentation area, manually label the sweat gland target segmentation area on each initial image as the ground truth image in the image pair, and combine each initial image with the corresponding ground truth image to form an image pair.

[0069] Optionally, for each sample, the corresponding multiple initial images and the ground truth image corresponding to the initial object are combined to obtain an image pair. By this method, other initial images and their corresponding ground truth images are combined to obtain multiple image pairs.

[0070] Illustratively, an initial image corresponding to a skin sample has 100, then the sample can have 100 image pairs.

[0071] S203, integrating the image pairs corresponding to each skin sample to obtain a skin sweat gland optical coherence tomography image dataset.

[0072] Among them, the dataset combined by all image pairs corresponding to all sample objects is the skin sweat gland optical coherence tomography image dataset, and the preset proportion of the skin sweat gland optical coherence tomography image dataset can be respectively used as the training dataset, the validation dataset and the test dataset. For example, 80% of the skin sweat gland optical coherence tomography image dataset can be used as the training dataset, 10% of the skin sweat gland optical coherence tomography image dataset can be used as the validation dataset, and 10% of the skin sweat gland optical coherence tomography image dataset can be used as the test dataset.

[0073] Illustratively, if there are 100 sample objects, each sample object can obtain 100 image pairs corresponding to each sample object through the above S201-S203, and the skin sweat gland optical coherence tomography image dataset includes 100 image pairs.

[0074] Figure 5 The above SD-OCT system (spectral domain optical coherence tomography system) structure diagram is provided in the embodiments of the present application.

[0075] As Figure 5As shown, the SD-OCT system comprises a low-coherence light source, a signal detector, a fiber coupler, a collimating mirror, dispersion-matched glass, a lens, a collimating mirror, a scanning lens, a two-dimensional scanning galvanometer, a first polarization controller, a second polarization controller, and a mirror.

[0076] Optionally, the skin of the region to be measured can be scanned by the SD-OCT system to obtain a plurality of initial skin OCT images of the region to be measured corresponding to each sample.

[0077] Figure 6 A three-dimensional segmentation algorithm based on human skin optical coherence tomography images according to an embodiment of the present application is compared with a test set segmentation sample. Figure 4 From left to right are the original skin OCT image (left), the 3D sweat gland image manually labeled by a dermatologist (middle), and the predicted 3D sweat gland image (right). It can be seen intuitively that these three-dimensional prediction results not only segment the skin sweat glands with smooth and highly consistent surface boundaries, but also have an overall distribution close to the actual morphology, which is conducive to three-dimensional morphological observation of sweat glands.

[0078] Figure 7 A result schematic diagram of a plurality of preferred embodiments of a three-dimensional segmentation algorithm based on human skin optical coherence tomography images according to an embodiment of the present application in clinical practice. Figure 7 The twelve three-dimensional skin sweat gland segmentation images show the three-dimensional segmentation results of different skin sweat glands. It can be seen that the segmentation results clearly show the spiral structure of the sweat glands, can preserve surface details, and the segmented surface boundaries exhibit high consistency. This side proves that the three-dimensional segmentation method based on human skin sweat gland optical coherence tomography imaging images provided by the embodiment of the present application can effectively represent global and local features to meet the diversified needs of medical diagnosis with high precision.

[0079] Figure 8 A system structure schematic diagram of initial model training provided by an embodiment of the present application.

[0080] Figure 9 A structure block diagram of an electronic device 300 provided by an embodiment of the present application. As shown, the electronic device can include a processor 301 and a memory 302. Figure 9

[0081] Optionally, a bus 303 can also be included, wherein the memory 302 is configured to store machine-readable instructions executable by the processor 301, and when the electronic device 300 is running, the processor 301 communicates with the memory 302 through the bus 303, and the machine-readable instructions are executed by the processor 601 to perform the method steps in the above method embodiments.

[0082] ​The embodiment of the application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program.

[0083] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and the device described above can refer to the corresponding process in the method embodiment, and the application will not be repeated herein. In the several embodiments provided in the application, it should be understood that the disclosed system, device and method can be implemented by other ways. The device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and the actual implementation can be another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed mutual elements can be indirect coupling or communication connection through some communication interface, device or module, and can be electrical, mechanical or other forms.

[0084] In addition, each functional unit in each embodiment of the application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. When the functions are realized in the form of software functional units and sold or used as independent products, the software functional units can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the application essentially or the parts of the technical solutions that make contributions to the prior art or the parts of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0085] The above is only a specific embodiment of the application, but the protection scope of the application is not limited to this, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the application, which should be covered in the protection scope of the application.

Claims

1. A method for acquiring the three-dimensional morphology of sweat glands on human skin, characterized in that: include: Use spectral domain optical coherence tomography system to obtain the initial OCT image of the skin in the area to be examined; The skin OCT image of the initial area to be inspected is input into a pre-trained deep learning-based three-dimensional segmentation skin sweat gland network model. This model performs three-dimensional segmentation on the skin OCT image of the initial area to be inspected to obtain the three-dimensional segmentation result of the sweat glands.

2. The method for acquiring the three-dimensional morphology of sweat glands of human skin according to claim 1, characterized in that: The three-dimensional segmentation skin sweat gland network model includes an encoding module, an ECA attention module, a decoding module and a saliency layer; the encoding module includes an embedding layer and multiple encoding layers connected in sequence, the decoding module includes multiple decoding layers connected in sequence, the number of layers of the encoding module is the same as the number of layers of the decoding module, the ECA attention module is connected after the embedding layer and each encoding layer, and the output of the ECA attention module is jump-connected to the corresponding decoding layer in the decoding module; the saliency layer includes a 3*3*3 convolutional layer and a 1*1*1 convolutional layer, the image to be segmented is input into the 3*3*3 convolutional layer in the encoding module and the saliency layer respectively, the output of the 3*3*3 convolutional layer is ConCate-operated with the output of the decoding module and then input into the 1*1*1 convolutional layer to obtain the segmented image.

3. The method for acquiring the three-dimensional morphology of sweat glands of human skin according to claim 2, characterized in that: The encoding module includes an embedding layer and four encoding layers, namely the first encoding layer, the second encoding layer, the third encoding layer and the fourth encoding layer. Each encoding layer has the same structure, including two Swin Transformer modules and one Patch Merging layer connected in sequence; the ECA attention module connected after the embedding layer is the first ECA attention module; The output of the Patch Merging layer of each encoding layer is connected to the ECA attention module of the corresponding layer, namely the second ECA attention module, the third ECA attention module, the fourth ECA attention module, and the fifth ECA attention module.

4. The method for acquiring the three-dimensional morphology of sweat glands of human skin according to claim 2, characterized in that: The ECA attention modules connected to the Embedding layer and each encoding layer are respectively the first ECA attention module, the second ECA attention module, and the third ECA attention module, fourth ECA attention module and fifth ECA attention module; The decoding module includes 5 decoding layers connected in sequence, namely the first decoding layer, the second decoding layer, the third decoding layer, the fourth decoding layer and the fifth decoding layer; the input of the first decoding layer is the output of the fifth ECA attention module corresponding to the encoding layer, which is input into the second decoding layer together with the output of the fourth ECA attention module after the upsampling operation. The second decoding layer includes a DFF module and an upsampling operation; the third decoding layer, the fourth decoding layer and the fifth decoding layer have the same structure and all include a ConCate operation and an upsampling operation. The ConCate operation is used to perform channel splicing on the output of the previous decoding layer and the output of the ECA attention module of the corresponding layer. The upsampling operation is used to upsample the output after channel splicing. The upsampled output is the output of the decoding layer of this layer.

5. The method for acquiring the three-dimensional morphology of sweat glands of human skin according to claim 4, characterized in that: The first ECA attention module, the second ECA attention module and the third ECA attention module are further connected to a residual module; in the third decoding layer, the fourth decoding layer and the fifth decoding layer, a residual module is also provided between the Concate operation and the upsampling operation.

6. The method for acquiring the three-dimensional morphology of sweat glands of human skin according to claim 4, characterized in that: A DLK module is also connected after the fourth ECA attention module and the fifth ECA attention module; the input of the first decoding layer is the output of the DLK module after the fifth ECA attention module, which is input into the second decoding layer together with the output of the DLK module after the fourth ECA attention module after the upsampling operation of the first decoding layer; a DLK module is also provided between the DFF module and the upsampling operation of the second decoding layer.

7. The method for acquiring the three-dimensional morphology of sweat glands of human skin according to claim 6, characterized in that: The mixed loss function of the three-dimensional segmentation skin sweat gland network model is L SweatGland , which is expressed as follows: L SweatGland =αL BCE +βL Dice Among them, α and β are weight coefficients, L BCE is the binary cross entropy loss function, L Dice is the Dice loss function. The expressions of the two loss functions are as follows: Where N is the number of samples, Pi is the predicted probability of the i-th sample, and yi is the true label of the i-th sample.

8. The method for acquiring the three-dimensional morphology of sweat glands of human skin according to claim 1, characterized in that: Also includes: The initial skin OCT image of the area to be inspected is preprocessed, where the preprocessing includes but is not limited to one or more of cropping, flipping, resizing, and contrast adjustment.

9. A system for acquiring the three-dimensional morphology of sweat glands on human skin, characterized in that: include: An image acquisition module is used to acquire an initial skin OCT image of the area to be examined output by the optical coherence tomography system; A preprocessing module, configured to preprocess the initial skin OCT image of the area to be inspected, wherein the preprocessing includes but is not limited to one or more of cropping, flipping, resizing, and contrast adjustment; A segmentation module is used to input the pre-processed initial skin OCT image of the area to be inspected into a pre-trained deep learning-based 3D segmentation skin sweat gland network model. The deep learning model performs 3D segmentation on the pre-processed initial skin OCT image of the area to be inspected to obtain a 3D segmentation result of the sweat glands. The three-dimensional segmentation skin sweat gland network model includes an encoding module, an ECA attention module, a decoding module and a saliency layer; the encoding module includes an embedding layer and multiple encoding layers connected in sequence, the decoding module includes multiple decoding layers connected in sequence, the number of layers of the encoding module is the same as the number of layers of the decoding module, the ECA attention module is connected after the embedding layer and each encoding layer, and the output of the ECA attention module is jump-connected to the corresponding decoding layer in the decoding module; the saliency layer includes a 3*3*3 convolutional layer and a 1*1*1 convolutional layer, the image to be segmented is input into the 3*3*3 convolutional layer in the encoding module and the saliency layer respectively, the output of the 3*3*3 convolutional layer is ConCate-operated with the output of the decoding module and then input into the 1*1*1 convolutional layer to obtain the segmented image.

10. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program executable by the processor, and when the processor executes the computer program, the method for acquiring the three-dimensional morphology of sweat glands on human skin according to any one of claims 1 to 8 is implemented.

Images