Share：
Share this content in WeChat
X
[Chinese] [PDF] 4 1
• Clinical Article •
A meta-learning-based MRI multimodal classification model for differentiating osteoarthritis with synovitis from rheumatoid arthritis
LI Shenghu  LI Jianhua  PAN Xiang  XIA Peng 

DOI:10.12015/issn.1674-8034.2025.11.024.


[Abstract] Objective This study integrates MRI radiomics with deep learning techniques to construct a classification model for accurate prediction of osteoarthritis (OA) and rheumatoid arthritis (RA) with synovitis.Materials and Methods Through a retrospective analysis of knee MRI data from patients diagnosed with OA or RA between January 2018 and December 2024, eligible scans were selected based on inclusion/exclusion criteria to establish the dataset. To address the challenge of small-sample learning in medical imaging, an improved ResNet3D-18 model based on model-agnostic meta-learning (MAML) was developed, enabling rapid task adaptation in small-sample scenarios to enhance classification performance. Three-dimensional gradient-weighted class activation mapping (Grad-CAM) was employed to interpret prediction results. After anonymizing clinical information, two radiologists independently annotated the MRI scans, with discrepancies resolved by consensus. Model performance was assessed using five-fold cross-validation.Results The dataset comprised 56 RA patients (60 knees) and 56 OA patients (61 knees). Under limited sample conditions, the model demonstrated superior performance in single-modality classification using T1WI (accuracy: 86.2%, AUC: 0.914) compared to T2WI (accuracy: 82.8%, AUC: 0.875). The multimodal model integrating T1WI and T2WI achieved optimal classification (accuracy: 97.5%, AUC: 0.975), outperforming manual classification (accuracy: 94.2%, AUC: 0.932). Grad-CAM heatmaps revealed that the model's attention patterns were highly consistent with the clinical-pathological characteristics of both diseases.Conclusions By integrating MRI radiomics with deep learning, the proposed classification model effectively overcomes the limitation of insufficient training data through the MAML strategy, enabling accurate and reliable prediction of OA and RA with synovitis in the knee joint. This study provides new technical support and a theoretical foundation for early clinical diagnosis, personalized treatment, and prognostic assessment.
[Keywords] rheumatoid arthritis；artificial intelligence；deep learning；magnetic resonance imaging；synovitis；meta-learning；multimodal fusion

LI Shenghu1   LI Jianhua1   PAN Xiang2   XIA Peng1*  

1 Department of Radiology, Wuxi Hospital of Traditional Chinese Medicine, Wuxi 214071, China

2 School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi 214122, China

Corresponding author: XIA P, E-mail: coatway@sohu.com

Conflicts of interest   None.

Received  2025-05-11
Accepted  2025-09-10
DOI: 10.12015/issn.1674-8034.2025.11.024
DOI:10.12015/issn.1674-8034.2025.11.024.

[1]
DI MATTEO A, BATHON J M, EMERY P. Rheumatoid arthritis[J]. Lancet, 2023, 402(10416): 2019-2033. DOI: 10.1016/S0140-6736(23)01525-8.
[2]
FINCKH A, GILBERT B, HODKINSON B, et al. Global epidemiology of rheumatoid arthritis[J]. Nat Rev Rheumatol, 2022, 18(10): 591-602. DOI: 10.1038/s41584-022-00827-y.
[3]
TIAN X P, WANG Q, JIANG N, et al. Chinese guidelines for the diagnosis and treatment of rheumatoid arthritis: 2024 update[J]. Rheumatol Immunol Res, 2025, 5(4): 189-208. DOI: 10.1515/rir-2024-0028.
[4]
COURTIES A, KOUKI I, SOLIMAN N, et al. Osteoarthritis year in review 2024: epidemiology and therapy[J]. Osteoarthr Cartil, 2024, 32(11): 1397-1404. DOI: 10.1016/j.joca.2024.07.014.
[5]
CHOUDHURY S R, PRAKASH M, SHARMA M, et al. Magnetic resonance imaging of knee arthropathies: A pictorial review[J/OL]. J Clin Orthop Trauma, 2024, 61: 102872 [2025-05-05]. https://pubmed.ncbi.nlm.nih.gov/39816717/. DOI: 10.1016/j.jcot.2024.102872.
[6]
LI S H, WANG L L. Research progress of artificial intelligence in imaging evaluation of rheumatoid arthritis[J]. Chin J Magn Reson Imag, 2024, 15(8): 229-234. DOI: 10.12015/issn.1674-8034.2024.08.037.
[7]
LI S F, CAO P H, LI J, et al. Integrating radiomics and neural networks for knee osteoarthritis incidence prediction[J]. Arthritis Rheumatol, 2024, 76(9): 1377-1386. DOI: 10.1002/art.42915.
[8]
FEIERABEND M, WOLFGART J M, PRASTER M, et al. Applications of machine learning and deep learning in musculoskeletal medicine: a narrative review[J/OL]. Eur J Med Res, 2025, 30(1): 386 [2025-06-12]. https://pubmed.ncbi.nlm.nih.gov/40375335/. DOI: 10.1186/s40001-025-02511-9.
[9]
NOZAKI T, HASHIMOTO M, UEDA D, et al. Recent topics in musculoskeletal imaging focused on clinical applications of AI: How should radiologists approach and use AI [J]. Radiol Med, 2025, 130(5): 587-597. DOI: 10.1007/s11547-024-01947-z.
[10]
ALETAHA D, NEOGI T, SILMAN A J, et al. 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative[J]. Arthritis Rheum, 2010, 62(9): 2569-2581. DOI: 10.1002/art.27584.
[11]
Chinese Medical Association Orthopaedic Surgery Branch. Guidelines for diagnosis and treatment of osteoarthritis (2007 edition)[J]. Chin J Clin, 2008, 36(1): 28-30. DOI: 10.3760/j.issn:0253-2352.2007.10.016.
[12]
HUNTER D J, GUERMAZI A, LO G H, et al. Evolution of semi-quantitative whole joint assessment of knee OA: MOAKS (MRI Osteoarthritis Knee Score)[J]. Osteoarthritis Cartilage, 2011, 19(8): 990-1002. DOI: 10.1016/j.joca.2011.05.004.
[13]
ØSTERGAARD M, PETERFY C G, BIRD P, et al. The OMERACT rheumatoid arthritis magnetic resonance imaging (MRI) scoring system: updated recommendations by the OMERACT MRI in arthritis working group[J]. J Rheumatol, 2017, 44(11): 1706-1712. DOI: 10.3899/jrheum.161433.
[14]
TAN H, KANG W L, FAN Q J, et al. Intravoxel incoherent motion diffusion-weighted MR imaging findings of infrapatellar fat pad signal abnormalities: comparison between symptomatic and asymptomatic knee osteoarthritis[J]. Acad Radiol, 2023, 30(7): 1374-1383. DOI: 10.1016/j.acra.2022.11.010.
[15]
SCHIRATTI J B, DUBOIS R, HERENT P, et al. A deep learning method for predicting knee osteoarthritis radiographic progression from MRI[J/OL]. Arthritis Res Ther, 2021, 23(1): 262 [2025-06-15]. https://pubmed.ncbi.nlm.nih.gov/34663440/. DOI: 10.1186/s13075-021-02634-4.
[16]
YIN R, CHEN H, TAO T Q, et al. Expanding from unilateral to bilateral: a robust deep learning-based approach for predicting radiographic osteoarthritis progression[J]. Osteoarthritis Cartilage, 2024, 32(3): 338-347. DOI: 10.1016/j.joca.2023.11.022.
[17]
FOLLE L, BAYAT S, KLEYER A, et al. Advanced neural networks for classification of MRI in psoriatic arthritis, seronegative, and seropositive rheumatoid arthritis[J]. Rheumatology (Oxford), 2022, 61(12): 4945-4951. DOI: 10.1093/rheumatology/keac197.
[18]
WANG Q Z, YAO M Y, SONG X H, et al. Automated segmentation and classification of knee synovitis based on MRI using deep learning[J]. Acad Radiol, 2024, 31(4): 1518-1527. DOI: 10.1016/j.acra.2023.10.036.
[19]
GAO J, SONG S. A hierarchical feature extraction and multimodal deep feature integration-based model for autism spectrum disorder identification[J]. IEEE J Biomed Health Inform, 2025, 29(7): 4920-4931. DOI: 10.1109/jbhi.2025.3540894.
[20]
LI Y H, EL HABIB DAHO M, CONZE P H, et al. A review of deep learning-based information fusion techniques for multimodal medical image classification[J/OL]. Comput Biol Med, 2024, 177: 108635 [2025-05-17]. https://pubmed.ncbi.nlm.nih.gov/38796881/. DOI: 10.1016/j.compbiomed.2024.108635.
[21]
SUI J, ZHI D M, CALHOUN V D. Data-driven multimodal fusion: approaches and applications in psychiatric research[J/OL]. Psychoradiology, 2023, 3: kkad026 [2025-05-17]. https://pubmed.ncbi.nlm.nih.gov/38143530/. DOI: 10.1093/psyrad/kkad026.
[22]
ALI G, ANWAR M, NAUMAN M, et al. Lyme rashes disease classification using deep feature fusion technique[J/OL]. Skin Res Technol, 2023, 29(11): e13519 [2025-05-17]. https://pubmed.ncbi.nlm.nih.gov/38009027/. DOI: 10.1111/srt.13519.
[23]
CAO C, SONG J, SU R, et al. Structure-constrained deep feature fusion for chronic otitis media and cholesteatoma identification[J/OL]. Multimed Tools Appl, 2023: 1-21 [2025-05-17]. https://pubmed.ncbi.nlm.nih.gov/37362730/. DOI: 10.1007/s11042-023-15425-7.
[24]
GUPTA P, JAIN N. Segmentation-based fusion of CT and MR images[J]. J Imaging Inform Med, 2024, 37(5): 2635-2648. DOI: 10.1007/s10278-024-01078-x.
[25]
HU J P, ZHENG C Y, YU Q L, et al. DeepKOA: a deep-learning model for predicting progression in knee osteoarthritis using multimodal magnetic resonance images from the osteoarthritis initiative[J]. Quant Imaging Med Surg, 2023, 13(8): 4852-4866. DOI: 10.21037/qims-22-1251.
[26]
SONG L W, LI C P, TAN L L, et al. A deep learning model to enhance the classification of primary bone tumors based on incomplete multimodal images in X-ray, CT, and MRI[J/OL]. Cancer Imaging, 2024, 24(1): 135 [2025-05-05]. https://pubmed.ncbi.nlm.nih.gov/39390604/. DOI: 10.1186/s40644-024-00784-7.
[27]
ZHANG D, LI Y N, LI C L, et al. Multimodal radiomics and deep learning models for predicting early femoral head deformity in LCPD[J/OL]. Eur J Radiol, 2024, 181: 111793 [2020-05-05]. https://pubmed.ncbi.nlm.nih.gov/39454426/. DOI: 10.1016/j.ejrad.2024.111793.
[28]
IQBAL S, QURESHI A N, ALHUSSEIN M, et al. Hybrid deep spatial and statistical feature fusion for accurate MRI brain tumor classification[J/OL]. Front Comput Neurosci, 2024, 18: 1423051 [2025-05-17]. https://pubmed.ncbi.nlm.nih.gov/38978524/. DOI: 10.3389/fncom.2024.1423051.
[29]
YOU S H, CHO Y, KIM B, et al. Synthetic temporal bone CT generation from UTE-MRI using a cycleGAN-based deep learning model: advancing beyond CT-MR imaging fusion[J]. Eur Radiol, 2025, 35(1): 38-48. DOI: 10.1007/s00330-024-10967-2.
[30]
HUMBERT-VIDAN L, PATEL V, KING A P, et al. Comparison of deep-learning multimodality data fusion strategies in mandibular osteoradionecrosis NTCP modelling using clinical variables and radiation dose distribution volumes[J/OL]. Phys Med Biol, 2024, 69(20) [2025-05-10]. https://pubmed.ncbi.nlm.nih.gov/39357529/. DOI: 10.1088/1361-6560/ad8290.

PREV Delta MRI-based radiomics for predicting risk factors in cervical cancer patients after neoadjuvant chemotherapy
NEXT Experimental study on diagnosis of metabolic dysfunction-associated steatohepatitis using different models of multi-b-value diffusion-weighted magnetic resonance imaging
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn