Muscular dystrophy typing diagnosis equipment and system based on medical image and storage medium

Abstract

The invention discloses a muscular dystrophy typing diagnosis device and system based on medical image.The device comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor. The method comprises the following steps: acquiring a multi-modal magnetic resonance imaging sequence of a hip region and a thigh region of a target object; inputting the multi-modal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain three-dimensional segmentation masks of the multiple target muscles; based on the multi-modal magnetic resonance imaging sequence and the three-dimensional segmentation mask, extracting a quantitative image feature set related to muscular dystrophy typing; and inputting the quantitative image feature set into a preset three-dimensional classification model to obtain a muscular dystrophy typing diagnosis result of the target object. According to the method, the three-dimensional segmentation masks of the multiple target muscles are obtained through the preset three-dimensional segmentation neural network model, and then the quantitative image feature set is extracted, so that the muscular dystrophy typing diagnosis efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a diagnostic device, system and storage medium for muscular dystrophy based on medical imaging. Background Technology

[0002] Dystrophinopathy is a recessive inherited muscle disease caused by a pathogenic mutation in the dystrophin (Dys) gene on the short arm of the X chromosome, Xp21. It is the most common type of muscular dystrophy in children. Progressive muscular dystrophy is a group of heterogeneous genetic disorders characterized by progressive weakness and atrophy of skeletal muscles. This rare disease is a neuromuscular disorder characterized by progressive, symmetrical muscle weakness and atrophy, caused by mutations in the dystrophin gene. It is an X-linked recessive genetic disorder that primarily affects males. Duchenne / Becker muscular dystrophy is inherited in an X-linked recessive manner, with no significant difference in incidence across countries, regions, and ethnicities, occurring in approximately 1 in 3,600–6,000 male births.

[0003] Both Duchenne muscular dystrophy (DMD) and Becker muscular dystrophy (BMD) present with progressive muscle weakness, initially affecting the proximal lower limbs, followed by more widespread muscle weakness. DMD is characterized by early onset, severe condition, rapid progression, and poor prognosis. Patients typically develop gait abnormalities between 3 and 5 years of age, gradually lose the ability to walk between 10 and 12 years of age, and die around 25 to 30 years of age due to heart and respiratory failure, making it the most common fatal muscle disease in children. In contrast, BMD partially expresses Dysplasia, usually has a later onset and slower progression, and patients may still be able to walk for 15 to 20 years after onset, resulting in a longer survival period and a relatively better prognosis. Previous studies have shown that early and long-term use of glucocorticoids can prolong the independent walking time in DMD, improve cardiopulmonary function, enhance the quality of life, and extend lifespan. Current consensus and guidelines regarding DMD both domestically and internationally recommend the long-term use of glucocorticoids in the early stages of DMD, after age 3 and before clinical function begins to decline. For BMD, considering the side effects of long-term use of glucocorticoids and the potential benefits of treatment, international studies generally do not recommend the use or long-term use of glucocorticoids in BMD.

[0004] However, due to the similar selective distribution patterns of fatty infiltration, atrophy, and hypertrophy in muscles between DMD and BMD, current methods of differentiation present certain challenges. Furthermore, conventional MRI imaging is subjective and cannot distinguish subtle differences between diseases. Clinical manual interpretation of MRI images ultimately leads to subjective variations, resulting in low diagnostic accuracy and efficiency. Therefore, how to improve the efficiency and accuracy of early classification and diagnosis of muscular dystrophy based on medical imaging is an urgent problem to be solved. Summary of the Invention

[0005] The main objective of this invention is to provide a medical imaging-based diagnostic device, system, and storage medium for muscular dystrophy, aiming to solve the technical problem of how to improve the efficiency and accuracy of muscular dystrophy classification diagnosis.

[0006] To achieve the above objectives, the present invention provides a medical imaging-based muscular dystrophy typing and diagnostic device. The device includes a memory, a processor, and a medical imaging-based muscular dystrophy typing and diagnostic program stored in the memory and running on the processor. When executed by the processor, the medical imaging-based muscular dystrophy typing and diagnostic program performs the following steps: Acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target subject; The multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles; Based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask, a quantitative image feature set related to muscular dystrophy classification is extracted; The quantitative image feature set is input into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

[0007] Optionally, the step of inputting the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles includes: The multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model. The encoder in the preset three-dimensional segmentation neural network model downsamples the multimodal magnetic resonance imaging sequence to extract multi-scale high-level semantic features. The multi-scale high-level semantic features are upsampled by the decoder in the preset three-dimensional segmentation neural network model, and semantic features from the corresponding level of the encoder are fused during the upsampling process to generate a three-dimensional segmentation mask for multiple target muscles.

[0008] Optionally, the step of extracting a quantitative image feature set related to muscular dystrophy classification based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask includes: The region of interest for each target muscle is determined based on the three-dimensional segmentation mask; Based on the fat signal images and water signal images in the multimodal magnetic resonance imaging sequence, the fat fraction of each region of interest is determined; Based on the three-dimensional segmentation mask and the multimodal magnetic resonance imaging sequence, image omics features of each region of interest are extracted, and the image omics features include at least one of histogram features, texture features, shape features and wavelet features; The fat fraction of each target muscle is fused with the radiomics features to obtain a quantitative image feature set associated with muscular dystrophy classification.

[0009] Optionally, after the step of extracting the radiomics features of each region of interest based on the three-dimensional segmentation mask and the multimodal magnetic resonance imaging sequence, the method further includes: The radiomics features are standardized to obtain standardized radiomics features. The standardized radiomics features are subjected to PCA dimensionality reduction to obtain the dimensionality-reduced radiomics features. Accordingly, the step of fusing the fat fraction of each target muscle with the radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification includes: The fat fraction of each target muscle is fused with the dimensionality-reduced radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification.

[0010] Optionally, before the step of acquiring multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target object, the method further includes: Acquire a sequence of sample images and an initial 3D segmentation neural network model, and preprocess the sample image sequence to obtain a training dataset; The initial 3D segmentation neural network model is trained based on the training dataset to obtain the training results; The parameters of the initial 3D segmentation neural network model are optimized based on the weighted cross-entropy loss function and the training results to obtain a preset 3D segmentation neural network model.

[0011] Optionally, after the step of inputting the quantitative image feature set into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object, the method further includes: The preset three-dimensional classification model is evaluated using preset model evaluation indicators to obtain evaluation results. The preset model evaluation indicators include accuracy, sensitivity, and F1 score. The formula for calculating the accuracy is: ; The formula for calculating the sensitivity is: ; The formula for calculating the F1 score is: ; in, Indicates accuracy rate. Indicates sensitivity, This represents the F1 score, and Precision represents the accuracy rate. TP indicates a true positive, TN indicates a true negative, FP indicates a false positive, and FN indicates a false negative.

[0012] Optionally, after the step of acquiring multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target object, the method further includes: The intensity normalization process is performed on the multimodal magnetic resonance imaging sequence to obtain a normalized multimodal magnetic resonance imaging sequence. The normalized multimodal magnetic resonance imaging sequence is cropped to obtain the cropped multimodal magnetic resonance imaging sequence. The cropped multimodal magnetic resonance imaging sequence is resampled based on a preset standardized resolution to obtain a preprocessed multimodal magnetic resonance imaging sequence. Accordingly, the step of inputting the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles includes: The preprocessed multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles.

[0013] Optionally, the multimodal magnetic resonance imaging sequence is a T2-weighted Dixon sequence, which includes images of four modalities: in-phase, out-of-phase, aqueous phase, and fatty phase; the preset three-dimensional classification model is a pre-trained Xception3D classifier.

[0014] Furthermore, to achieve the above objectives, the present invention also proposes a medical imaging-based diagnostic system for muscular dystrophy, the system comprising: The image acquisition module is used to acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target object; The muscle segmentation module is used to input the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles. The feature extraction module is used to extract a quantitative image feature set related to muscular dystrophy classification based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask. The result output module is used to input the quantitative image feature set into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

[0015] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing a medical image-based muscular dystrophy typing diagnostic program, wherein the medical image-based muscular dystrophy typing diagnostic program, when executed by a processor, performs the following steps: Acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target subject; The multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles; Based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask, a quantitative image feature set related to muscular dystrophy classification is extracted; The quantitative image feature set is input into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

[0016] This invention discloses a method for acquiring multimodal magnetic resonance imaging (MRI) sequences of the buttock and thigh regions of a target subject; inputting the MRI sequences into a preset three-dimensional segmentation neural network model to obtain three-dimensional segmentation masks for multiple target muscles; extracting quantitative image feature sets related to muscular dystrophy classification based on the MRI sequences and the three-dimensional segmentation masks; and inputting the quantitative image feature sets into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result for the target subject. Because this invention obtains three-dimensional segmentation masks for multiple target muscles through a preset three-dimensional segmentation neural network model, then extracts quantitative image feature sets related to muscular dystrophy classification based on the MRI sequences and the three-dimensional segmentation masks, and finally obtains the muscular dystrophy classification diagnosis result through a preset three-dimensional classification model, compared with existing technologies, this invention improves the efficiency and accuracy of muscular dystrophy classification diagnosis. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention. Figure 2This is an example image acquisition diagram of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention; Figure 3 This is a schematic diagram of the model structure of the preset three-dimensional segmentation neural network model in the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention. Figure 4 This is a flowchart illustrating the second embodiment of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention. Figure 5 This is a schematic diagram of the parallel processing architecture of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention. Figure 6 This is a flowchart illustrating the third embodiment of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention. Figure 7 This is a structural block diagram of the first embodiment of the medical imaging-based muscular dystrophy classification and diagnostic system of the present invention; Figure 8 This is a schematic diagram of the structure of a medical imaging-based muscular dystrophy typing diagnostic device, which is part of the hardware operating environment of the embodiment of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0020] This invention provides a medical imaging-based diagnostic device for muscular dystrophy, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention.

[0021] In this embodiment, the medical image-based muscular dystrophy typing and diagnostic device includes: a memory, a processor, and a medical image-based muscular dystrophy typing and diagnostic program stored in the memory and running on the processor. When the medical image-based muscular dystrophy typing and diagnostic program is executed by the processor, steps S10 to S40 are implemented: Step S10: Acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target subject.

[0022] It should be noted that the executing entity in this embodiment can be a computer server device with data processing, network communication, and program execution functions applied in the muscular dystrophy classification and diagnosis scenario, such as a server, tablet computer, or personal computer, or an electronic device capable of performing the above functions (such as a muscular dystrophy classification and diagnosis device based on medical images). The following uses a muscular dystrophy classification and diagnosis device based on medical images as an example to illustrate this embodiment and the subsequent embodiments.

[0023] It should be understood that multimodal magnetic resonance imaging sequence can refer to an advanced magnetic resonance imaging technique that can acquire a set of raw data in a single magnetic resonance scan through specific radio frequency pulse and gradient field design, and simultaneously reconstruct multiple images that are physically independent and complementary in information using post-processing algorithms.

[0024] It should be noted that, in this embodiment, the aforementioned multimodal magnetic resonance imaging sequence is a T2-weighted Dixon sequence, which includes images of four modalities: in-phase, out-of-phase, aqueous phase, and fat phase. Inputting these four modalities into a preset three-dimensional segmentation neural network model facilitates high-precision automatic muscle segmentation.

[0025] For example, refer to Figure 2 , Figure 2This diagram illustrates an example of image acquisition for the medical imaging-based muscular dystrophy classification and diagnostic device of this invention. The process begins in part (a) on the left, where raw T2 Dixon sequences (i.e., T2-weighted Dixon sequences) of the target subjects are acquired using magnetic resonance imaging (MRI). Then, in part (b) in the middle, this raw data undergoes a rigorous data screening and integration process: First, the data is filtered based on conditions such as age, resulting in two subsets: high-quality sequences (108 cases, >5 years old) and sequences providing high-quality sequences (120 cases, >8 years old). Next, these two subsets are merged into a dataset of 228 cases. Finally, after removing unlabeled samples (2 cases), a high-quality, labeled dataset of 226 cases is obtained for all subsequent analyses. This screening process ensures the reliability of the data for subsequent model training and testing. Section (c) on the right shows example images of specific multimodal sequences generated from the processed data, including in-phase, out-of-phase, water, and lipid images. These images collectively constitute a multimodal magnetic resonance imaging sequence. This workflow provides a high-quality, standardized data foundation for subsequent automatic muscle segmentation, quantitative image feature extraction, and final subtyping diagnosis. The entire workflow reflects the key preprocessing steps from raw data acquisition and quality control to standardized image generation, which are important prerequisites for achieving accurate and automated diagnosis.

[0026] In specific implementation, to improve the accuracy and stability of the final generated muscular dystrophy classification diagnosis results, after the step of acquiring multimodal magnetic resonance imaging (MRI) sequences of the target subject's buttock and thigh regions, the method further includes: performing intensity normalization processing on the multimodal MRI sequences to obtain normalized multimodal MRI sequences; cropping the normalized multimodal MRI sequences to obtain cropped multimodal MRI sequences; and resampling the cropped multimodal MRI sequences based on a preset normalized resolution to obtain preprocessed multimodal MRI sequences.

[0027] It should be understood that intensity normalization can refer to standardizing the grayscale value (i.e., signal intensity) of each pixel (voxel) in an image to eliminate or reduce non-pathological signal intensity variations caused by differences in scanning equipment models, scanning parameter settings, and individual patient differences. For example, it can scale the intensity values ​​of all images to a fixed range (such as between 0 and 1), or adjust them based on the statistical characteristics of the images themselves (such as mean and standard deviation) to ensure that the intensity distribution of all input data remains consistent.

[0028] Image cropping refers to removing irrelevant or redundant parts of an image. Resampling refers to changing the spatial resolution of an image (i.e., the actual physical size represented by each voxel), uniformly converting all images with different original resolutions to the same, preset voxel grid size. For example, in this embodiment, multimodal magnetic resonance imaging sequences can be uniformly preprocessed: intensity normalization, image cropping, and resampling to 128 voxels. 128 A standardized resolution of 128 is used to accommodate the processing needs of subsequent network models.

[0029] Step S20: Input the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles.

[0030] Accordingly, the step of inputting the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles includes: inputting the preprocessed multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles.

[0031] It should be understood that the preset 3D segmentation neural network model can refer to a trained deep learning model that can automatically identify and delineate the boundaries of a specific tissue (in this case, muscle) from 3D medical images. In this embodiment, the preset 3D segmentation neural network model can be an encoder-decoder structure based on U-Net or its variants (such as UNETR++).

[0032] In a specific implementation, the multimodal magnetic resonance imaging sequence can be input into a preset three-dimensional segmentation neural network model. The encoder in the preset three-dimensional segmentation neural network model downsamples the multimodal magnetic resonance imaging sequence to extract multi-scale high-level semantic features. The decoder in the preset three-dimensional segmentation neural network model upsamples the multi-scale high-level semantic features and fuses the semantic features from the corresponding level of the encoder during the upsampling process to generate a three-dimensional segmentation mask for multiple target muscles.

[0033] It should be noted that multi-scale high-level semantic features can refer to the abstract data representation with rich semantic information that a pre-defined three-dimensional segmentation neural network model learns and retains in its deep structure, which can simultaneously represent local details to global structure.

[0034] It should be understood that muscles in muscular dystrophy are often accompanied by fatty infiltration and atrophy, resulting in blurred boundaries between muscle and adipose tissue on magnetic resonance imaging. This embodiment upsamples multi-scale high-level semantic features through a decoder and fuses semantic features from the corresponding level of the encoder during the upsampling process. This allows for a better understanding and inference of these unclear anatomical boundaries, thereby achieving reliable segmentation even in pathological conditions.

[0035] It should be noted that the target muscles can refer to multiple specific muscle groups located in the lower limbs (hips and thighs) that are medically significant for diagnosing muscular dystrophy (especially DMD / BMD classification). In this embodiment, the target muscles may include the gluteus maximus, tensor fasciae latae, vastus lateralis, vastus intermedius, vastus medialis, rectus femoris, sartorius, adductor longus, adductor magnus, gracilis, semimembranosus, semitendinosus, long head of biceps femoris, and short head of biceps femoris, totaling 14 muscles. This embodiment further expands the segmentation range to 14 muscles, including the gluteus maximus, adductor magnus, and vastus intermedius, ensuring maximum information capture of disease-related areas. This broad muscle coverage provides more robust quantitative features, effectively addressing the limitation of existing deep learning methods that are restricted to limited anatomical regions.

[0036] It should be explained that a 3D segmentation mask can refer to a standardized digital map that marks the "boundary and name of each muscle" after the multimodal magnetic resonance imaging sequence of the buttock and thigh regions of a target object is automatically and pixel-level identified using a preset 3D segmentation neural network model.

[0037] For example, refer to Figure 3 , Figure 3 This is a schematic diagram of the pre-set three-dimensional segmentation neural network model in the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention. The pre-set three-dimensional segmentation neural network model includes an encoder (such as...). Figure 3 (a) in the middle), decoder (e.g.) Figure 3 (b) and the bottleneck layer.

[0038] The encoder consists of multiple context blocks, whose main function is to downsample the input data (i.e., multimodal magnetic resonance imaging sequences), extract multi-scale high-level semantic features, and capture contextual information. The encoder extracts semantic features layer by layer and significantly reduces the number of parameters in deeper modules through a weight-sharing strategy, improving computational efficiency. Context Block 1 contains multiple 3×3×3 convolutional operations (Conv k=3, for adapting to 3D data) and introduces the nonlinear LeakyReLU activation function, which inherits the advantages of ReLU's high computational efficiency and non-saturation. It also employs the regularization method Dropout3d to prevent overfitting. Feature fusion is achieved through voxel-level addition (⊕), and the output is finally processed by normalization (Norm, for stable training) and LeakyReLU. Context Blocks 2-4 have a structure basically the same as Context Block 1, but adjustments have been made to details such as the number of convolutional layers and dropout operations to progressively downsample and extract deeper contextual features.

[0039] The decoder consists of multiple localization blocks. It upsamples the features extracted by the encoder, gradually restoring the spatial resolution of the data and achieving accurate spatial localization. It innovatively employs a multi-scale feature fusion strategy, extracting and upsampling features from intermediate layers, then stacking them voxel-by-voxel to the final output, further refining muscle boundaries and improving the segmentation accuracy of 14 muscles (such as the gluteus maximus and adductor magnus). The model outputs a 3D segmentation map (i.e., a 3D segmentation mask) with 14 channels. Localization Blocks 1-3 include operations such as 3×3×3 convolution (Conv k=3), normalization, LeakyReLU activation, and upsampling to upsample and refine the features from the encoder, gradually restoring spatial information. Simultaneously, it fuses features from different stages of the encoder by concatenating (©) along the channel dimensions. Localization Block 4 has a relatively simple structure, mainly using 3×3×3 convolution (Conv k=3), normalization, and LeakyReLU activation to finally refine the features, preparing them for the output. In the later stages of the decoder, the feature map resolution is gradually restored to the input or target size through 1×1×1 convolution (Conv k=1, adjusting the number of channels) and upsampling operations, combined with voxel-level addition (⊕); at the same time, features from different stages are fused to ensure output accuracy.

[0040] The bottleneck layer is located between the encoder and the decoder, and it plays a role in feature integration and transition. It usually compresses the feature dimension first and then expands it to balance the amount of computation and the feature representation ability.

[0041] Step S30: Based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask, extract a quantitative image feature set related to the classification of muscular dystrophy.

[0042] It should be noted that a quantitative image feature set can refer to a set of features extracted from multimodal magnetic resonance imaging sequences and three-dimensional segmentation masks that can quantitatively characterize the pathological state of muscles and be used to distinguish different subtypes of muscular dystrophy (such as DMD and BMD).

[0043] In this embodiment, the quantitative image feature set may include fat fraction and radiomics features. Fat fraction is a clinically significant indicator that can be directly calculated from the physical properties of multimodal magnetic resonance imaging sequences and is used to quantify the degree of fat infiltration in muscles. Radiomics features may include at least one of histogram features, texture features, shape features, and wavelet features.

[0044] Step S40: Input the quantitative image feature set into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

[0045] It should be noted that the preset 3D classification model can refer to a pre-trained computer algorithm model capable of automatically distinguishing between DMD and BMD based on input image features. In this embodiment, the preset 3D classification model refers to the pre-trained Xception3D classifier. By inputting fat fraction and radiomics features as a quantitative image feature set into the preset 3D classification model, the robustness and clinical reliability of the model can be enhanced, and accurate classification of DMD and BMD can be achieved.

[0046] It should be understood that the classification diagnosis of muscular dystrophy can refer to the judgment conclusion on the subtype classification of the specific disease dystrophinopathy, that is, to distinguish between Duchenne muscular dystrophy (DMD) and Benedictine muscular dystrophy (BMD).

[0047] In a specific implementation, in order to evaluate the model performance of the preset three-dimensional classification model, after the step of inputting the quantitative image feature set into the preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object, the method further includes: using preset model evaluation indicators to evaluate the preset three-dimensional classification model and obtain evaluation results. The preset model evaluation indicators include accuracy, sensitivity and F1 score.

[0048] The formula for calculating the accuracy is: ; The formula for calculating the sensitivity is: ; The formula for calculating the F1 score is: ; in, Indicates accuracy rate. Indicates sensitivity, This represents the F1 score, and Precision represents the accuracy rate. TP (True Positive) represents a true positive, meaning the DMD patient was correctly predicted and classified as DMD; TN (True Negative) represents a true negative, meaning the BMD patient was correctly predicted and classified as BMD; FP (False Positive) represents a false positive, meaning the BMD patient was incorrectly predicted and classified as DMD; FN (False Negative) represents a false negative, meaning the DMD patient was incorrectly predicted and classified as BMD.

[0049] It should be noted that accuracy Sensitivity measures the proportion of samples correctly predicted by the model out of the total sample size; it is the most intuitive and commonly used metric. The F1 score measures the proportion of samples that the model correctly identifies as positive out of all actual positive samples. It is the harmonic mean of precision and sensitivity, where precision measures how many samples predicted as positive by the model are actually positive. A high F1 score avoids the extremes of pursuing only high sensitivity (leading to many false positives, i.e., low precision) or only high precision (leading to many false negatives, i.e., low sensitivity). A high F1 score indicates that the model captures the most true positive samples without sacrificing accuracy, i.e., it correctly predicts DMD classifications to the greatest extent possible.

[0050] This embodiment discloses a method for acquiring multimodal magnetic resonance imaging (MRI) sequences of the buttock and thigh regions of a target subject. The MRI sequences are input into a preset three-dimensional segmentation neural network model to obtain three-dimensional segmentation masks for multiple target muscles. Based on the MRI sequences and the three-dimensional segmentation masks, a quantitative image feature set related to muscular dystrophy classification is extracted. This quantitative image feature set is then input into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result for the target subject. Because this embodiment obtains three-dimensional segmentation masks for multiple target muscles through a preset three-dimensional segmentation neural network model, then extracts a quantitative image feature set related to muscular dystrophy classification based on the MRI sequences and the three-dimensional segmentation masks, and finally obtains the muscular dystrophy classification diagnosis result through a preset three-dimensional classification model, this embodiment improves the efficiency and accuracy of muscular dystrophy classification diagnosis compared to existing technologies.

[0051] refer to Figure 4 , Figure 4 This is a flowchart illustrating the second embodiment of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention.

[0052] Based on the first embodiment described above, in this embodiment, step S30 includes steps S301 to S304: Step S301: Determine the region of interest for each of the target muscles based on the three-dimensional segmentation mask.

[0053] Step S302: Based on the fat signal image and water signal image in the multimodal magnetic resonance imaging sequence, determine the fat fraction of each region of interest.

[0054] Step S303: Based on the three-dimensional segmentation mask and the multimodal magnetic resonance imaging sequence, extract the image omics features of each region of interest. The image omics features include at least one of histogram features, texture features, shape features, and wavelet features.

[0055] Step S304: The fat fraction of each target muscle is fused with the radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification.

[0056] Understandably, a region of interest (ROI) refers to a specific area in an image that is delineated or marked and requires specialized analysis and processing. In the specific implementation, a 3D segmentation mask (i.e., a 14-muscle mask) of multiple target muscles output from a pre-defined 3D segmentation neural network model is used as the ROI, combined with fat and water signal images from a multimodal magnetic resonance imaging sequence. The fat fraction of the ROI is calculated according to the clinical fat fraction formula. ; In the formula, FF represents the relative proportion of fat within the region of interest (ROI). Fat is the total fat voxel intensity within the ROI. Water is the total water voxel intensity within the ROI. This formula accurately calculates the fat content within each muscle's ROI; it is a unitless ratio ranging from 0 to 1. The quantitative FF values ​​for the 14 muscles provide clinicians with an objective and interpretable indicator of the severity of muscle disease, serving as an important basis for subsequent DMD and BMD classification. The FF value represents the fat content in the muscle and is presented as a percentage.

[0057] It should be understood that fat fraction (FF) primarily reflects the overall proportion of muscle tissue replaced by fat. Radiomic features (such as texture features) can capture microscopic spatial distribution patterns and heterogeneity during fat infiltration that are indistinguishable to the human eye (e.g., whether fat is uniformly distributed or distributed in a speckled or grid-like pattern). This embodiment provides a precise, objective, and quantitative value between 0-100% for each muscle by calculating the fat fraction (FF), effectively eliminating subjectivity.

[0058] Understandably, a single fat fraction can be affected by scanning parameters, and a single radiomics feature may overfit. By combining these two factors, the model can make a comprehensive judgment. For example, even if a muscle has a low fat fraction (FF) value, but its texture features show early abnormal changes, the model may still issue an accurate warning, thereby improving the accuracy, sensitivity, and reliability of early diagnosis and subtyping diagnosis (DMD and BMD).

[0059] For example, refer to Figure 5 , Figure 5 This diagram illustrates the parallel processing architecture of the medical imaging-based muscular dystrophy (DMD) classification diagnostic device of this invention. First, MRI images (i.e., multimodal magnetic resonance imaging sequences) are acquired. Then, the MRI images are input into a 3D UNet model (i.e., a pre-defined three-dimensional segmentation neural network model) to obtain segmentation results (i.e., three-dimensional segmentation masks for multiple target muscles). This process can be supervised and optimized using a segmentation loss function to ensure segmentation accuracy. Simultaneously, the MRI images enter the FF calculation branch, which calculates FF scores based on the MRI images. Subsequently, the segmentation results and FF scores are jointly input into a classification branch (i.e., a pre-defined three-dimensional classification model), outputting the diagnostic results of DMD / BMD (i.e., muscular dystrophy classification diagnostic results), and the model is optimized using a classification loss function.

[0060] To further eliminate redundancy and improve efficiency, in a specific implementation, after the step of extracting the radiomics features of each region of interest based on the three-dimensional segmentation mask and the multimodal magnetic resonance imaging sequence, the implementation may further include: standardizing the radiomics features to obtain standardized radiomics features; and performing PCA dimensionality reduction on the standardized radiomics features to obtain dimensionality-reduced radiomics features.

[0061] Accordingly, the step of fusing the fat fraction of each target muscle with the radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification includes: fusing the fat fraction of each target muscle with the dimensionality-reduced radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification.

[0062] This embodiment discloses a method for determining the regions of interest (ROIs) of each target muscle based on a 3D segmentation mask; determining the fat fraction of each ROI based on fat and water signal images from a multimodal magnetic resonance imaging (MRI) sequence; extracting radiomics features of each ROI based on the 3D segmentation mask and the multimodal MRI sequence, wherein the radiomics features include at least one of histogram features, texture features, shape features, and wavelet features; and fusing the fat fraction of each target muscle with the radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification. Because this embodiment fuses the fat fraction of each target muscle with radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification, compared to existing technologies, this embodiment achieves multi-dimensional and interpretable quantitative feature fusion, enhancing the robustness and accuracy of the pre-defined 3D classification model, thereby improving the accuracy and reliability of muscular dystrophy classification diagnosis.

[0063] refer to Figure 6 , Figure 6 This is a flowchart illustrating the third embodiment of the medical imaging-based muscular dystrophy classification and diagnostic device of the present invention.

[0064] Based on the above embodiments, in this embodiment, before step S10, steps S01 to S03 are further included: Step S01: Obtain the sample image sequence and the initial 3D segmentation neural network model, and preprocess the sample image sequence to obtain the training dataset.

[0065] Step S02: Train the initial 3D segmentation neural network model based on the training dataset to obtain the training results.

[0066] Step S03: Optimize the parameters of the initial 3D segmentation neural network model based on the weighted cross-entropy loss function and the training results to obtain a preset 3D segmentation neural network model.

[0067] It should be understood that the preset 3D segmentation neural network model is essentially a 3D CNN based on the U-Net architecture, focusing on accurately extracting multi-scale features in the hip and thigh muscle segmentation task. When training the initial 3D segmentation neural network model, the SGD optimizer can be selected, and the momentum (e.g., 0.9) can be set. This more robust optimization strategy helps accelerate convergence and overcome local optima. The initial learning rate can be set to 0.01, but considering the large number of parameters and memory limitations of the 3D model, a step decay mechanism is used to ensure stable model convergence. The weight decay can be set in the range of 10⁻⁴ to 10⁻⁵ to limit model complexity and prevent overfitting to the training data.

[0068] Furthermore, since the sample sizes of DMD and BMD in the training dataset may be imbalanced, weighted cross-entropy loss can be used to improve the model's ability to identify marginal classes by penalizing minority or hard-to-classify samples. The weighted random sampler, working in conjunction with the loss function, ensures that minority classes receive sufficient attention during training. Due to the limited memory of the developed hardware, batch size and training epochs can be set (e.g., batch size set to 1, training epochs set to 80), and an early stopping strategy can be employed to ensure the optimal preset 3D segmentation neural network model is obtained without overfitting.

[0069] This embodiment discloses acquiring a sample image sequence and an initial 3D segmentation neural network model, preprocessing the sample image sequence to obtain a training dataset, training the initial 3D segmentation neural network model based on the training dataset to obtain training results, and optimizing the parameters of the initial 3D segmentation neural network model according to the weighted cross-entropy loss function and the training results to obtain a preset 3D segmentation neural network model. Because this invention optimizes the parameters of the initial 3D segmentation neural network model according to the weighted cross-entropy loss function and the training results to obtain a preset 3D segmentation neural network model, compared with the prior art, this embodiment solves the problem of imbalanced sample numbers between DMD and BMD through the weighted cross-entropy loss function, enabling the preset 3D segmentation neural network model to better focus on minority class samples and improving the accuracy of muscle segmentation by the preset 3D segmentation neural network model.

[0070] Furthermore, embodiments of the present invention also propose a storage medium storing a medical image-based muscular dystrophy typing diagnostic program, wherein the medical image-based muscular dystrophy typing diagnostic program, when executed by a processor, performs the following steps: Acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target subject; The multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles; Based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask, a quantitative image feature set related to muscular dystrophy classification is extracted; The quantitative image feature set is input into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

[0071] Reference Figure 7 , Figure 7 This is a structural block diagram of the first embodiment of the muscular dystrophy classification and diagnostic system based on medical imaging of the present invention.

[0072] like Figure 7 As shown, the muscular dystrophy classification and diagnosis system based on medical imaging proposed in this embodiment of the invention includes: an image acquisition module 701, a muscle segmentation module 702, a feature extraction module 703, and a result output module 704.

[0073] The image acquisition module 701 is used to acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target object.

[0074] The muscle segmentation module 702 is used to input the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles.

[0075] The feature extraction module 703 is used to extract a quantitative image feature set related to the classification of muscular dystrophy based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask.

[0076] The result output module 704 is used to input the quantitative image feature set into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

[0077] The image acquisition module 701 is further configured to perform intensity normalization processing on the multimodal magnetic resonance imaging sequence to obtain a normalized multimodal magnetic resonance imaging sequence; perform image cropping on the normalized multimodal magnetic resonance imaging sequence to obtain a cropped multimodal magnetic resonance imaging sequence; and resample the cropped multimodal magnetic resonance imaging sequence based on a preset normalized resolution to obtain a preprocessed multimodal magnetic resonance imaging sequence.

[0078] The muscle segmentation module 702 is further configured to input the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model, downsample the multimodal magnetic resonance imaging sequence through the encoder in the preset three-dimensional segmentation neural network model to extract multi-scale high-level semantic features, upsample the multi-scale high-level semantic features through the decoder in the preset three-dimensional segmentation neural network model, and fuse the semantic features from the corresponding level of the encoder during the upsampling process to generate a three-dimensional segmentation mask for multiple target muscles.

[0079] The result output module 704 is also used to evaluate the preset three-dimensional classification model using preset model evaluation indicators to obtain evaluation results. The preset model evaluation indicators include accuracy, sensitivity, and F1 score.

[0080] This system embodiment discloses the acquisition of multimodal magnetic resonance imaging (MRI) sequences of the buttock and thigh regions of a target subject; inputting the multimodal MRI sequences into a preset three-dimensional segmentation neural network model to obtain three-dimensional segmentation masks for multiple target muscles; extracting quantitative image feature sets related to muscular dystrophy classification based on the multimodal MRI sequences and the three-dimensional segmentation masks; and inputting the quantitative image feature sets into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result for the target subject. Because this system embodiment obtains three-dimensional segmentation masks for multiple target muscles through a preset three-dimensional segmentation neural network model, then extracts quantitative image feature sets related to muscular dystrophy classification based on the multimodal MRI sequences and the three-dimensional segmentation masks, and finally obtains the muscular dystrophy classification diagnosis result through a preset three-dimensional classification model, compared with existing technologies, this system embodiment improves the efficiency and accuracy of muscular dystrophy classification diagnosis.

[0081] Based on the first embodiment of the medical imaging-based muscular dystrophy classification and diagnosis system of the present invention, a second embodiment of the medical imaging-based muscular dystrophy classification and diagnosis system of the present invention is proposed.

[0082] In this embodiment, the feature extraction module 703 is further configured to: determine the region of interest (ROI) of each target muscle based on the three-dimensional segmentation mask; determine the fat fraction of each ROI based on the fat signal image and water signal image in the multimodal magnetic resonance imaging sequence; extract radiomics features of each ROI based on the three-dimensional segmentation mask and the multimodal magnetic resonance imaging sequence, wherein the radiomics features include at least one of histogram features, texture features, shape features, and wavelet features; and fuse the fat fraction of each target muscle with the radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification.

[0083] The feature extraction module 703 is further configured to standardize the radiomics features to obtain standardized radiomics features; and to perform PCA dimensionality reduction on the standardized radiomics features to obtain dimensionality-reduced radiomics features.

[0084] Other embodiments or specific implementations of the medical imaging-based muscular dystrophy classification and diagnostic system of the present invention can be referred to the above-described device embodiments, and will not be repeated here.

[0085] This application provides a medical image-based muscular dystrophy typing and diagnostic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the steps corresponding to the medical image-based muscular dystrophy typing and diagnostic procedure in the above embodiment 1.

[0086] The following is for reference. Figure 8 This document illustrates a structural schematic diagram of a medical imaging-based muscular dystrophy typing diagnostic device suitable for implementing embodiments of this application. The medical imaging-based muscular dystrophy typing diagnostic device in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 8 The illustrated medical imaging-based muscular dystrophy typing diagnostic device is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0087] like Figure 8As shown, the medical image-based muscular dystrophy typing diagnostic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the medical image-based muscular dystrophy typing diagnostic device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the medical image-based muscular dystrophy typing diagnostic device to wirelessly or wiredly communicate with other devices to exchange data. Although the figures show medical image-based muscular dystrophy typing diagnostic devices with various systems, it should be understood that implementing or possessing all of the systems shown is not required. More or fewer systems may be implemented alternatively.

[0088] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0089] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0090] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0091] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system 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 system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0092] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0093] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0094] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the protection scope of the present invention.

Claims

1. A diagnostic device for muscular dystrophy based on medical imaging, characterized in that, The device includes: a memory, a processor, and a medical image-based muscular dystrophy typing diagnostic program stored in the memory and running on the processor. When the medical image-based muscular dystrophy typing diagnostic program is executed by the processor, it performs the following steps: Acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target subject; The multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles; Based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask, a quantitative image feature set related to muscular dystrophy classification is extracted; The quantitative image feature set is input into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

2. The device as described in claim 1, characterized in that, The step of inputting the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles includes: The multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model. The encoder in the preset three-dimensional segmentation neural network model downsamples the multimodal magnetic resonance imaging sequence to extract multi-scale high-level semantic features. The multi-scale high-level semantic features are upsampled by the decoder in the preset three-dimensional segmentation neural network model, and semantic features from the corresponding level of the encoder are fused during the upsampling process to generate a three-dimensional segmentation mask for multiple target muscles.

3. The device as described in claim 1, characterized in that, The step of extracting a quantitative image feature set related to muscular dystrophy classification based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask includes: The region of interest for each target muscle is determined based on the three-dimensional segmentation mask; Based on the fat signal images and water signal images in the multimodal magnetic resonance imaging sequence, the fat fraction of each region of interest is determined; Based on the three-dimensional segmentation mask and the multimodal magnetic resonance imaging sequence, image omics features of each region of interest are extracted, and the image omics features include at least one of histogram features, texture features, shape features and wavelet features; The fat fraction of each target muscle is fused with the radiomics features to obtain a quantitative image feature set associated with muscular dystrophy classification.

4. The device as described in claim 3, characterized in that, After the step of extracting the radiomics features of each region of interest based on the three-dimensional segmentation mask and the multimodal magnetic resonance imaging sequence, the method further includes: The radiomics features are standardized to obtain standardized radiomics features. The standardized radiomics features are subjected to PCA dimensionality reduction to obtain the dimensionality-reduced radiomics features. Accordingly, the step of fusing the fat fraction of each target muscle with the radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification includes: The fat fraction of each target muscle is fused with the dimensionality-reduced radiomics features to obtain a quantitative image feature set related to muscular dystrophy classification.

5. The device as described in claim 1, characterized in that, Before the step of acquiring multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target object, the method further includes: Acquire a sequence of sample images and an initial 3D segmentation neural network model, and preprocess the sample image sequence to obtain a training dataset; The initial 3D segmentation neural network model is trained based on the training dataset to obtain the training results; The parameters of the initial 3D segmentation neural network model are optimized based on the weighted cross-entropy loss function and the training results to obtain a preset 3D segmentation neural network model.

6. The device as described in claim 1, characterized in that, After the step of inputting the quantitative image feature set into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object, the method further includes: The preset three-dimensional classification model is evaluated using preset model evaluation indicators to obtain evaluation results. The preset model evaluation indicators include accuracy, sensitivity, and F1 score. The formula for calculating the accuracy is: ； The formula for calculating the sensitivity is: ； The formula for calculating the F1 score is: ； in, Indicates accuracy rate. Indicates sensitivity, This represents the F1 score, and Precision represents the accuracy rate. TP indicates a true positive, TN indicates a true negative, FP indicates a false positive, and FN indicates a false negative.

7. The device according to any one of claims 1-6, characterized in that, Following the step of acquiring multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target object, the method further includes: The intensity normalization process is performed on the multimodal magnetic resonance imaging sequence to obtain a normalized multimodal magnetic resonance imaging sequence. The normalized multimodal magnetic resonance imaging sequence is cropped to obtain a cropped multimodal magnetic resonance imaging sequence. The cropped multimodal magnetic resonance imaging sequence is resampled based on a preset standardized resolution to obtain a preprocessed multimodal magnetic resonance imaging sequence. Accordingly, the step of inputting the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles includes: The preprocessed multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles.

8. The device as described in claim 1, characterized in that, The multimodal magnetic resonance imaging sequence is a T2-weighted Dixon sequence, which includes images of four modalities: in-phase, out-of-phase, aqueous phase, and fatty phase; the preset three-dimensional classification model is a pre-trained Xception3D classifier.

9. A medical imaging-based classification and diagnostic system for muscular dystrophy, characterized in that, The system includes: The image acquisition module is used to acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target object; The muscle segmentation module is used to input the multimodal magnetic resonance imaging sequence into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles. The feature extraction module is used to extract a quantitative image feature set related to muscular dystrophy classification based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask. The result output module is used to input the quantitative image feature set into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

10. A storage medium, characterized in that, The storage medium stores a medical image-based muscular dystrophy classification diagnostic program, which, when executed by a processor, performs the following steps: Acquire multimodal magnetic resonance imaging sequences of the buttock and thigh regions of the target subject; The multimodal magnetic resonance imaging sequence is input into a preset three-dimensional segmentation neural network model to obtain a three-dimensional segmentation mask for multiple target muscles; Based on the multimodal magnetic resonance imaging sequence and the three-dimensional segmentation mask, a quantitative image feature set related to muscular dystrophy classification is extracted; The quantitative image feature set is input into a preset three-dimensional classification model to obtain the muscular dystrophy classification diagnosis result of the target object.

Images