Periorbital capacity judgment method and system based on machine learning and medium

Abstract

The invention relates to a machine learning-based periorbital capacity judgment method and system and a medium. The method comprises the following steps of: acquiring tomographic image data containing a periorbital region through CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) equipment; preprocessing the acquired sectional image data, wherein the preprocessing comprises noise removal, gray level normalization and region-of-interest extraction; inputting the preprocessed sectional image data into a deep learning model based on a lightweight U-Net network for layer-by-layer processing, and generating a binarization segmentation mask of an orbital region; the deep learning model carries out transfer learning based on the pre-training model; overlapping each layer of sectional image data after segmentation to generate a three-dimensional model of the orbital region; calculating the total capacity value of the three-dimensional model; and outputting the calculated total capacity value in a digital and graphical form. Compared with the prior art, the method for determining the periorbital capacity is rapid, accurate and low in cost, and has remarkable clinical application value and popularization potential.

Description

Technical Field

[0001] The present invention relates to the technical field of periorbital volume determination, and particularly to a method, system, and medium for periorbital volume determination based on machine learning. Background Art

[0002] The accurate determination of periorbital volume is of great significance in medical fields such as ophthalmology, dermatology, and plastic surgery. It can be used to assist in diagnosing diseases (such as thyroid-related ophthalmopathy, periorbital tumors, etc.) and evaluating the effects of surgeries and treatments (such as after periorbital tumor surgery, eyelid bag surgery, and beauty treatments such as periorbital radiofrequency laser). Currently, periorbital volume measurement mainly relies on imaging examinations (such as CT, MRI, and ultrasound), and is calculated through manual or semi-automatic segmentation. However, manual measurement has problems such as high time cost, large subjective errors, and high technical thresholds. Some small-area volume defects, such as tear troughs and eyelid bags, cannot be accurately measured by the above methods, which limits its application in large-scale medical scenarios. In recent years, the development of artificial intelligence, especially deep learning technologies (such as convolutional neural network CNN), has provided technical support for automatic image segmentation and accurate calculation of periorbital volume, but problems such as strong data dependence and insufficient model generalization still need to be solved.

[0003] Existing automated measurement schemes usually include steps such as image preprocessing, segmentation based on deep learning networks (such as U-Net), three-dimensional reconstruction and volume calculation, and result verification. Although these methods have improved efficiency to a certain extent, due to relying on high-quality labeled datasets and high-performance computing devices, it is difficult to promote them in ordinary medical institutions. At the same time, existing technologies mostly target single imaging modalities and are difficult to meet the needs of processing multi-modal images or low-quality images. In view of the above deficiencies, there is an urgent need to propose a more efficient and accurate method for periorbital volume determination to improve its universality and promotion. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects of the above existing technologies and provide a method, system, and medium for periorbital volume determination based on machine learning, so as to improve the measurement accuracy and efficiency of periorbital volume.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A method for periorbital volume determination based on machine learning includes the following steps:

[0007] Obtain tomographic image data containing the periorbital region through a CT or MRI device;

[0008] Perform preprocessing on the obtained tomographic image data, and the preprocessing includes noise removal, gray-scale normalization, and region of interest extraction;

[0009] Input the preprocessed tomographic image data into a deep learning model based on a lightweight U-Net network for layer-by-layer processing to generate a binary segmentation mask for the periorbital region; the deep learning model performs transfer learning based on a pre-trained model;

[0010] Overlay each layer of the segmented tomographic image data to generate a three-dimensional model of the periorbital region; calculate the total volume value of the three-dimensional model;

[0011] Output the calculated total volume value in numerical and graphical forms.

[0012] Further, the noise removal includes using Gaussian filtering or non-local means method to remove noise in the tomographic image data;

[0013] The gray normalization includes performing gray standardization on the tomographic image data, adjusting the brightness and contrast to enhance the segmentability of the periorbital region;

[0014] The region of interest extraction includes preliminarily calibrating the periorbital region through region growing or template matching techniques.

[0015] Further, the obtained tomographic image data is multi-modal data, including CT images and MRI images, and the multi-modal data is fused through a feature fusion algorithm to obtain the tomographic image data to be processed.

[0016] Further, the processing process of the feature fusion algorithm is image-level fusion, decision-level fusion or feature-level fusion;

[0017] The image-level fusion is: registering the CT image and the MRI image, and then performing image overlay to generate a fused tomographic image data for segmentation;

[0018] The decision-level fusion is: respectively processing the segmentation results of the CT image and the MRI image, and obtaining the final segmentation result through a weighted fusion strategy or a voting method;

[0019] The feature-level fusion is: extracting the texture features and edge features of the CT image and the MRI image, splicing the extracted features, and then inputting them into a deep learning model for segmentation processing.

[0020] Further, the method also includes comparing the calculated total volume value of the periorbital region with the corresponding manually measured value or standard value to verify the robustness and reliability of the method.

[0021] Further, the process of generating the three-dimensional model of the periorbital region and calculating the total volume value adopts a voxel integration algorithm, a contour interpolation method, a geometric shape fitting method or point cloud reconstruction;

[0022] The contour interpolation method is as follows: perform contour fitting on the segmented tomographic image data of each layer, generate a three-dimensional model through contour difference, and thus calculate the total volume value;

[0023] The geometric shape fitting method is as follows: fit the contour of the segmented area of the tomographic image data into a regular geometric shape, and then use the analytical volume formula to calculate the approximate total volume value;

[0024] The point cloud reconstruction is as follows: convert the segmented area of the tomographic image data into three-dimensional point cloud data, use the point cloud processing algorithm to reconstruct the orbital region, and perform volume calculation.

[0025] Furthermore, the method further includes: determining whether the acquired tomographic image data is defective, and if it is defective, using an image repair network or a deep learning repair model for patching;

[0026] Use adaptive median filtering or BM3D algorithm to remove noise from the acquired tomographic image data;

[0027] Extract both high-frequency and low-frequency features from the tomographic image data with low quality, and segment the orbital region through a dual-channel network model.

[0028] The present invention also provides an orbital volume judgment system based on machine learning, including:

[0029] A data acquisition and input module, used to acquire tomographic image data containing the orbital region through a CT or MRI device;

[0030] An image preprocessing module, used to preprocess the acquired tomographic image data, and the preprocessing includes noise removal, gray-scale normalization, and region of interest extraction;

[0031] A segmentation model module, used to input the preprocessed tomographic image data into a deep learning model based on a lightweight U-Net network for layer-by-layer processing to generate a binary segmentation mask of the orbital region; the deep learning model performs transfer learning based on a pre-trained model;

[0032] A three-dimensional reconstruction and volume calculation module, used to stack the segmented tomographic image data of each layer to generate a three-dimensional model of the orbital region; calculate the total volume value of the three-dimensional model;

[0033] A result output and verification module, used to output the calculated total volume value in numerical and graphical forms, and compare it with the corresponding manual measurement value or standard value to verify the robustness and reliability.

[0034] Furthermore, the noise removal includes using Gaussian filtering or non-local means method to remove noise in the tomographic image data;

[0035] The gray normalization includes performing gray standardization on the tomographic image data, adjusting the brightness and contrast to enhance the segmentability of the periorbital region;

[0036] The extraction of the region of interest includes preliminarily calibrating the periorbital region through region growing or template matching techniques;

[0037] The tomographic image data obtained by the data acquisition and input module is multi-modal data, including CT images and MRI images. The multi-modal data is fused through a feature fusion algorithm to obtain the tomographic image data to be processed;

[0038] The processing process of the feature fusion algorithm is image-level fusion, decision-level fusion or feature-level fusion;

[0039] The image-level fusion is as follows: registering the CT image and the MRI image, and then performing image superposition to generate the fused tomographic image data for segmentation;

[0040] The decision-level fusion is as follows: processing the segmentation results of the CT image and the MRI image respectively, and obtaining the final segmentation result through a weighted fusion strategy or a voting method;

[0041] The feature-level fusion is as follows: extracting the texture features and edge features of the CT image and the MRI image, splicing the extracted features, and then inputting them into a deep learning model for segmentation processing;

[0042] The process of the three-dimensional reconstruction and volume calculation module generating the three-dimensional model of the periorbital region and calculating the total volume value adopts a voxel integration algorithm, a contour interpolation method, a geometric shape fitting method or a point cloud reconstruction;

[0043] The contour interpolation method is as follows: performing contour fitting on each layer of the segmented tomographic image data, generating a three-dimensional model through contour interpolation, and thus calculating the total volume value;

[0044] The geometric shape fitting method is as follows: fitting the contour of the segmented region of the tomographic image data into a regular geometric shape, and then calculating the approximate total volume value using an analytical volume formula;

[0045] The point cloud reconstruction is as follows: converting the segmented region of the tomographic image data into three-dimensional point cloud data, reconstructing the periorbital region using a point cloud processing algorithm, and performing volume calculation;

[0046] The image preprocessing module further includes: judging whether the obtained tomographic image data is defective. If it is defective, an image repair network or a deep learning repair model is used for repair;

[0047] Adaptive median filtering or BM3D algorithm is used to remove noise from the obtained tomographic image data;

[0048] Extract both high - frequency and low - frequency features from tomographic image data of low quality, and segment the periorbital region through a dual - channel network model.

[0049] The present invention also provides a computer - readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the steps of the method as described above.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] (1) By integrating multi - modal image features, lightweight machine - learning models, and adaptive pre - processing techniques, the present invention proposes a more efficient and accurate method for periorbital volume determination to improve its universality and promotion.

[0052] Lightweight machine - learning model: The present invention designs a lightweight deep - learning segmentation model (such as an improved U - Net), which can run efficiently on ordinary computing devices while ensuring the accuracy and robustness of segmentation. By optimizing the network structure, the computational complexity is significantly reduced, making it suitable for grass - roots medical scenarios with limited resources.

[0053] Adaptive image pre - processing technology: Introduce an adaptive pre - processing technology that can automatically adjust processing parameters according to the noise level and gray - scale distribution of the image, thereby improving the segmentation effect of images of different qualities. It includes techniques such as noise removal, gray - scale normalization, and region - of - interest extraction, and is especially suitable for low - quality images and complex backgrounds.

[0054] Multi - modal image feature fusion: Support the processing of multi - modal medical images such as CT and MRI. By using a feature fusion algorithm, the advantages of different modalities are effectively combined to improve the applicability of the model in multiple scenarios. In the case of low - quality or partially missing images, multi - modal complementary information is used to improve the segmentation accuracy.

[0055] (2) The present invention also proposes an optimization process for three - dimensional reconstruction and volume calculation algorithms: Based on an optimization algorithm of voxel integration method and tomographic image superposition, it can quickly construct a three - dimensional periorbital region model and accurately calculate the periorbital volume. Improve the efficiency and accuracy of volume calculation and meet the processing requirements of large - scale image data.

[0056] (3) The present invention also proposes a technology for reducing data dependence: Adopt technologies such as transfer learning and self - supervised learning to reduce the dependence on high - quality labeled data and achieve high - precision image segmentation and volume measurement under limited data conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic flowchart of a method for periorbital volume determination based on machine learning provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention generally described and illustrated in the accompanying drawings herein can be arranged and designed in a variety of different configurations.

[0059] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but is merely representative of selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0060] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not require further definition and explanation in subsequent drawings.

[0061] Embodiment 1

[0062] As Figure 1 shown, this embodiment provides an orbital volume determination method based on machine learning, including the following steps:

[0063] S1: Obtain tomographic image data including the orbital region through a CT or MRI device;

[0064] The obtained tomographic image data can be imported into the system for storage, and the image format supports common medical image standards such as DICOM.

[0065] Preferably, the obtained tomographic image data can be multi-modal data, including CT images and MRI images. The multi-modal data is fused through a feature fusion algorithm to obtain the tomographic image data to be processed;

[0066] S2: Preprocess the obtained tomographic image data, and the preprocessing includes noise removal, gray-scale normalization, and region of interest extraction;

[0067] Specifically, noise removal includes using Gaussian filtering or non-local means method to remove noise in the tomographic image data;

[0068] Gray-scale normalization includes performing gray-scale standardization on the tomographic image data, adjusting brightness and contrast to enhance the segmentability of the orbital region;

[0069] Region of interest extraction includes preliminarily calibrating the orbital region through region growing or template matching technology to reduce the computational burden;

[0070] S3: Input the preprocessed tomographic image data into a deep learning model based on a lightweight U-Net network, and process the preprocessed image data layer by layer to generate a binary segmentation mask for the periorbital region; the deep learning model uses transfer learning technology to optimize the segmentation performance through a pre-trained model and improve the adaptability to different image data;

[0071] S4: Stack each layer of the segmented tomographic image data to generate a three-dimensional model of the periorbital region; calculate the total volume value of the three-dimensional model through a voxel integration algorithm;

[0072] S5: Output the calculated total volume value in numerical and graphical forms for doctors' reference; compare the results with manual measurements or standard values to verify the robustness and reliability of the method.

[0073] Optionally, the following solutions can be adopted for the above deep learning model:

[0074] a. Improvement based on classical segmentation networks:

[0075] Mask R-CNN: Use a Region Proposal Network (RPN) to locate the periorbital region and then perform mask segmentation to improve the segmentation accuracy.

[0076] DeepLab series: Process the periorbital region through atrous convolution to improve the segmentation effect of the model on edge details.

[0077] Swin Transformer: Introduce a segmentation network based on the Transformer architecture to enhance the modeling ability for large-size images.

[0078] b. Traditional machine learning methods: Implement preliminary segmentation based on region growing algorithms and threshold segmentation methods, and then optimize the segmentation boundary in combination with Support Vector Machine (SVM) or K-means clustering algorithms. Fit the image edge through an active contour model (Snake algorithm) to extract the periorbital region.

[0079] c. Based on weakly supervised learning: Combine a small amount of labeled data and automatically generate the segmentation region through weakly supervised segmentation methods (such as CAM-like visualization techniques) to reduce the dependence on high-quality labeled data.

[0080] Optionally, the following solutions can be adopted for the process of image preprocessing:

[0081] a. Based on frequency domain methods: Use Fourier transform and wavelet transform to perform frequency domain denoising and edge enhancement on the image to improve the quality and segmentation effect of the image.

[0082] b. Based on histogram equalization: Improve the gray-scale distribution of the image through histogram equalization, enhance the image contrast, and facilitate the model to extract edge features.

[0083] c. Image super-resolution reconstruction: Use deep learning super-resolution models such as SRCNN and ESRGAN to reconstruct low-resolution images and improve the quality of the input images.

[0084] Optionally, the process of multi-modal image fusion can be as follows:

[0085] a. Image-level fusion: Register CT and MRI images, superimpose multi-modal images, and generate a fused image with comprehensive information for segmentation.

[0086] b. Decision-level fusion: Process the segmentation results of CT and MRI images respectively, and obtain a more accurate final segmentation result through a weighted fusion strategy or voting method.

[0087] c. Feature-level fusion: Extract features such as texture and edges of different modal images, splice the features, and then input them into a machine learning model for segmentation and analysis.

[0088] Optionally, the process of 3D reconstruction and volume calculation can also be as follows:

[0089] a. Based on contour interpolation method: Fit the contours of the segmented continuous tomographic images, and generate a 3D model through contour interpolation method to reduce the calculation amount.

[0090] b. Based on geometric shape fitting method: Fit the contours of the segmented region into an ellipsoid or other regular geometric shapes, and perform approximate calculation using the analytical volume formula.

[0091] c. Based on point cloud reconstruction: Convert the segmented region into 3D point cloud data, use point cloud processing algorithms (such as Poisson reconstruction) to reconstruct the periorbital region, and perform volume calculation.

[0092] Preferably, the following solutions can also be added to reduce data dependence:

[0093] a. Transfer learning solution: Use a pre-trained medical image segmentation model (such as a model trained on other anatomical parts) for transfer learning to reduce the need for labeled data.

[0094] b. Data augmentation technology: Expand the existing dataset through data augmentation technologies such as rotation, flipping, cropping, and adding noise to improve the training effect of the model.

[0095] c. Synthetic data generation: Use a generative adversarial network (GAN) to generate high-quality virtual image data as a supplement to the training set to improve the robustness and generalization ability of the model.

[0096] Preferably, the processing of low-quality images can also be added to improve the adaptability of the method, as follows:

[0097] a. Image restoration: For defective or blurred images, use an image restoration network (such as Poisson restoration) or a deep learning restoration model (such as a deep residual network) for patching.

[0098] b. Noise adaptive removal: Use adaptive median filtering or the BM3D algorithm to remove noise from low-quality images.

[0099] c. Dual-channel processing: Simultaneously extract high-frequency and low-frequency features from low-quality images, and segment the periorbital region through a dual-channel network model to improve the processing effect.

[0100] The above is the introduction of the method embodiment. The following further illustrates the solution of the present invention through a system embodiment.

[0101] This embodiment also relates to a periorbital volume judgment system based on machine learning, including:

[0102] A data acquisition and input module for obtaining tomographic image data containing the periorbital region through CT or MRI equipment;

[0103] An image preprocessing module for preprocessing the obtained tomographic image data, and the preprocessing includes noise removal, gray normalization, and region of interest extraction;

[0104] A segmentation model module for inputting the preprocessed tomographic image data into a deep learning model based on a lightweight U-Net network for layer-by-layer processing to generate a binary segmentation mask of the periorbital region; the deep learning model performs transfer learning based on a pre-trained model;

[0105] A three-dimensional reconstruction and volume calculation module for superimposing each layer of the segmented tomographic image data to generate a three-dimensional model of the periorbital region; calculating the total volume value of the three-dimensional model;

[0106] A result output and verification module for outputting the calculated total volume value in digital and graphical forms, and comparing it with the corresponding manually measured value or standard value to verify the robustness and reliability.

[0107] Optionally, noise removal includes using Gaussian filtering or non-local means method to remove noise from the tomographic image data;

[0108] Gray normalization includes performing gray standardization on the tomographic image data to adjust the brightness and contrast to enhance the segmentability of the periorbital region;

[0109] Region of interest extraction includes preliminarily calibrating the periorbital region through region growing or template matching techniques;

[0110] The tomographic image data obtained by the data acquisition and input module is multi-modal data, including CT images and MRI images. The multi-modal data is fused through a feature fusion algorithm to obtain the tomographic image data to be processed;

[0111] The processing process of the feature fusion algorithm is image-level fusion, decision-level fusion or feature-level fusion;

[0112] Image-level fusion is: registering the CT image and the MRI image, and then performing image superposition to generate a fused tomographic image data for segmentation;

[0113] Decision-level fusion is: respectively processing the segmentation results of the CT image and the MRI image, and obtaining the final segmentation result through a weighted fusion strategy or a voting method;

[0114] Feature-level fusion is: extracting the texture features and edge features of the CT image and the MRI image, splicing the extracted features, and then inputting them into a deep learning model for segmentation processing;

[0115] The process of the three-dimensional reconstruction and volume calculation module generating the three-dimensional model of the orbital region and calculating the total volume value adopts a voxel integration algorithm, a contour interpolation method, a geometric shape fitting method or point cloud reconstruction;

[0116] The contour interpolation method is: performing contour fitting on each layer of the segmented tomographic image data, generating a three-dimensional model through contour difference, and thus calculating the total volume value;

[0117] The geometric shape fitting method is: fitting the contour of the segmented area of the tomographic image data into a regular geometric shape, and then using an analytical volume formula to approximately calculate the total volume value;

[0118] Point cloud reconstruction is: converting the segmented area of the tomographic image data into three-dimensional point cloud data, using a point cloud processing algorithm to reconstruct the orbital region, and performing volume calculation;

[0119] The image preprocessing module further includes: judging whether the obtained tomographic image data is defective, and if so, repairing it using an image repair network or a deep learning repair model;

[0120] Adaptive median filtering or BM3D algorithm is used to remove noise from the obtained tomographic image data;

[0121] For tomographic image data with low quality, high-frequency and low-frequency features are extracted simultaneously, and the orbital region is segmented through a dual-channel network model.

[0122] It should be noted that the specific content and beneficial effects of the system of the present application can be seen in the above method embodiments, and will not be repeated here.

[0123] This embodiment also provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the steps of the above-mentioned method for judging orbital volume based on machine learning.

[0124] In the context of the present invention, the computer-readable storage medium may be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium may be a machine-readable signal medium or a machine-readable storage medium. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0125] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. A periorbital volume determination method based on machine learning, characterized in that: The following steps are involved: Obtain tomographic image data including the periorbital area by CT or MRI equipment; Preprocessing the acquired tomographic image data includes noise removal, grayscale normalization and region of interest extraction; The preprocessed tomographic image data is input into a deep learning model based on a lightweight U-Net network for layer-by-layer processing to generate a binary segmentation mask of the periorbital area; the deep learning model performs transfer learning based on the pretrained model; The segmented tomographic image data of each layer are superimposed to generate a three-dimensional model of the periorbital area; Calculating a total capacity value of the three-dimensional model; The calculated total capacity value is output in numerical and graphical form.

2. A periorbital volume determination method based on machine learning according to claim 1, characterized in that: The noise removal includes using Gaussian filtering or non-local mean method to remove noise in the tomographic image data; The grayscale normalization includes grayscale normalization of the tomographic image data, adjusting brightness and contrast to enhance the segmentability of the periorbital area; The extraction of the region of interest includes preliminarily calibrating the periorbital region by region growing or template matching technology.

3. The method for determining periorbital volume based on machine learning according to claim 1, characterized in that: The acquired tomographic image data is multimodal data, including CT images and MRI images. The multimodal data are fused by a feature fusion algorithm to obtain the tomographic image data to be processed.

4. The method for determining periorbital volume based on machine learning according to claim 3, characterized in that: The processing process of the feature fusion algorithm is image level fusion, decision level fusion or feature level fusion; The image-level fusion is as follows: registering the CT image and the MRI image, and then superimposing the images to generate fused tomographic image data for segmentation; The decision-level fusion is as follows: the segmentation results of the CT image and the MRI image are processed respectively, and the final segmentation result is obtained through a weighted fusion strategy or a voting method; The feature-level fusion is as follows: extracting texture features and edge features of CT images and MRI images, splicing the extracted features, and inputting them into a deep learning model for segmentation processing.

5. The method for determining periorbital volume based on machine learning according to claim 1, characterized in that: The method further includes comparing the calculated total volume value of the periorbital area with a corresponding manually measured value or a standard value to verify the robustness and reliability of the method.

6. The method for determining periorbital volume based on machine learning according to claim 1, characterized in that: The process of generating the three-dimensional model of the periorbital area and calculating the total volume value adopts a voxel integration algorithm, a contour interpolation method, a geometric shape fitting method or a point cloud reconstruction; The contour interpolation method is: contour fitting is performed on each layer of tomographic image data after segmentation, and a three-dimensional model is generated through contour difference, thereby calculating the total capacity value; The geometric shape fitting method is: fitting the segmented area contour of the tomographic image data into a regular geometric shape, and then using the analytical volume formula to calculate the approximate total capacity value; The point cloud reconstruction is as follows: converting the segmented area of ​​the tomographic image data into three-dimensional point cloud data, reconstructing the periorbital area using a point cloud processing algorithm, and performing volume calculation.

7. The method for determining periorbital volume based on machine learning according to claim 1, characterized in that: The method further includes: determining whether the acquired tomographic image data has defects, and if so, repairing the defects using an image repair network or a deep learning repair model; Adaptive median filtering or BM3D algorithm is used to remove noise from the acquired tomographic image data; High-frequency and low-frequency features are extracted simultaneously for low-quality tomographic image data, and the periorbital area is segmented using a dual-channel network model.

8. A periorbital volume determination system based on machine learning, characterized in that: include: A data acquisition and input module, used for acquiring tomographic image data including the periorbital area through CT or MRI equipment; An image preprocessing module is used to preprocess the acquired tomographic image data, and the preprocessing includes noise removal, grayscale normalization and region of interest extraction; A segmentation model module is used to input the preprocessed tomographic image data into a deep learning model based on a lightweight U-Net network for layer-by-layer processing to generate a binary segmentation mask of the periorbital area; the deep learning model performs transfer learning based on the pretrained model; The three-dimensional reconstruction and volume calculation module is used to superimpose the segmented tomographic image data of each layer to generate a three-dimensional model of the periorbital area; Calculating a total capacity value of the three-dimensional model; The result output and verification module is used to output the calculated total capacity value in digital and graphical form, and compare it with the corresponding manual measurement value or standard value to verify the robustness and reliability.

9. The periorbital volume determination system based on machine learning according to claim 8, characterized in that: The noise removal includes using Gaussian filtering or non-local mean method to remove noise in the tomographic image data; The grayscale normalization includes grayscale normalization of the tomographic image data, adjusting brightness and contrast to enhance the segmentability of the periorbital area; The extraction of the region of interest includes preliminarily calibrating the periorbital region by region growing or template matching technology; The tomographic image data acquired by the data acquisition and input module is multimodal data, including CT images and MRI images, and the multimodal data is fused by a feature fusion algorithm to obtain tomographic image data to be processed; The processing process of the feature fusion algorithm is image level fusion, decision level fusion or feature level fusion; The image-level fusion is as follows: registering the CT image and the MRI image, and then superimposing the images to generate fused tomographic image data for segmentation; The decision-level fusion is as follows: the segmentation results of the CT image and the MRI image are processed respectively, and the final segmentation result is obtained through a weighted fusion strategy or a voting method; The feature-level fusion is as follows: extracting texture features and edge features of CT images and MRI images, splicing the extracted features, and inputting them into a deep learning model for segmentation processing; The three-dimensional reconstruction and volume calculation module generates a three-dimensional model of the periorbital area and calculates the total volume value by using a voxel integration algorithm, a contour interpolation method, a geometric shape fitting method or point cloud reconstruction; The contour interpolation method is: contour fitting is performed on each layer of tomographic image data after segmentation, and a three-dimensional model is generated through contour difference, thereby calculating the total capacity value; The geometric shape fitting method is: fitting the segmented area contour of the tomographic image data into a regular geometric shape, and then using the analytical volume formula to calculate the approximate total capacity value; The point cloud reconstruction comprises: converting the segmented area of ​​the tomographic image data into three-dimensional point cloud data, reconstructing the periorbital area using a point cloud processing algorithm, and performing volume calculation; The image preprocessing module further includes: determining whether the acquired tomographic image data has defects, and if so, repairing it using an image repair network or a deep learning repair model; Adaptive median filtering or BM3D algorithm is used to remove noise from the acquired tomographic image data; High-frequency and low-frequency features are extracted simultaneously for low-quality tomographic image data, and the periorbital area is segmented using a dual-channel network model.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is used by a processor to execute the method according to any one of claims 1 to 7.

Images