Pelvic organ prolapse staging method, device, apparatus, medium and program product

Abstract

The invention provides a pelvic organ prolapse staging method, device and equipment, a medium and a program product, and relates to the technical field of medical image processing. The method comprises the following steps: acquiring a first pelvic magnetic resonance image to be integrally staging; inputting the first pelvic magnetic resonance imaging into an image segmentation model to obtain an image segmentation result output by the image segmentation model; the image segmentation result comprises segmentation mask information of a target pelvic floor area and / or the area of the target pelvic floor area, the target pelvic floor area is a cross-sectional area below a coccygeus PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor area comprises a tailbone levator plate and a perineum bottom edge; and inputting an image segmentation result into the integral pelvic organ prolapse staging model to obtain an integral pelvic organ prolapse staging result output by the integral pelvic organ prolapse staging model. According to the method, the stable relevance between the target pelvic floor area and the integral pelvic organ prolapse staging result is considered, so that the accuracy of integral pelvic organ prolapse staging is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a pelvic organ prolapse staging method, device, equipment, medium and program product. Background Art

[0002] Pelvic Organ Prolapse (POP) refers to the abnormal descent of pelvic organs due to lack of effective support, including bladder prolapse, uterine or vaginal prolapse, rectal prolapse, etc., which is a pelvic dysfunction disease. Among them, quantitative description of the severity of POP in clinical practice is crucial for accurately assessing the severity of the disease, guiding the selection of treatment plans, and evaluating the treatment effect. Therefore, it is necessary to quantitatively describe the severity of pelvic organ prolapse.

[0003] The traditional method is to assess the severity of pelvic organ prolapse through physical examination. However, doctors are required to manually measure multiple reference systems, such as the sacrococcygeal subpubic line, the pubic midline or other anatomical landmarks, which makes the assessment of pelvic organ prolapse time-consuming, that is, it leads to low efficiency in the assessment of pelvic organ prolapse, and requires doctors to have professional knowledge of pelvic floor anatomy. In addition, different clinical experiences and professional backgrounds among different doctors lead to a lack of objectivity and consistency in the diagnostic results, that is, it leads to reduced accuracy in the assessment of pelvic organ prolapse.

[0004] At present, in order to improve the efficiency and accuracy of pelvic organ prolapse assessment, clinicians often use imaging technology as an auxiliary means of assessment. Specifically, pelvic floor MRI (Magnetic Resonance Imaging) is obtained, and then the PCL (Pubococcygeal line) is found on the pelvic floor MRI, and the PCL is used as a reference line for pelvic organ prolapse assessment; however, the existing assessment method does not consider the influence of the interaction of the overall pelvic floor structure, resulting in low accuracy in pelvic organ prolapse assessment, which may lead to misjudgment. Summary of the invention

[0005] The present invention provides a pelvic organ prolapse staging method, device, equipment, medium and program product, which are used to solve the defect of low accuracy in the evaluation of pelvic organ prolapse in the prior art and achieve high-accuracy pelvic organ prolapse staging.

[0006] The present invention provides a pelvic organ prolapse staging method, comprising: Obtain the first pelvic floor MRI to be staged overall; Inputting the first pelvic floor magnetic resonance imaging into an image segmentation model, obtaining an image segmentation result output by the image segmentation model; the image segmentation result includes segmentation mask information of a target pelvic floor region and / or an area of ​​a target pelvic floor region, the target pelvic floor region is a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor region includes the coccygeal levator plate and the edge of the perineum; Inputting the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model; Wherein, the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to evaluate the severity of overall pelvic organ prolapse.

[0007] According to a pelvic organ prolapse staging method provided by the present invention, the image segmentation result is input into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model, including: Inputting the first pelvic floor magnetic resonance imaging and the image segmentation result into an input layer of the overall pelvic organ prolapse staging model to obtain a multimodal fusion feature output by the input layer; The multimodal fusion features are input into the overall pelvic organ prolapse staging layer in the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging layer.

[0008] According to a pelvic organ prolapse staging method provided by the present invention, the obtaining of a first pelvic floor magnetic resonance imaging to be staged as a whole comprises: Obtain native pelvic floor magnetic resonance imaging; The original pelvic floor magnetic resonance imaging is cropped to obtain the first pelvic floor magnetic resonance imaging; the first pelvic floor magnetic resonance imaging includes a plurality of preset organ regions and / or a plurality of preset body regions.

[0009] According to a pelvic organ prolapse staging method provided by the present invention, the cropping of the original pelvic floor magnetic resonance imaging to obtain the first pelvic floor magnetic resonance imaging comprises: Inputting the original pelvic floor magnetic resonance imaging into a key point detection model to obtain a pubic point detection result and a sacrococcygeal point detection result output by the key point detection model; the key point detection model is used to detect the pubic point and sacrococcygeal point in the original pelvic floor magnetic resonance imaging; Determine the sacrococcygeal subpubic point SCIPP line based on the pubic point detection result and the sacrococcygeal point detection result; Performing a rotation transformation on the original pelvic floor magnetic resonance imaging based on the SCIPP line to obtain a transformed original pelvic floor magnetic resonance imaging, so that the angle between the SCIPP line and the horizontal line of the transformed original pelvic floor magnetic resonance imaging is a preset fixed angle; Based on the pubic point detection result and the sacrococcygeal point detection result, on the transformed original pelvic floor magnetic resonance imaging, the edges are expanded outwards with the pubic point and the sacrococcygeal point as starting points, respectively, to obtain a rectangular area; The first pelvic floor magnetic resonance imaging is obtained by cropping the transformed original pelvic floor magnetic resonance imaging based on the rectangular area.

[0010] According to a pelvic organ prolapse staging method provided by the present invention, the pelvic organ prolapse staging method further comprises: Obtain a second pelvic floor MRI to be used for pelvic staging; Inputting the second pelvic floor magnetic resonance imaging into a key point detection model, obtaining a pubic symphysis lower edge detection result, a distal coccygeal joint detection result, a first key point detection result, a second key point detection result, and a third key point detection result output by the key point detection model; the key point detection model is used to detect the pubic symphysis lower edge, the distal coccygeal joint, the first key point, the second key point, and the third key point in the second pelvic floor magnetic resonance imaging; Determining a target PCL in the second pelvic floor magnetic resonance imaging based on the detection result of the lower edge of the pubic symphysis and the detection result of the distal coccygeal joint; Determine, based on the first key point detection result, the second key point detection result, and the third key point detection result, a first vertical distance between the first key point and the target PCL, a second vertical distance between the second key point and the target PCL, and a third vertical distance between the third key point and the target PCL; Determine an anterior pelvic prolapse staging result based on the first vertical distance, determine a middle pelvic prolapse staging result based on the second vertical distance, and determine a posterior pelvic prolapse staging result based on the third vertical distance; Among them, the whole pelvic cavity is divided into anterior pelvic cavity, middle pelvic cavity and posterior pelvic cavity, the first key point is the prolapse reference point of the anterior pelvic cavity, the second key point is the prolapse reference point of the middle pelvic cavity, and the third key point is the prolapse reference point of the posterior pelvic cavity.

[0011] According to a pelvic organ prolapse staging method provided by the present invention, the second pelvic floor magnetic resonance imaging is input into a key point detection model to obtain a pubic symphysis lower edge detection result, a distal coccygeal joint detection result, a first key point detection result, a second key point detection result, and a third key point detection result output by the key point detection model, including: Inputting the second pelvic floor magnetic resonance imaging into a key point detection model, obtaining an initial pubic symphysis lower edge detection result, an initial distal coccygeal joint detection result, an initial first key point detection result, an initial second key point detection result, and an initial third key point detection result output by the key point detection model; The initial pubic symphysis lower edge detection result is input into the first key point fine-tuning model corresponding to the pubic symphysis lower edge, and the pubic symphysis lower edge detection result output by the first key point fine-tuning model is obtained; the initial distal coccyx joint detection result is input into the second key point fine-tuning model corresponding to the distal coccyx joint, and the distal coccyx joint detection result output by the second key point fine-tuning model is obtained; the initial first key point detection result is input into the third key point fine-tuning model corresponding to the first key point, and the first key point detection result output by the third key point fine-tuning model is obtained; the initial second key point detection result is input into the fourth key point fine-tuning model corresponding to the second key point, and the second key point detection result output by the fourth key point fine-tuning model is obtained; the initial third key point detection result is input into the fifth key point fine-tuning model corresponding to the third key point, and the third key point detection result output by the fifth key point fine-tuning model is obtained.

[0012] According to a pelvic organ prolapse staging method provided by the present invention, the step of obtaining a second pelvic floor magnetic resonance imaging to be used for pelvic staging includes: Obtain native pelvic floor magnetic resonance imaging; Inputting the original pelvic floor magnetic resonance imaging into a target detection model to obtain the second pelvic floor magnetic resonance imaging output by the target detection model; Among them, the second pelvic floor magnetic resonance imaging is a rectangular area surrounding the lower edge of the pubic symphysis, the distal coccyx joint, the first key point, the second key point and the third key point; the size of the second pelvic floor magnetic resonance imaging is a preset fixed value.

[0013] The present invention also provides a pelvic organ prolapse staging device, comprising: An imaging acquisition module, used to acquire a first pelvic floor magnetic resonance imaging to be staged overall; an image segmentation module, configured to input the first pelvic floor magnetic resonance imaging into an image segmentation model, and obtain an image segmentation result output by the image segmentation model; the image segmentation result includes segmentation mask information of a target pelvic floor region and / or an area of ​​a target pelvic floor region, the target pelvic floor region being a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor region including the coccygeal levator plate and the edge of the perineum; A prolapse staging module, used for inputting the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model; Wherein, the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to evaluate the severity of overall pelvic organ prolapse.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned pelvic organ prolapse staging methods is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for staging pelvic organ prolapse as described in any one of the above is implemented.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned pelvic organ prolapse staging methods.

[0017] The pelvic organ prolapse staging method, device, equipment, medium and program product provided by the present invention obtain a first pelvic floor magnetic resonance image to be staged as a whole, input the first pelvic floor magnetic resonance image into an image segmentation model, and obtain an image segmentation result output by the image segmentation model, and the image segmentation result includes segmentation mask information of the target pelvic floor area and / or the area of ​​the target pelvic floor area, and the target pelvic floor area is the cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance image, and the boundary of the target pelvic floor area includes the coccygeal levator plate and the perineal floor edge, based on this, the target pelvic floor area covers all pelvic organs, muscles and ligaments. belt, thereby enabling a deeper understanding of the pelvic anatomical structure, and further providing a way to comprehensively evaluate the overall prolapse of the pelvic organs, and ultimately improving the accuracy of the overall pelvic organ prolapse staging; at the same time, the image segmentation result is input into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model, and the image segmentation result includes the segmentation mask information of the target pelvic floor area and / or the area of ​​the target pelvic floor area, taking into account the stable correlation between the target pelvic floor area and the overall pelvic organ prolapse staging result, thereby improving the accuracy of the overall pelvic organ prolapse staging. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is one of the flow charts of the pelvic organ prolapse staging method provided by the present invention.

[0020] Figure 2 Schematic diagram of the target pelvic floor area provided by the present invention.

[0021] Figure 3 This is the second schematic flow chart of the pelvic organ prolapse staging method provided by the present invention.

[0022] Figure 4 It is a cropped schematic diagram of pelvic floor magnetic resonance imaging provided by the present invention.

[0023] Figure 5 This is the third flow chart of the pelvic organ prolapse staging method provided by the present invention.

[0024] Figure 6 It is a schematic diagram of key point detection provided by the present invention.

[0025] Figure 7 It is a schematic diagram of the process of the cascade convolutional neural network provided by the present invention.

[0026] Figure 8 It is a schematic structural diagram of the pelvic organ prolapse staging device provided by the present invention.

[0027] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] At present, in order to improve the efficiency and accuracy of pelvic organ prolapse assessment, clinicians often use imaging technology as an auxiliary means of assessment. For example, the existing MRI measurement methods such as PCL measurement are used to assess pelvic organ prolapse; however, the existing technology only focuses on a few specific key points and does not consider the influence of the interaction of the overall pelvic floor structure, resulting in low accuracy in pelvic organ prolapse assessment, which may lead to misjudgment.

[0030] In view of the above problems, the present invention proposes the following embodiments. Figure 1-Figure 7 The present invention is described in terms of the staging method for pelvic organ prolapse.

[0031] Figure 1 FIG. 1 is one of the flow charts of the pelvic organ prolapse staging method provided by the present invention, such as Figure 1 As shown, the pelvic organ prolapse staging method includes the following steps 110 , 120 and 130 .

[0032] Step 110: Acquire a first pelvic floor magnetic resonance imaging image to be staged as a whole.

[0033] Here, the first pelvic floor Magnetic Resonance Imaging (MRI) is used to determine the overall pelvic organ prolapse staging result, that is, it is a pelvic floor MRI of a person to be staging the overall pelvic organ prolapse.

[0034] In one embodiment, the first pelvic floor magnetic resonance imaging is a mid-sagittal MRI, that is, the mid-sagittal MRI is the key layer for diagnosis.

[0035] Furthermore, all DICOM (Digital Imaging and Communications in Medicine) images were converted into PNG (Portable Network Graphics) format, and the grayscale values ​​were replicated on each channel to convert the single-channel DICOM images into three-channel RGB images.

[0036] It should be understood that magnetic resonance imaging (MRI) has the advantages of high resolution, large field of view, clear soft tissue definition and non-invasiveness, and the probe does not touch the perineum or vagina. It plays an important role in the screening, monitoring, diagnosis, staging and treatment response evaluation of POP.

[0037] Furthermore, the first pelvic floor MRI is a dynamic MRI. Based on this, the characteristics of POP can be clearly observed directly on the dynamic MRI under the Valsalva maneuver, so that dynamic MRI can not only be used for the diagnosis of overall pelvic organ prolapse, but also provide excellent visualization of pelvic organs and structures.

[0038] Step 120: input the first pelvic floor magnetic resonance imaging into an image segmentation model to obtain an image segmentation result output by the image segmentation model.

[0039] The image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area. The target pelvic floor area is a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor area includes the coccygeal levator plate and the perineal floor edge.

[0040] Specifically, the target pelvic floor area is the area under the pubococcygeal line (AUPCL), that is, the area under the pubococcygeal line. Further, the target pelvic floor area is the area occupied by the pelvic floor organs, defined as the cross-sectional area below the PCL (Pubococcygeal Line) under the midsagittal image obtained by the maximum Valsalva maneuver, with the boundaries including the coccygeal levator plate and the edge of the perineum, and tracing from the lower edge of the pubic symphysis backward along the shallow edge of these structures to the last coccygeal joint.

[0041] It should be understood that the targeted pelvic floor regions provide a comprehensive way to assess the entire pelvic organ prolapse, which encompasses all pelvic organs, muscles, and ligaments, allowing for a deeper understanding of the pelvic anatomy and, in turn, improving the accuracy of the overall pelvic organ prolapse staging.

[0042] In one embodiment, the image segmentation model is trained based on sample pelvic floor magnetic resonance imaging and its corresponding image segmentation result labels.

[0043] For example, a sample pelvic floor magnetic resonance imaging dataset is established based on a large number of subjects, and the size of each image is (Unit: mm), and the midsagittal MRI of each subject at rest (Rest) and under Valsalva maneuver was selected as the sample pelvic floor MRI. Among them, all DICOM images were converted to PNG format, and the grayscale values ​​were copied on each channel to convert the single-channel DICOM image into a three-channel RGB image. In order to prevent data leakage, the sample pelvic floor MRI dataset was divided into training set, validation set and test set in a ratio of 6:2:2 based on the case as the division unit, that is, to ensure that the midsagittal MRI of the same case at rest and under Valsalva maneuver is in the same dataset.

[0044] For example, 3D Slicer software is used to annotate the segmentation mask and perform one-hot encoding to facilitate subsequent training of the image segmentation model. Figure 2 The black area in the middle pelvic floor is the marked target pelvic floor area.

[0045] Furthermore, the number of sample pelvic floor MRI images is usually small, based on which data augmentation is performed on the sample pelvic floor MRI images, thereby increasing the diversity of training data and the generalization ability of the model. The data augmentation method may include but is not limited to: random resizing of images, random cropping, and random horizontal flipping.

[0046] In one embodiment, the image segmentation model is a Mask2Former segmentation model with multi-dimensional collaborative attention.

[0047] Exemplarily, the image segmentation model includes a backbone feature extraction network, a pixel decoder, a Transformer decoder, and a semantic reasoning module connected in sequence. The backbone feature extraction network is used to extract low-resolution semantic features from the image; the pixel decoder is used to generate multi-scale deformable features and output pixel-by-pixel embeddings for predicting semantic category probabilities; the Transformer decoder is used to receive multi-scale deformable features from the pixel decoder and output a set of semantic region queries; the semantic reasoning module is used to integrate semantic category predictions and mask predictions to generate the final semantic segmentation results.

[0048] For example, the backbone feature extraction network uses a deep residual network ResNet-101, which can capture complex semantic information in images. At the same time, its residual connection helps alleviate the gradient vanishing and gradient exploding problems. Furthermore, given that the number of sample pelvic floor MRIs is relatively small, the use of transfer learning methods and models pre-trained on other datasets not only saves training time, but also reduces the risk of overfitting and improves the generalization ability of the model, thereby improving the segmentation accuracy of the image segmentation model.

[0049] Furthermore, considering that the Valsalva maneuver will cause obvious displacement and morphological changes of the pelvic floor organs, pelvic floor MRI is more likely to produce motion artifacts than the resting state, which reduces the image clarity and the recognition of the anatomical structure, which may lead to increased measurement errors. Based on this, in order to better segment the target pelvic floor area and extract its contour, we improved the backbone feature extraction network based on ResNet-101, introduced a novel multi-dimensional collaborative attention mechanism, and directly embedded the multi-dimensional collaborative attention module after the third 1×1 convolutional layer in the bottleneck residual block of ResNet-101, namely MCA-ResNet101.

[0050] Among them, the Multidimensional Collaborative Attention (MCA) module consists of three parallel branches, which are responsible for capturing the feature dependencies between the spatial dimensions (H, W) and the channel dimension (C). Each branch performs feature integration on the input feature map. The spatial branch (W, top branch) first rotates the feature map 90 degrees counterclockwise along the H axis, and then captures the attention dependencies along the spatial W dimension through squeezing and excitation transformation, followed by sigmoid activation to obtain the attention weights of the W dimension; finally, the attention weights are combined with the original features to obtain the enhanced features. Similarly, the features enhanced by attention on another spatial dimension (H, middle branch) can be obtained. The channel branch learns the attention between channels through identity mapping, squeezing and excitation transformation, and combines it with the original features to output the enhanced features. Finally, the features from the three branches are averaged to obtain the final optimized feature map.

[0051] For example, three feature maps of different scales (1 / 32, 1 / 16, and 1 / 8 of the original image size, respectively) extracted from the backbone feature extraction network are converted into feature maps with 256 channels through a 1×1 convolution operation. Subsequently, the multi-scale deformable attention module in the pixel decoder generates multi-scale deformable features with unchanged channel number and resolution. Finally, the 1 / 8 scale deformable features are upsampled by 2 times and added to the 1 / 4 scale features in the backbone feature extraction network to obtain a pixel-by-pixel embedding.

[0052] Among them, the Transformer decoder of the embodiment of the present invention is different from the standard decoder. The cross attention part adopts mask attention, which only focuses on the semantically relevant feature area predicted by the query, rather than the entire feature map, which accelerates the convergence of the Transformer-based model. In addition, exchanging the order of self-attention and mask attention further improves the computational efficiency. There are three decoders in total, each of which has three layers, and each layer uses one of the three multi-scale deformable features from the pixel decoder in turn, from the lowest resolution to the highest resolution for processing. Sinusoidal position embedding is added to the deformable features of each scale, and learnable scale-level embedding is used to distinguish different feature levels. The predicted semantically relevant mask is input into the Transformer decoder together with the image features for mask attention calculation. Finally, the Transformer decoder divides the image into N regions and generates N semantic region query outputs Q in parallel. The semantic region query output Q independently generates semantic category predictions and semantic feature embeddings. The semantic category prediction is obtained by a linear classifier and a softmax function. After Q is converted into semantic feature embedding through two layers of MLP (Multi-Layer Perceptron), the semantic feature embedding is dot-producted with the pixel-by-pixel embedding, and sigmoid activation is used to obtain the semantic segmentation prediction result.

[0053] Exemplarily, during the training of the image segmentation model, 40,000 training cycles are performed. The optimizer uses stochastic gradient descent, with a learning rate of 0.01, a momentum of 0.9, and a weight decay of 0.0005. At the same time, the learning rate scheduler is a polynomial decay, which decays the learning rate from the initial value to the minimum value of 0.0001, and the decay process follows an exponent of 0.9. Furthermore, evaluation indicators such as Dice similarity coefficient, precision, recall rate, and F1 value (F1 Score) can be used to evaluate the performance of the image segmentation model.

[0054] The image segmentation result includes segmentation mask information of the target pelvic floor region and / or the area of ​​the target pelvic floor region.

[0055] In a specific embodiment, the image segmentation model outputs segmentation mask information of the target pelvic floor region, and the pixel value can be automatically calculated based on the segmentation mask information to obtain the area of ​​the target pelvic floor region.

[0056] Step 130: input the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model.

[0057] The overall pelvic organ prolapse staging model is used to assess the severity of overall pelvic organ prolapse. In one embodiment, the overall pelvic organ prolapse staging model is trained based on the sample image segmentation results and the corresponding overall pelvic organ prolapse staging result labels.

[0058] In one embodiment, the overall pelvic organ prolapse staging results are divided into multiple different stages. For example, pelvic organ prolapse staging is performed based on the POP-Q (Pelvic Organ Prolapse Quantification) staging system, that is, the overall pelvic organ prolapse staging results are divided into five stages 0-IV, that is, POP is divided into five stages 0-IV. Among them, in the POP-Q staging system, the staging is determined according to the most severe prolapse site such as the anterior and posterior vaginal wall, cervix or fornix. The POP-Q staging system is divided into stages 0-IV, where stage 0 indicates no prolapse, stage I indicates that the distal end of the prolapse is on the inner side of the hymen edge, and the distal end of the prolapse is <1 cm away from the hymen edge; stage II indicates that the distal end of the prolapse is on the inner or outer side of the hymen edge, and the distal end of the prolapse is within 1 cm of the hymen edge; stage III indicates that the distal end of the prolapse is on the outer side of the hymen edge, and the distal end of the prolapse is >1 cm away from the hymen edge, but the distal end of the prolapse is < (total vaginal length - 2) cm; stage IV indicates that the entire vagina or organ is completely everted and completely prolapsed, and the distal end of the prolapse exceeds the hymen edge by a distance > (total vaginal length - 2) cm.

[0059] It should be understood that determining the overall pelvic organ prolapse staging results is a comprehensive assessment of the overall situation of pelvic floor prolapse, which helps to quickly understand the approximate severity of the patient's pelvic floor prolapse, and has certain guiding significance for the initial diagnosis of the condition and the selection of the general direction of the treatment plan (such as conservative treatment or surgical treatment). For example, for POP-Q staging, mild prolapse (stage I-II) may be more inclined to try conservative treatment first, such as pelvic floor muscle training, while severe prolapse (stage III-IV) may need to consider surgical intervention.

[0060] In one embodiment, when the image segmentation result includes the area of ​​the target pelvic floor region, the area of ​​the target pelvic floor region is input into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model.

[0061] Furthermore, since the sparse feature vector of the target pelvic floor area is difficult for the neural network model to learn, the overall pelvic organ prolapse staging model in the embodiment of the present invention also applies an Autoencoder, which can automatically learn the potential feature representation of the data, extract key features in the low-dimensional space through the encoding and decoding process, achieve dimensionality reduction and denoising, and effectively deal with the redundancy and noise problems that may exist in numerical features. It can also generate synthetic data when the data is unbalanced to balance the data set, enhance the model's recognition ability for minority classes, and improve the generalization ability when facing new data.

[0062] Furthermore, since the characteristic vector of the target pelvic floor area is a continuous variable, it needs to be normalized or standardized to facilitate model learning. There are many parameters to adjust when machine learning modeling, such as the number of hidden layers, the number of neural network units, the optimization method, the learning rate, etc. The grid search method is used to search for the optimal parameters to achieve the best result.

[0063] In another embodiment, when the image segmentation result includes segmentation mask information of the target pelvic floor area, the segmentation mask information of the target pelvic floor area is input into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model.

[0064] In another embodiment, the first pelvic floor magnetic resonance imaging and image segmentation results are input into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging results output by the overall pelvic organ prolapse staging model.

[0065] It should be noted that the correlation analysis shows that the correlation coefficient between the target pelvic floor area and POP-Q is significantly better than the correlation coefficient between the PCL level (i.e., the method using PCL as the reference line) and POP-Q. In terms of reliability assessment, the prolapse staging method of the embodiment of the present invention has good intra-assessor reliability and inter-assessor reliability. Through ordered logistic regression analysis, it is clear that the target pelvic floor area is significantly correlated with the POP-Q stage, and the correlation of the target pelvic floor area is more stable and reliable. Therefore, the embodiment of the present invention, as a novel and comprehensive pelvic organ prolapse staging method, is closely related to the clinical POP-Q stage, has good evaluation reliability, and shows great potential in improving the accuracy of POP diagnosis, optimizing treatment strategies, and improving patient prognosis, thereby achieving high-accuracy pelvic organ prolapse staging.

[0066] The pelvic organ prolapse staging method provided by the embodiment of the present invention obtains the first pelvic floor magnetic resonance imaging to be staged as a whole, inputs the first pelvic floor magnetic resonance imaging into the image segmentation model, obtains the image segmentation result output by the image segmentation model, and the image segmentation result includes the segmentation mask information of the target pelvic floor area and / or the area of ​​the target pelvic floor area, and the target pelvic floor area is the cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor area includes the coccygeal levator plate and the perineal floor edge, based on which, the target pelvic floor area covers all pelvic organs, muscles and ligaments, so that It can provide a deeper understanding of the pelvic anatomical structure, thereby providing a way to comprehensively evaluate the overall pelvic organ prolapse, and ultimately improve the accuracy of the overall pelvic organ prolapse staging; at the same time, the image segmentation result is input into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model, and the image segmentation result includes the segmentation mask information of the target pelvic floor area and / or the area of ​​the target pelvic floor area, taking into account the stable correlation between the target pelvic floor area and the overall pelvic organ prolapse staging result, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0067] Based on any of the above embodiments, in the method, the above step 130 includes step 131 and step 132.

[0068] Step 131: input the first pelvic floor magnetic resonance imaging and the image segmentation result into an input layer of the overall pelvic organ prolapse staging model to obtain a multimodal fusion feature output by the input layer.

[0069] Here, multimodal data is first input into the input layer of the overall pelvic organ prolapse staging model, so that the model can learn comprehensive feature representation from the beginning, thereby improving the staging accuracy of the overall pelvic organ prolapse staging model. That is, the strategy of early feature fusion is adopted to integrate the features of different modalities at the data input level.

[0070] It should be noted that, considering that the first pelvic floor magnetic resonance imaging can present detailed anatomical information of the pelvic region in an all-round and high-resolution manner, from the grayscale differences of different tissues to the spatial position relationship between organs, the first pelvic floor magnetic resonance imaging is also input into the overall pelvic organ prolapse staging model, thereby improving the accuracy of the overall pelvic organ prolapse staging. And the segmentation mask information of the target pelvic floor area and / or the area of ​​the target pelvic floor area are used as quantitative indicators of the pelvic area area to provide information for POP staging from the area dimension, thereby utilizing the complementarity between images and numerical data of different modalities to improve the accuracy of the overall pelvic organ prolapse staging.

[0071] In one embodiment, the segmentation mask information of the first pelvic floor magnetic resonance imaging and the target pelvic floor area is input into the input layer of the overall pelvic organ prolapse staging model to obtain the multimodal fusion features output by the input layer. The segmentation mask information is usually used to highlight or extract specific areas in the image. Adding the segmentation mask information on the basis of the first pelvic floor magnetic resonance imaging can help focus on the key areas related to POP staging and exclude other interference information, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0072] In another embodiment, the first pelvic floor magnetic resonance imaging and the area of ​​the target pelvic floor region are input into the input layer of the overall pelvic organ prolapse staging model to obtain a multimodal fusion feature output by the input layer. By combining the first pelvic floor magnetic resonance imaging and the area of ​​the target pelvic floor region, the accuracy of the overall pelvic organ prolapse staging is improved by fusing the two information sources, using the comprehensive information of the first pelvic floor magnetic resonance imaging and the specific indicators of the area of ​​the target pelvic floor region.

[0073] In another embodiment, the first pelvic floor magnetic resonance imaging, the segmentation mask information of the target pelvic floor region, and the area of ​​the target pelvic floor region are input into the input layer of the overall pelvic organ prolapse staging model to obtain the multimodal fusion features output by the input layer. By comprehensively utilizing the first pelvic floor magnetic resonance imaging, the segmentation mask information of the target pelvic floor region, and the area of ​​the target pelvic floor region, POP can be staged in a comprehensive manner from multiple dimensions to the maximum extent, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0074] Step 132: input the multimodal fusion features into the overall pelvic organ prolapse staging layer in the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging layer.

[0075] It should be understood that using multimodal data together as input to the overall pelvic organ prolapse staging model can improve the accuracy of the overall pelvic organ prolapse staging compared to the model based on a single data source.

[0076] The pelvic organ prolapse staging method provided by the embodiment of the present invention inputs the first pelvic floor magnetic resonance imaging and image segmentation results into the input layer of the overall pelvic organ prolapse staging model to obtain the multimodal fusion features output by the input layer, so that the model can learn the comprehensive feature representation from the beginning, thereby improving the staging accuracy of the overall pelvic organ prolapse staging model; inputs the multimodal fusion features into the overall pelvic organ prolapse staging layer in the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging results output by the overall pelvic organ prolapse staging layer, thereby using the multimodal data together as the input of the overall pelvic organ prolapse staging model, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0077] Based on any of the above embodiments, in the method, the above step 110 includes: step 111 and step 112.

[0078] Step 111, obtaining original pelvic floor magnetic resonance imaging.

[0079] Here, the original pelvic floor magnetic resonance imaging may be the acquired pelvic floor magnetic resonance imaging, and thus preprocessing the original pelvic floor magnetic resonance imaging may improve the accuracy of the overall pelvic organ prolapse staging.

[0080] Step 112: cropping the original pelvic floor magnetic resonance imaging to obtain the first pelvic floor magnetic resonance imaging.

[0081] Wherein, the first pelvic floor magnetic resonance imaging includes a plurality of preset organ regions and / or a plurality of preset body regions.

[0082] For example, the preset organ regions include the bladder region, the rectum region and the uterus region; the preset site regions include the vagina region, that is, the region of interest including the three key organs (bladder, rectum and uterus) and the vagina is cropped in the original pelvic floor magnetic resonance imaging to remove irrelevant information.

[0083] In one embodiment, the original pelvic floor magnetic resonance imaging is input into the target detection model to obtain the first pelvic floor magnetic resonance imaging output by the target detection model.

[0084] It should be understood that the original pelvic floor MRI is cropped to obtain the first pelvic floor MRI actually used to determine the overall pelvic organ prolapse staging result, so as to remove irrelevant information, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0085] The pelvic organ prolapse staging method provided in the embodiment of the present invention crops the original pelvic floor magnetic resonance imaging to obtain the first pelvic floor magnetic resonance imaging, thereby removing irrelevant information and further improving the accuracy of the overall pelvic organ prolapse staging.

[0086] Based on any of the above embodiments, Figure 3 FIG. 2 is a flow chart of the pelvic organ prolapse staging method provided by the present invention. Figure 3 As shown, the above step 112 includes: step 1121 to step 1225.

[0087] Step 1121, inputting the original pelvic floor magnetic resonance imaging into the key point detection model, and obtaining the pubic point detection result and the sacrococcygeal point detection result output by the key point detection model.

[0088] The key point detection model is used to detect the pubic point and the sacrococcygeal point in the original pelvic floor magnetic resonance imaging. In one embodiment, the key point detection model is trained based on the sample original pelvic floor magnetic resonance imaging and its corresponding pubic point detection result label and sacrococcygeal point detection result label.

[0089] Further, the original pelvic floor magnetic resonance imaging is input into the key point detection model to obtain the initial pubic point detection result and the initial sacrococcygeal point detection result output by the key point detection model; the initial pubic point detection result is input into the first key point fine-tuning model corresponding to the pubic point to obtain the pubic point detection result output by the first key point fine-tuning model; the initial sacrococcygeal point detection result is input into the second key point fine-tuning model corresponding to the sacrococcygeal point to obtain the sacrococcygeal point detection result output by the second key point fine-tuning model. This embodiment can be specifically referred to the following embodiment, and will not be described in detail here.

[0090] Furthermore, the original pelvic floor magnetic resonance imaging is input into the target detection model to obtain the third pelvic floor magnetic resonance imaging output by the target detection model; the third pelvic floor magnetic resonance imaging is input into the key point detection model to obtain the pubic point detection result and the sacrococcygeal point detection result output by the key point detection model. The third pelvic floor magnetic resonance imaging is a rectangular area surrounding the pubic point and the sacrococcygeal point; the size of the third pelvic floor magnetic resonance imaging is a preset fixed value. This embodiment can be specifically referred to the following embodiment, and will not be described in detail here.

[0091] Step 1122, based on the pubic point detection result and the sacrococcygeal point detection result, determine the sacrococcygeal subpubic point SCIPP line.

[0092] Specifically, the two ends of the SCIPP (Sacrococcygeal Inferior Pubic Point) line are the pubic point and the sacrococcygeal point respectively.

[0093] Step 1123, performing a rotation transformation on the original pelvic floor MRI based on the SCIPP line to obtain a transformed original pelvic floor MRI, so that the angle between the SCIPP line and the horizontal line of the transformed original pelvic floor MRI is a preset fixed angle.

[0094] Here, the horizontal line is a straight line in the horizontal direction, that is, a line parallel to the X-axis. The preset fixed angle can be set according to actual needs, for example, 34 degrees.

[0095] Exemplarily, the original pelvic floor magnetic resonance imaging is subjected to a rotation transformation so that the angle between the SCIPP line and the horizontal line is 34 degrees.

[0096] It should be understood that the purpose of rotational transformation of the original pelvic floor MRI based on the SCIPP line is to eliminate the differences caused by different pelvic inclinations during scanning, that is, to align the original pelvic floor MRI, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0097] Step 1124, based on the pubic point detection result and the sacrococcygeal point detection result, the edges are expanded outwards with the pubic point and the sacrococcygeal point as starting points on the transformed original pelvic floor magnetic resonance imaging to obtain a rectangular area.

[0098] Specifically, the pubic point is used as the starting point, the first preset pixel is expanded horizontally, and the second preset pixel is expanded vertically; the sacrococcygeal point is used as the starting point, the third preset pixel is expanded horizontally, and the fourth preset pixel is expanded vertically; the maximum expanded edge is determined as a rectangular area, that is, the finally determined rectangular area includes the edge expanded outward from the pubic point and the edge expanded outward from the sacrococcygeal point. Exemplarily, the horizontal and vertical margins from the pubic point are expanded by 74.92 pixels and 75.33 pixels respectively, and the horizontal and vertical margins from the sacrococcygeal point are expanded by 59.19 pixels and 5.86 pixels respectively.

[0099] Step 1125: cropping the transformed original pelvic floor magnetic resonance imaging based on the rectangular region to obtain the first pelvic floor magnetic resonance imaging.

[0100] Specifically, the first pelvic floor magnetic resonance imaging is the rectangular area.

[0101] For ease of understanding, Figure 4 As shown, Figure 4 (a) is the original pelvic floor magnetic resonance imaging. Figure 4 (b) is the first pelvic floor magnetic resonance imaging cropped based on the rectangular area. Figure 4 Middle (c) is the first pelvic floor MRI cropped based on the target detection model.

[0102] The pelvic organ prolapse staging method provided by the embodiment of the present invention performs a rotation transformation on the original pelvic floor magnetic resonance imaging based on the SCIPP line to obtain the transformed original pelvic floor magnetic resonance imaging, so that the angle between the SCIPP line and the horizontal line of the transformed original pelvic floor magnetic resonance imaging is a preset fixed angle, thereby eliminating the difference caused by different pelvic inclinations during scanning, that is, aligning the original pelvic floor magnetic resonance imaging, thereby improving the cropping accuracy, and further improving the accuracy of the overall pelvic organ prolapse staging; at the same time, based on the pubic point detection result and the sacrococcygeal point detection result, the edges are expanded outward with the pubic point and the sacrococcygeal point as the starting points on the transformed original pelvic floor magnetic resonance imaging to obtain a rectangular area, and the transformed original pelvic floor magnetic resonance imaging is cropped based on the rectangular area to obtain a first pelvic floor magnetic resonance imaging including a plurality of preset organ areas and / or a plurality of preset part areas, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0103] Based on any of the above embodiments, Figure 5 FIG. 3 is a flow chart of the pelvic organ prolapse staging method provided by the present invention. Figure 5 As shown, the method further includes steps 510 to 550.

[0104] Step 510, obtaining a second pelvic floor magnetic resonance imaging to be used for pelvic staging.

[0105] Considering the complex structure of the pelvic floor, it is difficult to meet the needs of precise treatment by relying solely on the overall pelvic organ prolapse staging. For example, the female pelvis can be divided into three chambers from front to back based on the anatomical region. The anterior pelvic cavity includes the bladder and urethra, the middle pelvic cavity contains the uterus and vagina, and the posterior pelvic cavity contains the rectum and anal canal. Based on this, the pelvic staging of the embodiment of the present invention can more accurately point out the specific situation of pelvic floor prolapse in different anatomical positions, and prolapse in different areas may cause different symptoms. For example, anterior pelvic prolapse may cause urinary system symptoms such as urinary incontinence, middle pelvic prolapse may cause gynecological symptoms such as vaginal prolapse, and posterior pelvic prolapse may cause intestinal symptoms such as defecation disorders. By staging the pelvis, doctors can formulate more targeted treatment plans based on the degree of prolapse and corresponding symptoms in each area.

[0106] Here, the second pelvic floor MRI is used to determine each pelvic organ prolapse staging result, that is, it is a pelvic floor MRI of a person to be staging pelvic organ prolapse.

[0107] In one embodiment, the second pelvic floor magnetic resonance imaging is a mid-sagittal MRI, that is, the mid-sagittal MRI is the key layer for diagnosis.

[0108] Furthermore, the second pelvic floor MRI is dynamic MRI. Based on this, the characteristics of POP can be clearly observed directly on dynamic MRI under the Valsalva maneuver, so dynamic MRI can not only be used for the diagnosis of pelvic organ prolapse, but also provide excellent visualization of pelvic organs and structures.

[0109] For example, if model training is required based on sample pelvic floor MRI, a sample pelvic floor MRI dataset can be established based on more subjects, and the size of each image is (Unit: mm), and the midsagittal MRI of each subject at rest (Rest) and under Valsalva maneuver was selected as the sample pelvic floor MRI. Among them, all DICOM images were converted to PNG format, and the grayscale values ​​were copied on each channel to convert the single-channel DICOM image into a three-channel RGB image. In order to prevent data leakage, the sample pelvic floor MRI dataset was divided into training set, validation set and test set in a ratio of 6:2:2 based on the case as the division unit, that is, to ensure that the midsagittal MRI of the same case at rest and under Valsalva maneuver is in the same dataset.

[0110] Step 520, input the second pelvic floor magnetic resonance imaging into the key point detection model, and obtain the pubic symphysis lower edge detection result, the distal coccyx joint detection result, the first key point detection result, the second key point detection result and the third key point detection result output by the key point detection model.

[0111] Among them, the key point detection model is used to detect the lower edge of the pubic symphysis, the distal coccyx joint, the first key point, the second key point and the third key point in the second pelvic floor magnetic resonance imaging.

[0112] Among them, the whole pelvic cavity is divided into anterior pelvic cavity, middle pelvic cavity and posterior pelvic cavity, the first key point is the prolapse reference point of the anterior pelvic cavity, the second key point is the prolapse reference point of the middle pelvic cavity, and the third key point is the prolapse reference point of the posterior pelvic cavity.

[0113] Further, the first key point is the lowest prolapse reference point of the anterior pelvic cavity, the second key point is the lowest prolapse reference point of the middle pelvic cavity, and the third key point is the lowest prolapse reference point of the posterior pelvic cavity. Exemplarily, the first key point is the lowest edge of the bladder; the second key point is the anterior lower edge of the cervix; and the third key point is the anterior edge of the anorectal junction.

[0114] In one embodiment, the key point detection model is trained based on the sample second pelvic floor magnetic resonance imaging and its corresponding pubic symphysis lower edge detection result label, distal coccyx joint detection result label, first key point detection result label, second key point detection result label and third key point detection result label.

[0115] Exemplarily, each key point detection result label may be annotated in the following manner: Use software to sequentially annotate five key points on the second pelvic floor magnetic resonance imaging of each sample, and obtain five (X, Y) coordinate values ​​respectively.

[0116] Among them, the lower edge of the pubic symphysis can be the pubic point, and the distal coccygeal joint can be the sacrococcygeal point (the last sacrococcygeal joint).

[0117] Furthermore, when training the key point detection model, KLDiscretLoss (Kullback-Leibler discrete loss function) can be used to measure the difference between the model prediction result and the true label. It is based on the KL divergence principle and minimizes the KL divergence between the predicted distribution and the true distribution, so that the prediction of the key point detection model is closer to the actual situation. The definition formula of KL divergence is as follows: ; In the formula, represents the probability distribution of key point positions predicted by the model, represents the true distribution of key point locations, , They represent the probabilities of different discrete positions where key points may appear.

[0118] Furthermore, during the training of the key point detection model, the optimizer is selected, and the learning rate and weight decay coefficient are set to adjust the step size of the model parameter update and prevent overfitting. A linearly increasing learning rate is used in the early stage of training, and then the cosine annealing strategy is used to gradually reduce the learning rate so that the model can converge better during the training process. The learning rate is automatically scaled according to the actual training batch size to maintain the stability and consistency of the training.

[0119] Furthermore, considering that the amount of data of the sample second pelvic floor MRI is not large, the transfer learning method is used. First, the existing posture key point detection model is pre-trained based on a large amount of data from the human posture key point dataset, and then migrated to the sample second pelvic floor MRI dataset for further training, thereby improving the detection accuracy of the key points.

[0120] Step 530: Determine the target PCL in the second pelvic floor magnetic resonance imaging based on the detection result of the lower edge of the pubic symphysis and the detection result of the distal coccyx joint.

[0121] Specifically, the starting and ending points of the target PCL are the lower edge of the pubic symphysis and the distal coccygeal joint, respectively.

[0122] Step 540, based on the first key point detection result, the second key point detection result and the third key point detection result, determine the first vertical distance between the first key point and the target PCL, the second vertical distance between the second key point and the target PCL, and the third vertical distance between the third key point and the target PCL.

[0123] In a specific embodiment, if the first key point, the second key point or the third key point is below the PCL, the vertical distance is recorded as a positive value; if the first key point, the second key point or the third key point is above the PCL, the vertical distance is recorded as a negative value.

[0124] For example, Figure 6 As shown, Figure 6 The midpoint 1 represents the lower edge of the pubic symphysis. Figure 6 The midpoint 2 represents the distal coccygeal joint. Figure 6 The midpoint 3 represents the first key point, Figure 6 The midpoint 4 represents the second key point, Figure 6 The midpoint 5 represents the third key point, where Figure 6 The vertical distance and target PCL line are also shown.

[0125] Step 550, based on the first vertical distance, determine the anterior pelvic prolapse staging result, based on the second vertical distance, determine the middle pelvic prolapse staging result, and based on the third vertical distance, determine the posterior pelvic prolapse staging result.

[0126] Specifically, based on the first vertical distance and multiple first preset distance intervals, the anterior pelvic prolapse staging result is determined; based on the second vertical distance and multiple second preset distance intervals, the middle pelvic prolapse staging result is determined; based on the third vertical distance and multiple third preset distance intervals, the posterior pelvic prolapse staging result is determined.

[0127] For example, if the first key point is the lowest edge of the bladder, the second key point is the anterior lower edge of the cervix, and the third key point is the anterior edge of the anorectal junction, the staging criteria for the anterior, middle and posterior pelvic cavity with the PCL line as the reference line are as follows: Anterior pelvic cavity: when the first vertical distance between the lowest edge of the bladder and the PCL line is less than 1 cm, the staging result of anterior pelvic prolapse is grade 0; when the first vertical distance between the lowest edge of the bladder and the PCL line is between 1 cm and 3 cm (excluding 3 cm), the staging result of anterior pelvic prolapse is grade I; when the first vertical distance between the lowest edge of the bladder and the PCL line is between 3 cm and 6 cm, the staging result of anterior pelvic prolapse is grade II; when the first vertical distance between the lowest edge of the bladder and the PCL line is greater than 6 cm, the staging result of anterior pelvic prolapse is grade III; Mid-pelvic cavity: When the second vertical distance of the anterior lower edge of the cervix under the PCL line is less than 1 cm, the mid-pelvic cavity prolapse staging result is grade 0; when the second vertical distance of the anterior lower edge of the cervix under the PCL line is between 1 cm and 3 cm (excluding 3 cm), the mid-pelvic cavity prolapse staging result is grade I; when the second vertical distance of the anterior lower edge of the cervix under the PCL line is between 3 cm and 6 cm, the mid-pelvic cavity prolapse staging result is grade II; when the second vertical distance of the anterior lower edge of the cervix under the PCL line is greater than 6 cm, the mid-pelvic cavity prolapse staging result is grade III; Posterior pelvis: When the third vertical distance of the front edge of the anorectal junction under the PCL line is less than 1 cm, the posterior pelvic prolapse staging result is grade 0; when the third vertical distance of the front edge of the anorectal junction under the PCL line is between 1 cm and 3 cm (excluding 3 cm), the posterior pelvic prolapse staging result is grade I; when the third vertical distance of the front edge of the anorectal junction under the PCL line is between 3 cm and 6 cm, the posterior pelvic prolapse staging result is grade II; when the third vertical distance of the front edge of the anorectal junction under the PCL line is greater than 6 cm, the posterior pelvic prolapse staging result is grade III.

[0128] Further, the prolapse type is determined based on the anterior pelvic prolapse staging results, the middle pelvic prolapse staging results, and the posterior pelvic prolapse staging results. Specifically, the pelvic cavity with the most severe prolapse is determined based on the anterior pelvic prolapse staging results, the middle pelvic prolapse staging results, and the posterior pelvic prolapse staging results; the pelvic cavity with the most severe prolapse is taken as the prolapse type. For example, if the anterior pelvic prolapse staging result is grade III, the middle pelvic prolapse staging result is grade I, and the posterior pelvic prolapse staging result is grade 0, then the prolapse type is anterior pelvic prolapse. Further, the prolapse type, the anterior pelvic prolapse staging result, the middle pelvic prolapse staging result, and the posterior pelvic prolapse staging result are all recorded.

[0129] It should be understood that in pelvic staging, the prolapse condition of each part is evaluated independently, and each area has its own independent measurement indicators and staging criteria, so as to more accurately depict the prolapse situation of each pelvic area, that is, to improve the accuracy of pelvic prolapse staging.

[0130] It should be understood that with the accumulation of a large amount of pelvic floor MRI data in recent years, doctors are faced with the challenge of processing more scan information. The existing MRI-based POP quantitative staging is usually based on some reference lines and reference points to measure the key positions of pelvic organs. However, the traditional manual labeling of these reference lines and reference points is time-consuming and laborious, which greatly wastes the valuable working time of clinicians. Therefore, key point positioning can improve diagnostic efficiency and accuracy, and help complete the labeling work in a fast and consistent manner, which can reduce subjective interpretation, reduce dependence on doctors' clinical experience and expertise, and promote telemedicine diagnosis.

[0131] It should be understood that there is a complex pathological relationship between the three organs of the pelvic floor, the anterior, middle and posterior pelvic cavity. The degree of prolapse in the prior art is not accurate enough, has certain limitations, and cannot meet clinical needs. The correct diagnosis of the type of pelvic floor prolapse is crucial, because different types of prolapse require different repair methods. Wrong treatment may significantly increase the recurrence rate of prolapse. In addition, when a patient suffers from multiple types of prolapse at the same time, if only one of them is repaired, other untreated prolapse conditions may worsen. Therefore, the embodiments of the present invention can excellently describe the lesions at both qualitative and quantitative levels, so that it can prompt the doctor whether there is overall prolapse of the pelvic organs, and explain the severity of the overall prolapse, and can also prompt the doctor whether there is prolapse of the anterior, middle and posterior pelvic organs respectively, and explain the severity of prolapse of each pelvic cavity respectively.

[0132] The pelvic organ prolapse staging method provided in the embodiment of the present invention obtains a second pelvic floor magnetic resonance image to be staging the pelvis, and inputs the second pelvic floor magnetic resonance image into a key point detection model to obtain a pubic symphysis lower edge detection result, a distal coccygeal joint detection result, a first key point detection result, a second key point detection result, and a third key point detection result output by the key point detection model, so as to determine a target PCL in the second pelvic floor magnetic resonance image based on the pubic symphysis lower edge detection result and the distal coccygeal joint detection result, and to determine a first vertical distance between the first key point and the target PCL, a second vertical distance between the second key point and the target PCL, and a third key point detection result based on the first key point detection result, the second key point detection result, and the third key point detection result. The third vertical distance between the three key points and the target PCL is used to determine the anterior pelvic prolapse staging result based on the first vertical distance, the middle pelvic prolapse staging result based on the second vertical distance, and the posterior pelvic prolapse staging result based on the third vertical distance, so that the prolapse condition of each cavity part can be independently evaluated, so as to more finely depict the prolapse situation of each pelvic area, that is, to improve the accuracy of pelvic prolapse staging; at the same time, the overall pelvis is divided into anterior pelvic cavity, middle pelvic cavity and posterior pelvic cavity, the first key point is the prolapse reference point of the anterior pelvic cavity, the second key point is the prolapse reference point of the middle pelvic cavity, and the third key point is the prolapse reference point of the posterior pelvic cavity, so as to more accurately perform pelvic prolapse staging, that is, to improve the accuracy of pelvic prolapse staging.

[0133] Based on any of the above embodiments, in the method, the above step 520 includes: step 521 and step 522.

[0134] Step 521, input the second pelvic floor magnetic resonance imaging into the key point detection model, and obtain the initial pubic symphysis lower edge detection result, the initial distal coccyx joint detection result, the initial first key point detection result, the initial second key point detection result and the initial third key point detection result output by the key point detection model.

[0135] Step 522, input the initial pubic symphysis lower edge detection result into the first key point fine-tuning model corresponding to the pubic symphysis lower edge, and obtain the pubic symphysis lower edge detection result output by the first key point fine-tuning model, and input the initial distal coccyx joint detection result into the second key point fine-tuning model corresponding to the distal coccyx joint, and obtain the distal coccyx joint detection result output by the second key point fine-tuning model, and input the initial first key point detection result into the third key point fine-tuning model corresponding to the first key point, and obtain the first key point detection result output by the third key point fine-tuning model, and input the initial second key point detection result into the fourth key point fine-tuning model corresponding to the second key point, and obtain the second key point detection result output by the fourth key point fine-tuning model, and input the initial third key point detection result into the fifth key point fine-tuning model corresponding to the third key point, and obtain the third key point detection result output by the fifth key point fine-tuning model.

[0136] Considering the high complexity of coordinate regression of key points, it is difficult to accurately predict all key points at the same time within a high-precision range, that is, the key point detection model may still have errors. Based on this, the key point fine-tuning model is used to further fine-tune the initial predicted key point positions, so that each key point is independently refined by a convolutional neural network to achieve more accurate coordinate prediction, that is, to improve the detection accuracy of key points, and then improve the accuracy of pelvic prolapse staging.

[0137] Further, based on the initial detection result of the lower edge of the pubic symphysis, a first image block is intercepted on the second pelvic floor magnetic resonance imaging, and the first image block is input into the first key point fine-tuning model corresponding to the lower edge of the pubic symphysis, so as to obtain the detection result of the lower edge of the pubic symphysis output by the first key point fine-tuning model. Based on the initial detection result of the distal coccyx joint, a second image block is intercepted on the second pelvic floor magnetic resonance imaging, and the second image block is input into the second key point fine-tuning model corresponding to the distal coccyx joint, so as to obtain the detection result of the distal coccyx joint output by the second key point fine-tuning model. Based on the initial first key point detection result, a third image block is intercepted on the second pelvic floor magnetic resonance imaging, and the third image block is input into the third key point fine-tuning model corresponding to the first key point, so as to obtain the first key point detection result output by the third key point fine-tuning model. Based on the initial second key point detection result, a fourth image block is intercepted on the second pelvic floor magnetic resonance imaging, and the fourth image block is input into the fourth key point fine-tuning model corresponding to the second key point, so as to obtain the second key point detection result output by the fourth key point fine-tuning model. Based on the initial third key point detection result, a fifth image block is intercepted on the second pelvic floor magnetic resonance imaging, and the fifth image block is input into the fifth key point fine-tuning model corresponding to the third key point, so as to obtain the third key point detection result output by the fifth key point fine-tuning model. In other words, each key point further extracts key points through the image block extracted around its initial position by its corresponding key point fine-tuning model, thereby further improving the detection accuracy of the key points, and further improving the accuracy of pelvic prolapse staging. Among them, the image block can be a rectangular area.

[0138] It should be understood that based on the initial key point detection results, image blocks are intercepted and then input into the key point fine-tuning model. It can be regarded as a multi-level regression problem. The cascaded convolutional neural network is used to extract the input image of each subsequent network based on the output of the previous network. The convolutional neural network is used to learn the corresponding objective function in each stage, so as to further improve the detection accuracy of key points, and then further improve the accuracy of pelvic prolapse staging. The cascade architecture can utilize more training data and help prevent overfitting problems, so as to further improve the detection accuracy of key points, and then further improve the accuracy of pelvic prolapse staging.

[0139] It should be understood that each key point is fine-tuned by its corresponding key point fine-tuning model, that is, instead of using a unified key point fine-tuning model, so as to further improve the accuracy of key point detection and further improve the accuracy of pelvic prolapse staging.

[0140] In one embodiment, the key point fine-tuning model can be trained by a mean square error loss function.

[0141] It should be noted that the key point fine-tuning model can be evaluated by the mean radial error (MRE), standard deviation (SD) and success detection rate (SDR).

[0142] Among them, MRE is used to measure the average Euclidean distance between the predicted key point coordinates and the manually annotated key point coordinates. Its formula is as follows: ; In the formula, Indicates The average radial error of key points is It is In the image The radial error of key points (Euclidean distance), is the number of images. By calculating the average error of each key point in all images, the accuracy of the model in locating the key point can be evaluated. The smaller the MRE value, the more accurate the positioning.

[0143] The formula for standard deviation is as follows: ; In the formula, Indicates The standard deviation of the key points, Indicates The average radial error of key points is It is In the image The radial error of key points (Euclidean distance), is the number of images. SD can evaluate the stability of the model in locating key points. The smaller the SD value, the more stable the positioning and the smaller the fluctuation range.

[0144] Among them, the successful detection rate means that for a detected key point, if the radial error between the predicted value and the true value does not exceed δ mm, then it is considered to be successfully detected. The calculation formula for the successful detection rate of δ mm is as follows: ; In the formula, Indicates The predicted value of the key point, Indicates The true value of the key point, represents the set of predictions for all images, Represents a cardinality function.

[0145] The pelvic organ prolapse staging method provided in an embodiment of the present invention uses a key point fine-tuning model to perform further fine-tuning around the initially predicted key point positions, so that each key point is independently refined by a convolutional neural network to achieve more accurate coordinate prediction, that is, to improve the detection accuracy of the key points, and then to improve the accuracy of the pelvic prolapse staging; in other words, each key point is fine-tuned by its corresponding key point fine-tuning model, that is, instead of using a unified key point fine-tuning model, thereby further improving the detection accuracy of the key points, and then further improving the accuracy of the pelvic prolapse staging.

[0146] Based on any of the above embodiments, in the method, the above step 510 includes: step 511 and step 512.

[0147] Step 511, obtaining original pelvic floor magnetic resonance imaging.

[0148] Here, the original pelvic floor magnetic resonance imaging may be an acquired pelvic floor magnetic resonance imaging, and thus preprocessing the original pelvic floor magnetic resonance imaging may improve the accuracy of the overall pelvic organ prolapse staging. For example, the original pelvic floor magnetic resonance imaging is an original human abdominal MRI.

[0149] Step 512: input the original pelvic floor magnetic resonance imaging into a target detection model to obtain the second pelvic floor magnetic resonance imaging output by the target detection model.

[0150] The second pelvic floor MRI is a rectangular area surrounding the lower edge of the pubic symphysis, the distal coccygeal joint, the first key point, the second key point and the third key point; the size of the second pelvic floor MRI is a preset fixed value. In other words, the second pelvic floor MRI is defined as the minimum enclosing rectangle of all key points with a specified margin. That is, the second pelvic floor MRI is a rectangular pelvic floor area including the bladder, vagina, uterus, rectum, pubic bone and sacrococcygeal joint.

[0151] Exemplarily, the first key point is the lowest edge of the bladder, the second key point is the anterior lower edge of the cervix, and the third key point is the anterior edge of the anorectal junction.

[0152] It should be understood that the original pelvic floor magnetic resonance imaging is input into the target detection model, that is, the original pelvic floor magnetic resonance imaging is cropped to obtain the second pelvic floor magnetic resonance imaging actually used to determine the staging results of each pelvic prolapse, so as to remove irrelevant information, thereby improving the accuracy of pelvic organ prolapse staging. And the original pelvic floor magnetic resonance imaging is aligned to improve the accuracy of the overall pelvic organ prolapse staging.

[0153] Exemplarily, the object detection model is used to locate the pelvic floor region in the original pelvic floor magnetic resonance imaging.

[0154] It should be understood that the original pelvic floor magnetic resonance imaging is input into the target detection model to obtain the second pelvic floor magnetic resonance imaging output by the target detection model, and then the second pelvic floor magnetic resonance imaging is input into the key point detection model. It can be regarded as a multi-level regression problem. The cascaded convolutional neural network is used to extract the input image of each subsequent network based on the output of the previous network. The convolutional neural network is used to learn the corresponding objective function in each stage, so as to further improve the detection accuracy of the key points, and then further improve the accuracy of the pelvic prolapse staging. The cascade architecture can utilize more training data and help prevent overfitting problems, so as to further improve the detection accuracy of the key points, and then further improve the accuracy of the pelvic prolapse staging.

[0155] Exemplarily, during the training of the target detection model, a rectangular pelvic floor region including the bladder, vagina, uterus, rectum, pubic bone and sacrococcygeal joint is annotated in the sample original pelvic floor magnetic resonance imaging using annotation software to reduce noise interference of irrelevant information. The cross entropy loss is used as the loss function for model training. For example, a target detection model is trained using the Faster R-CNN target detection algorithm model.

[0156] The pelvic organ prolapse staging method provided by the embodiment of the present invention inputs the original pelvic floor magnetic resonance imaging into the target detection model, that is, crops the original pelvic floor magnetic resonance imaging to obtain a second pelvic floor magnetic resonance imaging that is actually used to determine each pelvic prolapse staging result, so as to remove irrelevant information, thereby improving the accuracy of pelvic organ prolapse staging; and the second pelvic floor magnetic resonance imaging is a rectangular area surrounding the lower edge of the pubic symphysis, the distal coccygeal joint, the first key point, the second key point and the third key point, so that the original pelvic floor magnetic resonance imaging is aligned in the above manner, thereby improving the accuracy of the overall pelvic organ prolapse staging.

[0157] To facilitate understanding of the above embodiments, a specific embodiment is described here. Figure 7 As shown, Figure 7The figure represents a three-stage cascade convolutional neural network; the first input is an original human abdominal MRI (original pelvic floor MRI). In the first stage, the pelvic area is located by using the target detection model (ROI positioning network), that is, the ROI (Region Of Interest) area is located to obtain a pelvic image (second pelvic floor MRI); in the second stage, the initial positions (initial coordinates) of the five key points are predicted simultaneously by the key point detection model (key point initial positioning network), that is, 1-5 marked in the figure; in the last stage, each key point is fine-tuned around the image block extracted from its initial position by its corresponding key point fine-tuning model (fine-tuning network-1 to fine-tuning network-5); the final predicted key points are annotated in the second pelvic floor MRI, that is, the output of key points 1 to key points 5 are merged.

[0158] Based on the above embodiments, the embodiments of the present invention can automatically grade the pelvic floor as a whole and in each pelvic cavity, provide doctors with more accurate diagnostic information, reduce the subjectivity of diagnosis, and thus improve the consistency of diagnosis. Specifically, the present invention automatically grades and classifies the pelvic floor based on deep learning and pressure MRI, and can describe the lesions excellently at both qualitative and quantitative levels. It not only automatically segments and quantifies based on the new measurement method AUPCL for the first time, but also adopts a multimodal fusion method to achieve the automatic grading of the overall degree of pelvic prolapse. And based on the key point detection method, the pelvic cavity prolapse grading method combined with the PCL (pubic tail line) and the lowest point of each cavity prolapse measurement has higher accuracy than the end-to-end grading method of deep learning in previous studies, and can more accurately determine the POP type and carry out targeted treatment. The embodiment of the present invention establishes a POP computer-assisted clinical diagnosis system based on deep learning and pressure MRI, and establishes a complete evaluation system from overall grading to specific cavity grading to prolapse type, providing doctors with comprehensive and accurate diagnostic information, reducing the subjectivity of diagnosis and improving the consistency of diagnosis.

[0159] The pelvic organ prolapse staging device provided by the present invention is described below. The pelvic organ prolapse staging device described below and the pelvic organ prolapse staging method described above can be referred to each other.

[0160] Figure 8 Schematic diagram of the structure of the pelvic organ prolapse staging device provided by the present invention, such as Figure 8 As shown, the pelvic organ prolapse staging device includes an imaging acquisition module 810 , an image segmentation module 820 and a prolapse staging module 830 .

[0161] The imaging acquisition module 810 is used to acquire a first pelvic floor magnetic resonance imaging to be staged as a whole.

[0162] The image segmentation module 820 is used to input the first pelvic floor magnetic resonance imaging into the image segmentation model to obtain the image segmentation result output by the image segmentation model; the image segmentation result includes the segmentation mask information of the target pelvic floor area and / or the area of ​​the target pelvic floor area, the target pelvic floor area is the cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor area includes the coccygeal levator plate and the perineal floor edge.

[0163] The prolapse staging module 830 is used to input the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model.

[0164] Wherein, the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to evaluate the severity of overall pelvic organ prolapse.

[0165] Fig. 9 An example of a physical structure diagram of an electronic device is shown in FIG. Fig. 9 As shown, the electronic device may include: a processor (processor) 910 , a communication interface (Communications Interface) 920 , a memory (memory) 930 and a communication bus 940 , wherein the processor 910 , the communication interface 920 , and the memory 930 communicate with each other through the communication bus 940 . The processor 910 can call the logic instructions in the memory 930 to execute the pelvic organ prolapse staging method, which includes: obtaining a first pelvic floor magnetic resonance imaging to be staged as a whole; inputting the first pelvic floor magnetic resonance imaging into an image segmentation model to obtain an image segmentation result output by the image segmentation model; the image segmentation result includes segmentation mask information of a target pelvic floor area and / or an area of ​​a target pelvic floor area, the target pelvic floor area is a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor area includes the coccygeal levator plate and the perineal floor edge; inputting the image segmentation result into the overall pelvic organ prolapse staging model to obtain an overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model; wherein the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to assess the severity of overall pelvic organ prolapse.

[0166] In addition, the logic instructions in the above-mentioned memory 930 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0167] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the pelvic organ prolapse staging method provided by the above-mentioned methods, the method comprising: obtaining a first pelvic floor magnetic resonance imaging to be staged as a whole; inputting the first pelvic floor magnetic resonance imaging into an image segmentation model to obtain an image segmentation result output by the image segmentation model; the image segmentation result includes segmentation mask information of a target pelvic floor area and / or an area of ​​a target pelvic floor area, the target pelvic floor area is a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor area includes the coccygeal levator plate and the perineal floor edge; inputting the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model; wherein the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to assess the severity of overall pelvic organ prolapse.

[0168] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented by a processor to execute the pelvic organ prolapse staging method provided by the above-mentioned methods when the computer program is executed, the method comprising: obtaining a first pelvic floor magnetic resonance imaging to be staged as a whole; inputting the first pelvic floor magnetic resonance imaging into an image segmentation model to obtain an image segmentation result output by the image segmentation model; the image segmentation result comprises segmentation mask information of a target pelvic floor area and / or an area of ​​a target pelvic floor area, the target pelvic floor area being a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor area comprising the coccygeal levator plate and the perineal floor edge; inputting the image segmentation result into an overall pelvic organ prolapse staging model to obtain an overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model; wherein the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to assess the severity of overall pelvic organ prolapse.

[0169] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0170] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for staging pelvic organ prolapse, characterized in that: include: Obtain the first pelvic floor MRI to be staged overall; Inputting the first pelvic floor magnetic resonance imaging into an image segmentation model to obtain an image segmentation result output by the image segmentation model; The image segmentation result includes segmentation mask information of a target pelvic floor region and / or an area of ​​a target pelvic floor region, wherein the target pelvic floor region is a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor region includes the coccygeal levator plate and the edge of the perineum floor; Inputting the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model; Wherein, the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to evaluate the severity of overall pelvic organ prolapse.

2. The method for staging pelvic organ prolapse according to claim 1, characterized in that: The step of inputting the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model comprises: Inputting the first pelvic floor magnetic resonance imaging and the image segmentation result into an input layer of the overall pelvic organ prolapse staging model to obtain a multimodal fusion feature output by the input layer; The multimodal fusion features are input into the overall pelvic organ prolapse staging layer in the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging layer.

3. The method for staging pelvic organ prolapse according to claim 1, characterized in that: The step of obtaining the first pelvic floor magnetic resonance imaging to be staged as a whole comprises: Obtain native pelvic floor magnetic resonance imaging; The original pelvic floor magnetic resonance imaging is cropped to obtain the first pelvic floor magnetic resonance imaging; the first pelvic floor magnetic resonance imaging includes a plurality of preset organ regions and / or a plurality of preset body regions.

4. The method for staging pelvic organ prolapse according to claim 3, characterized in that: The step of trimming the original pelvic floor magnetic resonance imaging to obtain the first pelvic floor magnetic resonance imaging includes: Inputting the original pelvic floor magnetic resonance imaging into a key point detection model to obtain a pubic point detection result and a sacrococcygeal point detection result output by the key point detection model; the key point detection model is used to detect the pubic point and sacrococcygeal point in the original pelvic floor magnetic resonance imaging; Determine the sacrococcygeal subpubic point SCIPP line based on the pubic point detection result and the sacrococcygeal point detection result; Performing a rotation transformation on the original pelvic floor magnetic resonance imaging based on the SCIPP line to obtain a transformed original pelvic floor magnetic resonance imaging, so that the angle between the SCIPP line and the horizontal line of the transformed original pelvic floor magnetic resonance imaging is a preset fixed angle; Based on the pubic point detection result and the sacrococcygeal point detection result, on the transformed original pelvic floor magnetic resonance imaging, the edges are expanded outwards with the pubic point and the sacrococcygeal point as starting points, respectively, to obtain a rectangular area; The first pelvic floor magnetic resonance imaging is obtained by cropping the transformed original pelvic floor magnetic resonance imaging based on the rectangular area.

5. The method for staging pelvic organ prolapse according to any one of claims 1 to 4, characterized in that: The pelvic organ prolapse staging method further comprises: Obtain a second pelvic floor MRI to be performed for pelvic staging; Inputting the second pelvic floor magnetic resonance imaging into a key point detection model, obtaining a pubic symphysis lower edge detection result, a distal coccygeal joint detection result, a first key point detection result, a second key point detection result, and a third key point detection result output by the key point detection model; the key point detection model is used to detect the pubic symphysis lower edge, the distal coccygeal joint, the first key point, the second key point, and the third key point in the second pelvic floor magnetic resonance imaging; Determining a target PCL in the second pelvic floor magnetic resonance imaging based on the detection result of the lower edge of the pubic symphysis and the detection result of the distal coccygeal joint; Determine, based on the first key point detection result, the second key point detection result, and the third key point detection result, a first vertical distance between the first key point and the target PCL, a second vertical distance between the second key point and the target PCL, and a third vertical distance between the third key point and the target PCL; Determine an anterior pelvic prolapse staging result based on the first vertical distance, determine a middle pelvic prolapse staging result based on the second vertical distance, and determine a posterior pelvic prolapse staging result based on the third vertical distance; Among them, the whole pelvic cavity is divided into anterior pelvic cavity, middle pelvic cavity and posterior pelvic cavity, the first key point is the prolapse reference point of the anterior pelvic cavity, the second key point is the prolapse reference point of the middle pelvic cavity, and the third key point is the prolapse reference point of the posterior pelvic cavity.

6. The method for staging pelvic organ prolapse according to claim 5, characterized in that: The step of inputting the second pelvic floor magnetic resonance imaging into the key point detection model to obtain the pubic symphysis lower edge detection result, the distal coccyx joint detection result, the first key point detection result, the second key point detection result and the third key point detection result output by the key point detection model includes: Inputting the second pelvic floor magnetic resonance imaging into a key point detection model, obtaining an initial pubic symphysis lower edge detection result, an initial distal coccygeal joint detection result, an initial first key point detection result, an initial second key point detection result, and an initial third key point detection result output by the key point detection model; The initial pubic symphysis lower edge detection result is input into the first key point fine-tuning model corresponding to the pubic symphysis lower edge, and the pubic symphysis lower edge detection result output by the first key point fine-tuning model is obtained; the initial distal coccyx joint detection result is input into the second key point fine-tuning model corresponding to the distal coccyx joint, and the distal coccyx joint detection result output by the second key point fine-tuning model is obtained; the initial first key point detection result is input into the third key point fine-tuning model corresponding to the first key point, and the first key point detection result output by the third key point fine-tuning model is obtained; the initial second key point detection result is input into the fourth key point fine-tuning model corresponding to the second key point, and the second key point detection result output by the fourth key point fine-tuning model is obtained; the initial third key point detection result is input into the fifth key point fine-tuning model corresponding to the third key point, and the third key point detection result output by the fifth key point fine-tuning model is obtained.

7. The method for staging pelvic organ prolapse according to claim 5, characterized in that: The step of obtaining a second pelvic floor magnetic resonance imaging to be used for pelvic staging includes: Obtain native pelvic floor magnetic resonance imaging; Inputting the original pelvic floor magnetic resonance imaging into a target detection model to obtain the second pelvic floor magnetic resonance imaging output by the target detection model; Among them, the second pelvic floor magnetic resonance imaging is a rectangular area surrounding the lower edge of the pubic symphysis, the distal coccyx joint, the first key point, the second key point and the third key point; the size of the second pelvic floor magnetic resonance imaging is a preset fixed value.

8. A pelvic organ prolapse staging device, characterized in that: include: An imaging acquisition module, used to acquire a first pelvic floor magnetic resonance imaging to be staged overall; an image segmentation module, configured to input the first pelvic floor magnetic resonance imaging into an image segmentation model, and obtain an image segmentation result output by the image segmentation model; the image segmentation result includes segmentation mask information of a target pelvic floor region and / or an area of ​​a target pelvic floor region, the target pelvic floor region being a cross-sectional area below the pubococcygeus line PCL in the first pelvic floor magnetic resonance imaging, and the boundary of the target pelvic floor region including the coccygeal levator plate and the edge of the perineum; A prolapse staging module, used for inputting the image segmentation result into the overall pelvic organ prolapse staging model to obtain the overall pelvic organ prolapse staging result output by the overall pelvic organ prolapse staging model; Wherein, the image segmentation model is used to segment the first pelvic floor magnetic resonance imaging to obtain the target pelvic floor area; the overall pelvic organ prolapse staging model is used to evaluate the severity of overall pelvic organ prolapse.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the pelvic organ prolapse staging method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for staging pelvic organ prolapse according to any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for staging pelvic organ prolapse according to any one of claims 1 to 7 is implemented.

Images