SPARCC automatic scoring method, system and equipment for sacroiliac arthritis and medium
Through the pre-trained network of the asymmetric encoder-decoder architecture and data augmentation technology, the automated SPARCC scoring of sacroiliac joint inflammation was achieved, which solved the problems of time-consuming and labor-intensive traditional methods and inconsistent scoring, and improved the efficiency and accuracy of scoring.
Patent Information
- Application Number
- CN202510694285.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-28
- Publication Date
- 2025-09-09
AI Technical Summary
The traditional SPARCC scoring method for sacroiliac joint inflammation relies on manual visual assessment, which is time-consuming and labor-intensive and has the problem of inconsistent scoring results.
A pre-trained network with an asymmetric encoder-decoder architecture, combined with data augmentation and transfer learning, automatically scores sacroiliac joint MRI images, predicts SPARCC scores layer by layer, and accumulates the total score, reducing dependence on labeled data and solving the problem of model overfitting.
It achieves efficient and accurate automated scoring of sacroiliac joint inflammation, reduces the subjective differences in scoring results, improves the consistency and accuracy of scoring, and meets the needs of clinical diagnosis.
Smart Images
Figure CN120612294A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of medical imaging artificial intelligence technology, and in particular to a sacroiliac joint inflammation SPARCC automated scoring method, system, device and medium. Background Art
[0002] Sacroiliitis, an inflammatory disease affecting the joint between the sacrum and ilium, is common in various rheumatic diseases, particularly ankylosing spondylitis (AS). To accurately assess the severity of sacroiliitis inflammation, the SPARCC (Spondyloarthritis Research Consortium of Canada) scoring system is widely used clinically. This system quantifies inflammatory activity in the sacroiliac joint based on MRI images, providing an important basis for physicians' diagnosis and treatment. However, the traditional SPARCC scoring method relies on visual assessment by radiologists, which is not only time-consuming and laborious but can also lead to inconsistent scoring results due to differences in physician experience and subjective judgment. Summary of the Invention
[0003] The purpose of the present invention is to provide a sacroiliac joint inflammation SPARCC automated scoring method, system, device and medium, which can reduce the dependence on labeled data, solve the overfitting problem of the CNN model, and reduce the interpretation bias caused by subjective differences in scorers.
[0004] To achieve the above object, the present invention provides the following solutions:
[0005] An automated scoring method for sacroiliac joint inflammation (SPARCC) comprising:
[0006] Acquire a target image; the target image is a coronal STIR sequence MRI image of the sacroiliac joint of a patient meeting the ASAS diagnostic criteria;
[0007] Preprocessing the target image to obtain a data-enhanced training set;
[0008] Based on the training set, tuning parameters of a pre-trained network composed of an asymmetric encoder-decoder architecture to obtain a trained automatic scoring model;
[0009] All MRI images of the sacroiliac joint in the oblique coronal STIR sequence are input into the automated scoring model, the inflammation SPARCC score is predicted layer by layer, and the images of the 6 layers with the highest scores are output and accumulated as the total score.
[0010] Optionally, the specific process of obtaining the target image is:
[0011] Set the scanning parameters and acquire the target image; the scanning parameters include layer thickness 5.0mm, layer spacing 1.0mm, number of layers 20, and FOV 380×380mm 2 .
[0012] Optionally, preprocessing the target image to obtain a data-enhanced training set specifically includes:
[0013] Performing standardization processing on the target image to obtain standard data; the standardization processing includes grayscale normalization, inter-layer alignment and continuous slice screening of the original DICOM data;
[0014] The standard data are subjected to double-blind SPARCC annotation by physicians to obtain a data-enhanced training set; the double-blind SPARCC annotation by physicians includes annotating the inflammation area according to the SPARCC standard, and adding random Gaussian noise and contrast adjustment to perform data enhancement.
[0015] Optionally, based on the training set, parameter tuning is performed on a pre-trained network composed of an asymmetric encoder-decoder architecture to obtain a trained automatic scoring model, specifically including:
[0016] Build a pre-trained network based on an asymmetric encoder-decoder architecture;
[0017] Based on a 75% high mask rate pre-training strategy, the training set is input into the pre-trained network, and training is performed with the goal of minimizing the loss between the network output and the labeled labels in the training set, and the trained network is determined as an automatic scoring model.
[0018] Optionally, the loss function of the automatic scoring model during training includes a mean square error loss function and a binary cross entropy loss function.
[0019] The present invention also provides a sacroiliac joint inflammation SPARCC automated scoring system, comprising:
[0020] An image acquisition unit is used to acquire a target image; the target image is a coronal STIR sequence MRI image of the sacroiliac joint of a patient meeting the ASAS diagnostic criteria;
[0021] A processing unit, configured to preprocess the target image to obtain a data-enhanced training set;
[0022] A model training unit is used to tune parameters of a pre-trained network composed of an asymmetric encoder-decoder architecture based on the training set to obtain a trained automatic scoring model;
[0023] The automated scoring unit is used to input all MRI images of the sacroiliac joint in the oblique coronal STIR sequence into the automated scoring model, predict the inflammation SPARCC score layer by layer, and output the images of the 6 layers with the highest scores, which are accumulated as the total score.
[0024] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-mentioned SPARCC automated scoring method for sacroiliac joint inflammation.
[0025] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned SPARCC automated scoring method for sacroiliac joint inflammation.
[0026] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0027] The present invention discloses a method, system, device, and medium for automated scoring of sacroiliac joint inflammation (SPARCC). The method comprises acquiring a target image; the target image is a STIR sequence MRI image of the sacroiliac joint of a patient meeting the ASAS diagnostic criteria; preprocessing the target image to obtain a data-enhanced training set; based on the training set, performing parameter tuning on a pre-trained network composed of an asymmetric encoder-decoder architecture to obtain a trained automated scoring model; inputting all STIR sequence MRI images of the sacroiliac joint into the automated scoring model, predicting the inflammation SPARCC score layer by layer, and outputting the images of the six layers with the highest scores, which are accumulated as a total score. The present invention can reduce the dependence on labeled data, solve the overfitting problem of the CNN model, and reduce the interpretation bias caused by subjective differences in scorers. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0029] Figure 1 Schematic diagram of the process of the SPARCC automated scoring method for sacroiliac joint inflammation of the present invention;
[0030] Figure 2 This is the overall process logic diagram of this embodiment;
[0031] Figure 3 This is a pre-training flow chart in this embodiment;
[0032] Figure 4 This is a fine-tuning flow chart in this embodiment;
[0033] Figure 5 is the image reconstruction diagram in this embodiment;
[0034] Figure 6 The violin plots for the experiments in this embodiment and the results of the score consistency test are shown in Figure 2. Part (a) is a schematic diagram of the overall score for the sacroiliac joint; Part (b) is a schematic diagram of the single-image score for the sacroiliac joint.
[0035] Figure 7 Schematic diagram of the ROC curve of the overall activity efficacy test in this embodiment. DETAILED DESCRIPTION
[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention.
[0037] The purpose of the present invention is to provide a sacroiliac joint inflammation SPARCC automated scoring method, system, device and medium, which can reduce the dependence on labeled data, solve the overfitting problem of the CNN model, and reduce the interpretation bias caused by subjective differences in scorers.
[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0039] like Figure 1 As shown, the present invention provides a SPARCC automated scoring method for sacroiliac joint inflammation, comprising:
[0040] Step 100: Acquire a target image; the target image is an oblique coronal STIR sequence MRI image of the patient's sacroiliac joint that meets the ASAS (Asia Spondyloarthritis Association International) diagnostic criteria.
[0041] Step 200: Preprocess the target image to obtain a data-enhanced training set.
[0042] Step 300: Based on the training set, perform parameter tuning on a pre-trained network composed of an asymmetric encoder-decoder architecture to obtain a trained automatic scoring model.
[0043] Step 400: All MRI images of the sacroiliac joint in the oblique coronal STIR sequence are input to the automated scoring model, and the inflammation SPARCC score (Spondyloarthritis Research Consortium of Canada score) is predicted layer by layer. The images of the 6 layers with the highest scores are output and accumulated as the total score.
[0044] As a specific embodiment, the specific process of acquiring the target image is as follows: setting scanning parameters and acquiring the target image; the scanning parameters include layer thickness 5.0 mm, layer spacing 1.0 mm, number of layers 20, FOV 380×380 mm 2 .
[0045] As a specific implementation, preprocessing the target image to obtain a data-enhanced training set specifically includes:
[0046] The target image is standardized to obtain standard data; the standardization includes grayscale normalization, inter-layer alignment, and continuous slice screening of the original DICOM data; the standard data is double-blindly annotated by a physician using SPARCC to obtain a data-enhanced training set; the double-blind SPARCC annotation by the physician includes annotating the inflammatory area according to the SPARCC standard, and adding random Gaussian noise and contrast adjustment to perform data enhancement.
[0047] As a specific implementation, based on the training set, parameters of a pre-trained network composed of an asymmetric encoder-decoder architecture are tuned to obtain a trained automated scoring model. This includes: constructing a pre-trained network based on the asymmetric encoder-decoder architecture; inputting the training set into the pre-trained network based on a 75% high mask rate pre-training strategy, training the network with the goal of minimizing the loss between the network output and the labeled labels in the training set, and determining the trained network as the automated scoring model. The loss functions used in the training of the automated scoring model include a mean squared error loss function and a binary cross entropy loss function.
[0048] Based on the above technical solution, the following Figure 2-Figure 7 And the embodiments shown in Table 1.
[0049] in, Figure 6 The ICC value of the sacroiliac joint single image score was 0.980; the ICC value of the sacroiliac joint overall image score was 0.988; the kappa value of the sacroiliac joint overall image grouped according to activity was 0.887. Figure 7 The blue, orange, and green curves in the figure represent the diagnostic efficacy for patients with remission, mild, and severe activity, respectively. Table 1 shows the overall activity efficacy test indicators of this embodiment, including AUC value, accuracy, precision, recall rate, and F1 value.
[0050] Table 1 Overall activity performance test indicators
[0051] AUC Accuracy Accuracy Recall F1 ease 0.932 0.955 0.955 0.955 0.955 Mild O.899 0.909 0.867 0.867 0.867 severe 0.929 0.977 0.857 1 0.933
[0052] S1 Data Preprocessing
[0053] Medical imaging acquisition: Obtain coronal STIR sequence MRI images of the sacroiliac joint of patients who meet the ASAS diagnostic criteria. Scan parameters include layer thickness 5.0 mm, layer spacing 1.0 mm, number of layers 20, and FOV 380 × 380 mm. 2 ;
[0054] Standardization processing: grayscale normalization (-1000 to 2000 HU), inter-slice alignment, and 6-slice continuous slice screening were performed on the original DICOM data;
[0055] Annotation and enhancement: Two radiologists annotated the inflammation areas according to the SPARCC standard, and reached consensus in case of disagreement. Random Gaussian noise (σ = 0.1) and contrast adjustment (± 20%) were added for data enhancement.
[0056] S2 model training
[0057] MAE pre-training stage:
[0058] Model architecture: ViT-Base is used as the encoder (hidden layer dimension 768, 12 attention heads) and a lightweight decoder (hidden layer dimension 512);
[0059] Masking strategy: 75% random block masking, retaining 25% visible blocks to capture the global context of medical images;
[0060] Loss function: pixel-level mean squared error (MSE) reconstruction loss, optimizer is AdamW (learning rate 1e-4, weight decay 0.05), training for 300 epochs;
[0061] Fine-tuning phase:
[0062] Parameter migration: freeze the first 8 layers of the encoder and only fine-tune the last 4 layers and the classification head;
[0063] Supervised training: using binary cross entropy loss function, learning rate reduced to 1e-5, batch size 32;
[0064] Regularization: Dropout rate 0.2, L2 regularization coefficient 1e-4.
[0065] S3 model inference
[0066] Input processing: continuous slices of the test set are fed into the model to generate a heat map of inflammation probability for each layer;
[0067] Score calculation: The bone marrow edema area in each layer was multi-classified according to the SPARCC rules (score ranged from 0 to 12), and the highest scores of the six layers were added together to obtain the patient's total score;
[0068] Verification results: The consistency between the model and radiologists' scores was evaluated by intraclass correlation coefficient (ICC) and Cohen's Kappa coefficient analysis. The consistency between the model and radiologists' diagnosis of sacroiliitis activity was evaluated by receiver operating characteristic (ROC) curve (AUC value), accuracy, precision, recall rate and F1 value. The model inference time was ≤0.5 seconds per case.
[0069] Therefore, it can be seen that this embodiment has the following beneficial effects:
[0070] This method uses a 75% high mask rate pre-training strategy, an asymmetric encoder-decoder architecture to learn medical image features, and combines transfer learning to achieve accurate quantitative assessment of inflammation with limited labeled data. Specifically, it includes: S1 standardized pre-processing of sacroiliac joint MRI data and double-blind SPARCC annotation by physicians; S2 uses the mean square error loss function to pre-train the MAE model and optimizes the binary cross entropy loss through fine-tuning; S3 inputs continuous STIR sequence images, predicts the inflammation score layer by layer, and accumulates the highest scores of 6 layers to output the patient's overall SPARCC score. The present invention significantly reduces the dependence on labeled data and solves the overfitting problem of the CNN model. The scoring results are consistent with experts at ICC>0.88, and the AUC values are all>0.89.
[0071] In addition, the present invention also provides a sacroiliac joint inflammation SPARCC automated scoring system, comprising:
[0072] An image acquisition unit is used to acquire a target image; the target image is a coronal STIR sequence MRI image of the sacroiliac joint of a patient meeting the ASAS diagnostic criteria;
[0073] A processing unit, configured to preprocess the target image to obtain a data-enhanced training set;
[0074] A model training unit is used to tune parameters of a pre-trained network composed of an asymmetric encoder-decoder architecture based on the training set to obtain a trained automatic scoring model;
[0075] The automated scoring unit is used to input all MRI images of the sacroiliac joint in the oblique coronal STIR sequence into the automated scoring model, predict the inflammation SPARCC score layer by layer, and output the images of the 6 layers with the highest scores, which are accumulated as the total score.
[0076] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-mentioned SPARCC automated scoring method for sacroiliac joint inflammation.
[0077] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned SPARCC automated scoring method for sacroiliac joint inflammation.
[0078] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0079] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.
Claims
1. A SPARCC automated scoring method for sacroiliac joint inflammation, characterized in that: include: Acquire the target image; The target image is a coronal STIR sequence MRI image of the sacroiliac joint of a patient who meets the ASAS diagnostic criteria; Preprocessing the target image to obtain a data-enhanced training set; Based on the training set, tuning parameters of a pre-trained network composed of an asymmetric encoder-decoder architecture to obtain a trained automatic scoring model; All MRI images of the sacroiliac joint in the oblique coronal STIR sequence are input into the automated scoring model, the inflammation SPARCC score is predicted layer by layer, and the images of the 6 layers with the highest scores are output and accumulated as the total score.
2. The SPARCC automated scoring method for sacroiliac joint inflammation according to claim 1, characterized in that: The specific process of obtaining the target image is: Set the scanning parameters and acquire the target image; the scanning parameters include layer thickness 5.0mm, layer spacing 1.0mm, number of layers 20, and FOV 380×380mm 2 .
3. The SPARCC automated scoring method for sacroiliac joint inflammation according to claim 1, characterized in that: The target image is preprocessed to obtain a data-enhanced training set, specifically including: Performing standardization processing on the target image to obtain standard data; the standardization processing includes grayscale normalization, inter-layer alignment and continuous slice screening of the original DICOM data; The standard data are subjected to double-blind SPARCC annotation by physicians to obtain a data-enhanced training set; the double-blind SPARCC annotation by physicians includes annotating the inflammation area according to the SPARCC standard, and adding random Gaussian noise and contrast adjustment to perform data enhancement.
4. The SPARCC automated scoring method for sacroiliac joint inflammation according to claim 1, characterized in that: Based on the training set, the parameters of the pre-trained network composed of an asymmetric encoder-decoder architecture are tuned to obtain a trained automatic scoring model, specifically including: Build a pre-trained network based on an asymmetric encoder-decoder architecture; Based on a 75% high mask rate pre-training strategy, the training set is input into the pre-trained network, and training is performed with the goal of minimizing the loss between the network output and the labeled labels in the training set, and the trained network is determined as an automatic scoring model.
5. The SPARCC automated scoring method for sacroiliac joint inflammation according to claim 4, characterized in that: The loss functions of the automatic scoring model during training include a mean square error loss function and a binary cross entropy loss function.
6. A SPARCC automated scoring system for sacroiliac joint inflammation, characterized in that: include: An image acquisition unit, used for acquiring a target image; The target image is a coronal STIR sequence MRI image of the sacroiliac joint of a patient who meets the ASAS diagnostic criteria; A processing unit, configured to preprocess the target image to obtain a data-enhanced training set; A model training unit is used to tune parameters of a pre-trained network composed of an asymmetric encoder-decoder architecture based on the training set to obtain a trained automatic scoring model; The automated scoring unit is used to input all MRI images of the sacroiliac joint in the oblique coronal STIR sequence into the automated scoring model, predict the inflammation SPARCC score layer by layer, and output the images of the 6 layers with the highest scores, which are accumulated as the total score.
7. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the SPARCC automated scoring method for sacroiliac joint inflammation according to any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that The device stores a computer program, which, when executed by a processor, implements the SPARCC automated scoring method for sacroiliac joint inflammation according to any one of claims 1 to 5.