Lightweight choroidal neovascularization OCT image segmentation method and system

Abstract

The present application discloses a light-weight choroidal neovascularization OCT image segmentation method and system, and relates to the technical field of image segmentation.The present application comprises: receiving a choroidal neovascularization OCT image for preprocessing, and dividing the image data into a training set and a test set; constructing an improved segmentation model based on a UNeXt network, wherein the improved segmentation model comprises an encoder, a decoder and a skip connection, and the depth separable convolution blocks in the encoder and the decoder are replaced with excellent convolution modules.The present application can capture features at different levels and different scales by adding excellent convolution modules after the first three layers of the encoder, fully excavate useful information in the feature map, and introduce a deep feature extraction module that fuses multiple attention mechanisms, which can significantly improve the feature extraction capability and the generalization ability of the model.Meanwhile, the excellent convolution module introduced can focus on important features while saving computing resources by performing efficient and lightweight convolution operations on only part of the channels.

Description

Technical Field

[0001] This invention relates to the field of image segmentation technology, specifically to a lightweight OCT image segmentation method and system for choroidal neovascularization. Background Technology

[0002] Age-related macular degeneration (AMD) is one of the leading causes of vision loss and blindness in the elderly. Statistics show that there were approximately 196 million AMD patients worldwide in 2020, and this number is projected to increase to 288 million by 2040. Choroidal neovascularization (CNV) is a hallmark of exudative AMD, and its development leads to retinal damage and vision loss. Therefore, early detection and treatment of CNV are crucial for preventing vision loss. Optical coherence tomography (OCT) is a non-invasive imaging technique widely used in ophthalmology for the detection and diagnosis of CNV. Interpretation of OCT images often relies on the physician's subjective experience; furthermore, manual CNV annotation is a highly labor-intensive and time-consuming task. Therefore, there is a significant need for automated segmentation of choroidal neovascularization. However, automated CNV segmentation remains challenging due to blurred CNV boundaries, significant morphological variations, speckle noise, and interference from other similar diseases.

[0003] To address the aforementioned issues, deep learning has been increasingly applied in medical image processing in recent years, particularly in the classification of disease and non-disease images, where it has made significant progress. The efficiency of convolutional neural networks (CNNs) in object detection, image recognition, and segmentation has been widely recognized, demonstrating their importance in OCT image analysis. In particular, U-shaped networks using CNNs, with their unique encoder-decoder structure and skip connection structure, densely integrate shallow and deep features, and the size of the UNet model matches the amount of medical image data. They have been widely applied to various tasks. In addition, many Transformer-based networks have also been proposed for medical image segmentation because they learn a global understanding of the image, which aids in segmentation. For example, TransUNet has become a leading medical image segmentation method.

[0004] However, traditional segmentation methods focus on improving network performance, but the overall computational complexity, inference time, or number of parameters are relatively large. To address this, we propose a lightweight OCT image segmentation method and system for choroidal neovascularization. Summary of the Invention

[0005] The purpose of this invention is to provide a lightweight OCT image segmentation method and system for choroidal neovascularization, which can improve the accuracy and efficiency of segmentation while reducing the amount of computation and parameters.

[0006] Example 1:

[0007] To achieve the above objectives, the present invention provides the following technical solution: a lightweight OCT image segmentation method for choroidal neovascularization, comprising the following steps:

[0008] The OCT images of choroidal neovascularization are received, preprocessed, and the image data is divided into training and testing sets.

[0009] An improved segmentation model based on the UNeXt network is constructed. The improved segmentation model includes an encoder, a decoder, and skip connections. The depthwise separable convolutional blocks in the encoder and decoder are replaced with superconvolutional modules. A depth feature extraction module is introduced at the output of the first three layers of the encoder, and a tokenized MLP block is introduced in both the encoder and decoder.

[0010] The improved segmentation model is trained using the training set, with the goal of minimizing the overall loss function. The parameters of the improved segmentation model are then adjusted based on the validation set, and a new round of training is conducted until an improved segmentation model that meets the standards is obtained.

[0011] An optimized and improved segmentation model is used to identify and segment the ultrasound image to be identified.

[0012] Furthermore, the received choroidal neovascularization OCT images are preprocessed, and the image data is divided into training and testing sets, as follows:

[0013] (1) Receive choroidal neovascularization OCT image data and manually annotate the affected areas of the patient's OCT images;

[0014] (2) Randomly select some of the above-annotated images as training set samples and the remaining images as test set samples. Perform data augmentation processing on the training set samples to expand the training samples.

[0015] Furthermore, the encoder comprises 5 layers, with the first 3 layers using the same convolutional blocks and the last 2 layers using tokenized MLP blocks. Convolutional downsampling is used between each layer to reduce the feature map.

[0016] Furthermore, each of the convolutional blocks comprises a superconvolutional module, a 2D convolutional layer, a batch normalization layer, and a ReLU activation function.

[0017] Furthermore, the superior convolution module includes partial convolution and phantom convolution. The superior convolution module divides the number of channels of the input feature map into two parts. One part is subjected to phantom convolution operation, while the other part remains unchanged. This is used to concentrate computational resources on more important feature extraction while retaining some original information.

[0018] The phantom convolution generates more feature maps through inexpensive operations. These feature maps supplement the original features with low computational cost. The entire process is specifically represented as follows:

[0019] Y = Concat(X) keep GhostConv(X) ghost ))

[0020] In the formula, X is the input feature map, X keep X represents a channel that remains unchanged. ghost Represents the Ghost channel, GhostConv(X) ghost ) represents the phantom convolution operation performed on the phantom convolution channel, and Concat means concatenating the unchanged channel with the feature map after phantom convolution to finally obtain the output feature map Y.

[0021] Furthermore, the deep feature extraction module includes a CBAM module and a parallel processing module. The parallel processing module is located at the output of the CBAM module. The CBAM module applies a channel attention mechanism to calculate a channel attention map through global pooling and fully connected layers, and then applies a spatial attention mechanism to calculate a spatial attention map through pooling and convolution, which is used to enhance the expressive power of the feature map.

[0022] Furthermore, the parallel processing module includes an AG module and an SE module. The AG module enhances the features of the target region and suppresses background noise by calculating attention coefficients. The SE module compresses spatial dimension information into a global description through global average pooling and global max pooling, then calculates the channel weights through a fully connected layer, and finally redistributes the weights to each channel of the feature map. The specific process is as follows:

[0023] F refined =(F CBAM ·σ(FC(GAP(F CBAM ))))+(F CBAM ·σ(ReLU(Conv 1×1 (F CBAM ))))

[0024] In the formula, F CBAM This represents the feature map after processing by the CBAM module, σ(FC(GAP(F)). CBAM ))) represents the global average pooling and fully connected layer processing of the SE module, σ(ReLU(Conv) 1×1 (F CBAM ))) represents the convolution and ReLU processing of the AG module, and the final F refined It is a fine feature map that combines the two.

[0025] Furthermore, the decoder uses the decoding part of UNeXt to reconstruct high-resolution feature maps, which includes 5 layers. The first 2 layers are tokenized MLP blocks used to gradually restore feature resolution, and the last 3 layers are convolutional blocks. The decoder uses bilinear interpolation for upsampling.

[0026] Furthermore, the BCEDiceLoss is used as the overall loss function, specifically the sum of the BCE Loss and Dice loss functions. The BCE loss function (Binary Cross-Entropy Loss) is used for binary classification problems, and its formula is as follows:

[0027]

[0028] The Dice loss function is primarily used to evaluate overlap, and its formula is as follows:

[0029]

[0030] The overall loss function BCEDiceLoss is a weighted sum of the two loss functions mentioned above, and its formula is as follows:

[0031] BCEDiceLoss=BCE Loss+Dice Loss

[0032] In the formula, N represents the number of samples, y i This represents the true label of the i-th sample. This represents the predicted value for the i-th sample.

[0033] According to another aspect of the present invention, the present invention provides a lightweight choroidal neovascularization OCT image segmentation system, comprising:

[0034] The receiving module is used to receive OCT images of choroidal neovascularization, perform preprocessing, and divide the image data into training and testing sets.

[0035] The model building module is used to build an improved segmentation model based on the UNeXt network. The improved segmentation model includes an encoder, a decoder, and skip connections. The depthwise separable convolutional blocks in the encoder and decoder are replaced with superconvolutional modules. A depth feature extraction module is introduced at the output of the first three layers of the encoder, and a tokenized MLP block is introduced in both the encoder and decoder.

[0036] The training module is used to train the improved segmentation model using the training set. The objective function for optimizing the improved segmentation model is to minimize the overall loss function. The parameters of the improved segmentation model are adjusted based on the validation set, and a new round of training is carried out until an improved segmentation model that meets the standard is obtained.

[0037] The segmentation output module is used to identify and segment the ultrasound image to be identified using an optimized and improved segmentation model.

[0038] The present invention has at least the following beneficial effects:

[0039] 1. This invention adds an excellent convolutional module after the first three encoder layers, which can capture features at different levels and scales, fully mine useful information in the feature maps, and the introduced deep feature extraction module integrates multiple attention mechanisms, which can significantly improve the feature extraction capability and the model's generalization capability.

[0040] 2. The superior convolution module introduced in this invention saves computational resources by performing efficient and lightweight convolution operations on only a portion of the channels, thus focusing on important features.

[0041] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the structure of the improved segmentation model in this invention;

[0043] Figure 2 This is a schematic diagram of the deep feature extraction module in this invention;

[0044] Figure 3 This is a schematic diagram of the structure of the superior convolution module in this invention;

[0045] Figure 4 This is a comparison diagram of the segmentation results of the method proposed in this invention with other methods. Detailed Implementation

[0046] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0047] Please see Figure 1-4 This invention provides a technical solution: a lightweight OCT image segmentation method for choroidal neovascularization, comprising the following steps:

[0048] S1. Receive and preprocess OCT images of choroidal neovascularization, and divide the image data into training and testing sets, as follows:

[0049] (S1.1) Receive choroidal neovascularization OCT image data and manually annotate the affected areas in the patient's OCT images;

[0050] (S2.2) Randomly select some of the above-annotated images as training set samples and the remaining images as test set samples. Perform data augmentation processing on the training set samples to expand the training samples.

[0051] It should be noted that the CNV optical coherence tomography data used in this embodiment came from Shanghai First People's Hospital. The OCT data included 1070 OCT ultrasound images of 63 CNV patients. All images were manually annotated by experienced retinal specialists to ensure the accuracy and reliability of the data annotation.

[0052] Furthermore, data augmentation processing of the training set samples includes horizontal flipping, vertical flipping, and random rotation to alleviate the overfitting phenomenon of deep neural networks.

[0053] S2. Construct an improved segmentation model based on the UNeXt network. The improved segmentation model includes an encoder, a decoder, and skip connections. The depthwise separable convolutional blocks in the encoder and decoder are replaced with superconvolutional modules. A depth feature extraction module is introduced at the output of the first three layers of the encoder. Tokenized MLP blocks are introduced in both the encoder and decoder.

[0054] For the technical solution of this embodiment, the encoder uses the encoding part of UNeXt as the feature extractor, which includes 5 stages. The first 3 stages are composed of the same convolutional blocks, and the last 2 stages use tokenized MLP blocks. This module achieves efficient feature representation and extraction by converting convolutional features into tokens and processing them through MLP. Convolutional downsampling is used between each stage to reduce the feature map, thereby increasing the receptive field of convolution.

[0055] Furthermore, each convolutional block contains a superconvolutional module, a convolutional layer, a batch normalization layer, and a ReLU activation function;

[0056] The superior convolutional module (SConv) includes partial convolution and ghost convolution. Based on the idea of ​​partial convolution, SConv divides the number of channels of the input feature map into two parts: one part is subjected to ghost convolution, while the other part remains unchanged. This division allows computational resources to be concentrated on more important feature extraction while retaining some original information. Ghost convolution generates more feature maps through inexpensive operations (such as pointwise convolution). These feature maps supplement the original features with lower computational cost. The entire process can be described by the following formula:

[0057] Y = Concat(X) keep GhostConv(X) ghost ))

[0058] In the formula, X is the input feature map, X keep X represents a channel that remains unchanged. ghost Represents the Ghost channel, GhostConv(X) ghost ) represents the Ghost convolution operation performed on the Ghost channel, and Concat means concatenating the unchanged channel with the feature map after Ghost convolution to finally obtain the output feature map Y;

[0059] To enhance the model's flexibility and adaptability, the improved model offers two forward propagation methods (forward_slicing and forward_split_cat) suitable for different application scenarios and needs. forward_slicing is used during inference, performing Ghost Convolution only on the segmented channels; forward_split_cat can be used during both training and inference, rearranging the input features to accommodate Ghost Convolution operations. Through these design approaches, SConv reduces computational resource requirements while maintaining or even improving model performance.

[0060] Regarding the technical solution of this embodiment, the deep feature extraction module includes a CBAM module and a parallel processing module. The parallel processing module includes an AG module and an SE module. The feature map extracted by the convolutional layer is processed by the CBAM module, and then processed by the parallel structure of the AG and SE modules. Different types of attention mechanisms are used to enhance the features of the target region, adjust the channel weights, and simultaneously consider the channel and spatial dimensions of attention, thereby fully mining the useful information in the feature map, as detailed below:

[0061] (1) The CBAM module first applies the channel attention mechanism, calculates the channel attention map through global pooling and fully connected layers, and then applies the spatial attention mechanism, calculates the spatial attention map through pooling and convolution, further enhancing the expressive power of the feature map;

[0062] (2) The feature map after CBAM processing is then processed by the parallel structure of the AG and SE modules. The AG module enhances the features of the target region and suppresses background noise by calculating the attention coefficient, while the SE module compresses the spatial dimension information into a global description by global average pooling and global max pooling. Then, the weights of the channels are calculated through the fully connected layer, and finally the weights are redistributed to each channel of the feature map. This can enhance the features of important channels and suppress unimportant channels.

[0063] The entire processing flow is described by the following formula:

[0064] F refined =(F CBAM ·σ(FC(GAP(F CBAM ))))+(F CBAM ·σ(ReLU(Conv 1×1 (F CBAM ))))

[0065] In the formula, F CBAM This represents the feature map after processing by the CBAM module, σ(FC(GAP(F)). CBAM ))) represents the global average pooling and fully connected layer processing of the SE module, σ(ReLU(Conv) 1×1 (F CBAM ))) represents the convolution and ReLU processing of the AG module, and the final F refined It is a fine-grained feature map that combines the two;

[0066] Specifically, by combining the CBAM, AG, and SE modules, the Deep Feature Extraction (DEM) module can comprehensively utilize three different attention mechanisms to process and enhance feature maps from multiple perspectives. Furthermore, the combination of various attention mechanisms helps improve the model's generalization ability. The CBAM module considers attention in both channel and spatial dimensions, enhancing the overall expressive power of the feature maps. The AG module enhances the features of the target region, enabling the model to focus more on key regions when processing different images. The SE module adjusts channel weights, allowing the model to better adapt to different channel features.

[0067] Despite the introduction of multiple attention mechanisms, the DEM block is designed to minimize computational complexity through parallel structures and efficient weight calculation methods. After the initial processing by the CBAM module, the results are fused into the AG and SE modules. This design not only improves the feature extraction effect but also maintains high computational efficiency.

[0068] Regarding the technical solution of this embodiment, the decoder uses the decoding part of UNeXt to reconstruct high-resolution feature maps, which includes 5 stages. The first 2 stages are Tokenized MLP blocks, which are used to gradually restore feature resolution. The last 3 stages are convolutional blocks, which are the same as the convolutional blocks in the encoder. Each convolutional block contains a superconvolutional module, a convolutional layer, a batch normalization layer, and a ReLU activation function. The decoder uses bilinear interpolation for upsampling to increase the size of the feature map.

[0069] S3. Train the improved segmentation model using the training set, with minimizing the overall loss function as the objective function for optimizing the improved segmentation model. Adjust the parameters of the improved segmentation model based on the validation set and conduct a new round of training until an improved segmentation model that meets the standards is obtained.

[0070] For the technical solution in this embodiment, BCEDiceLoss is used as the overall loss function, specifically the sum of BCELoss loss function and Dice loss function. The BCE loss function (Binary Cross-Entropy Loss) is used for binary classification problems, and its formula is as follows:

[0071]

[0072] The Dice loss function is primarily used to evaluate overlap, and its formula is as follows:

[0073]

[0074] BCEDiceLoss is a weighted sum of the two loss functions mentioned above, and its formula is as follows:

[0075] BCEDiceLoss=BCE Loss+Dice Loss

[0076] In the formula, N represents the number of samples, y i This represents the true label of the i-th sample. Representing the predicted value of the i-th sample, the BCEDiceLoss loss function combines binary cross-entropy loss and Dice loss. By adding these two loss functions, it can better handle the class imbalance problem and improve the accuracy of segmentation tasks.

[0077] It should be noted that the loss function can also be the standard BCEWithLogitsLoss. This embodiment does not limit it. Select the appropriate loss function according to the configuration and use CUDA to accelerate it.

[0078] Furthermore, the improved segmentation model in this embodiment is based on the PyTorch deep learning framework, with a GPU configuration of an NVIDIA RTX4090 with 24GB of memory. It is trained for a total of 300 epochs, using a multivariate learning rate strategy with an initial learning rate of 1e-4 and a batch size of 16 during training. During training, the Adam optimizer is used to optimize the model parameters, and the learning rate is gradually reduced through the CosineAnnealingLR scheduler.

[0079] S4. The optimized and improved segmentation model is used to identify and segment the ultrasound image to be identified.

[0080] To quantitatively evaluate the performance of the method proposed in this embodiment, two commonly used evaluation metrics in medical image segmentation, Dice coefficient (DSC) and intersection-over-union ratio (IoU), were used as the benchmarks for the experimental results. DSC is a function that evaluates the similarity between the predicted result and the manually annotated image, and it is defined as follows:

[0081]

[0082] Where TP (True Positive) represents the number of true positive pixels, FP (False Positive) represents the number of false positive pixels, and FN (False Negative) represents the number of false negative pixels.

[0083] The intersection-union ratio (IUR) is the result of dividing the overlapping portion of two regions by the combined portion of the two regions, and it is defined as follows:

[0084]

[0085] Where A is the predicted region and B is the ground truth region. |A∩B| represents the intersection area of ​​the predicted segmentation region and the ground truth segmentation region (i.e., the number of overlapping pixels), and |A∪B| represents the union area of ​​the predicted segmentation region and the ground truth segmentation region (i.e., the total number of pixels, including all pixels in both the predicted and ground truth regions).

[0086] To demonstrate the effectiveness of the modules, the UNeXt model was selected as the base network. The corresponding ablation experiment results are shown in Table 1. The results show that the DEM and SConv modules proposed in this embodiment are effective. As shown in Table 1, starting with the original model, the segmentation accuracy increased after adding the DEM module, indicating that the DEM module enhanced the feature extraction capability. Next, the Funnel module was added to the model, further improving the Dice coefficient and IoU, demonstrating the advantage of the Funnel module in further refining features. Finally, the SConv module was added to the improved model, and the Dice coefficient and IoU reached 0.8790 and 0.7846, respectively, indicating that the SConv module significantly improved the accuracy and robustness of feature extraction. The number of channels is a major hyperparameter that affects the number of parameters, complexity, and performance of the model. C1 to C5 represent the number of channels in different blocks. Typically, we set C1 = 32, C2 = 64, C3 = 128, C4 = 160, and C5 = 256.

[0087] Furthermore, by setting the number of channels for each block, the number and complexity of model parameters can be flexibly controlled to adapt to different tasks and datasets.

[0088] This embodiment compares the proposed method with TransUnet and UNet++. The segmentation results of the different methods are as follows: Figure 4 As shown, the method in this embodiment achieves more accurate results than other methods, while also reducing the number of parameters and computational cost. Table 2 presents the quantitative segmentation results of different methods. The results show that the method in this embodiment has the highest Dice coefficient and IoU coefficient for all CNV segmentations.

[0089] Table 1 Comparison of ablation experiment segmentation results

[0090]

[0091] Table 2 Performance Comparison with SOTA Medical Image Segmentation Model

[0092]

[0093] In summary, the lightweight choroidal neovascularization OCT image segmentation method based on the improved UNeXt structure proposed in this embodiment can improve the accuracy and efficiency of segmentation while reducing the amount of computation and parameters.

[0094] Example 2:

[0095] This invention proposes a lightweight OCT image segmentation system for choroidal neovascularization, comprising:

[0096] The receiving module is used to receive OCT images of choroidal neovascularization, perform preprocessing, and divide the image data into training and testing sets.

[0097] The model building module is used to build an improved segmentation model based on the UNeXt network. The improved segmentation model includes an encoder, a decoder, and skip connections. The depthwise separable convolutional blocks in the encoder and decoder are replaced with superconvolutional modules. A depth feature extraction module is introduced at the output of the first three layers of the encoder, and a tokenized MLP block is introduced in both the encoder and decoder.

[0098] The training module is used to train the improved segmentation model using the training set. The objective function for optimizing the improved segmentation model is to minimize the overall loss function. The parameters of the improved segmentation model are adjusted based on the validation set, and a new round of training is carried out until an improved segmentation model that meets the standard is obtained.

[0099] The segmentation output module is used to identify and segment the ultrasound image to be identified using an optimized and improved segmentation model.

[0100] Specifically, the receiving module, model building module, training module, and segmentation output module can be embedded into a computer processing system. The computer, based on the lightweight choroidal neovascularization OCT image segmentation method provided above, calls each of the above modules to complete the task of equivalent modeling. The receiving module, model building module, training module, and segmentation output module can perform operations according to the specific steps given in the lightweight choroidal neovascularization OCT image segmentation method described above.

[0101] It should be noted that the division of the various modules in the above system is merely a division of logical functions. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. These modules can be implemented entirely in software through processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, the receiving module can be a separate processing element or integrated into a chip in the aforementioned device. Alternatively, it can be stored as program code in the memory of the aforementioned device, and called and executed by a processing element of the device. The implementation of other modules is similar. Furthermore, these modules can be fully or partially integrated together or implemented independently. The processing element mentioned here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0102] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to form a system-on-a-chip (SOC).

[0103] It should be noted that, in this embodiment, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0104] For those skilled in the art, the specific meanings of the above terms in this invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on," "mounted on," "fixed to," or "set on" another element, it may be directly on the other element or there may be an intermediate element present. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element present. The terms "vertical," "horizontal," "upper," "lower," "left," "right," and similar expressions used in this embodiment are for illustrative purposes only and do not represent the only possible implementation.

[0105] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0106] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A lightweight OCT image segmentation method for choroidal neovascularization, characterized in that, Includes the following steps: The OCT images of choroidal neovascularization are received, preprocessed, and the image data is divided into training and testing sets. An improved segmentation model based on the UNeXt network is constructed. The improved segmentation model includes an encoder, a decoder, and skip connections. The depthwise separable convolutional blocks in the encoder and decoder are replaced with superconvolutional modules. A depth feature extraction module is introduced at the output of the first three layers of the encoder, and a tokenized MLP block is introduced in both the encoder and decoder. The improved segmentation model is trained using the training set, with the goal of minimizing the overall loss function. The parameters of the improved segmentation model are then adjusted based on the validation set, and a new round of training is conducted until an improved segmentation model that meets the standards is obtained. An optimized and improved segmentation model is used to identify and segment the ultrasound image to be identified. The deep feature extraction module includes a CBAM module and a parallel processing module. The parallel processing module is located at the output of the CBAM module. The CBAM module applies a channel attention mechanism, which calculates a channel attention map through global pooling and fully connected layers. Then, it applies a spatial attention mechanism, which calculates a spatial attention map through pooling and convolution, to enhance the expressive power of the feature map. The parallel processing module includes an AG module and an SE module. The AG module enhances the features of the target region and suppresses background noise by calculating attention coefficients. The SE module compresses spatial dimensional information into a global description through global average pooling and global max pooling, then calculates the channel weights through a fully connected layer, and finally redistributes the weights to each channel of the feature map. The specific process is as follows: In the formula, This represents the feature map after processing by the CBAM module. This represents the global average pooling and fully connected layer processing of the SE module. The convolution and ReLU processing of the AG module, the final... It is a fine feature map that combines the two.

2. The lightweight choroidal neovascularization OCT image segmentation method according to claim 1, characterized in that, OCT images of choroidal neovascularization were received and preprocessed, and the image data were divided into training and testing sets, as follows: (1) Receive choroidal neovascularization OCT image data and manually annotate the affected areas of the patient's OCT images; (2) Randomly select some of the above-annotated images as training set samples and the remaining images as test set samples. Perform data augmentation processing on the training set samples to expand the training samples.

3. The lightweight choroidal neovascularization OCT image segmentation method according to claim 2, characterized in that: The encoder consists of 5 layers. The first 3 layers are the same convolutional blocks, and the last 2 layers use tokenized MLP blocks. Convolutional downsampling is used between each layer to reduce the feature map.

4. The lightweight choroidal neovascularization OCT image segmentation method according to claim 3, characterized in that: Each of the convolutional blocks comprises a superconvolutional module, a 2D convolutional layer, a batch normalization layer, and a ReLU activation function.

5. The lightweight choroidal neovascularization OCT image segmentation method according to claim 4, characterized in that: The superior convolution module includes partial convolution and phantom convolution. The superior convolution module divides the number of channels of the input feature map into two parts. One part is subjected to phantom convolution operation, while the other part remains unchanged. This is used to concentrate computational resources on more important feature extraction while retaining some original information. The phantom convolution generates more feature maps through inexpensive operations. These feature maps supplement the original features with low computational cost. The entire process is specifically represented as follows: In the formula, It is the input feature map. Represents a channel that remains unchanged. Represents the Ghost channel. This represents the phantom convolution operation performed on the phantom convolution channel. This means concatenating the unchanged channels and the feature maps after phantom convolution together to obtain the final output feature map. .

6. The lightweight choroidal neovascularization OCT image segmentation method according to claim 5, characterized in that: The decoder uses the decoding part of UNeXt to reconstruct high-resolution feature maps, which includes 5 layers. The first 2 layers are tokenized MLP blocks used to gradually restore feature resolution, and the last 3 layers are convolutional blocks. The decoder uses bilinear interpolation for upsampling.

7. The lightweight choroidal neovascularization OCT image segmentation method according to claim 1, characterized in that, The overall loss function is BCE Loss, which is the sum of the BCE Loss and Dice Loss functions. The BCE loss function (Binary Cross-Entropy Loss) is used for binary classification problems, and its formula is as follows: The Dice loss function is primarily used to evaluate overlap, and its formula is as follows: The overall loss function BCEDiceLoss is a weighted sum of the two loss functions mentioned above, and its formula is as follows: In the formula, Represents the number of samples. Representing the The true label of each sample Representing the The predicted value for each sample.

8. A lightweight choroidal neovascularization OCT image segmentation system, used to implement the lightweight choroidal neovascularization OCT image segmentation method according to any one of claims 1 to 7, characterized in that, include: The receiving module is used to receive OCT images of choroidal neovascularization, perform preprocessing, and divide the image data into training and testing sets. The model building module is used to build an improved segmentation model based on the UNeXt network. The improved segmentation model includes an encoder, a decoder, and skip connections. The depthwise separable convolutional blocks in the encoder and decoder are replaced with superconvolutional modules. A depth feature extraction module is introduced at the output of the first three layers of the encoder, and a tokenized MLP block is introduced in both the encoder and decoder. The training module is used to train the improved segmentation model using the training set. The objective function for optimizing the improved segmentation model is to minimize the overall loss function. The parameters of the improved segmentation model are adjusted based on the validation set, and a new round of training is carried out until an improved segmentation model that meets the standard is obtained. The segmentation output module is used to identify and segment the ultrasound image to be identified using an optimized and improved segmentation model.

Images