Share：
Share this content in WeChat
X
[Chinese][PDF]6433
• Clinical Article •
A self-attention-based deep learning model predicts the progression of new bone formation in the sacroiliac joints of patients with axial spondylarthritis on MRI
LI Yi  SONG Liwen  ZHAO Yinghua 

Cite this article as LI Y, SONG L W, ZHAO Y H. A self-attention-based deep learning model predicts the progression of new bone formation in the sacroiliac joints of patients with axial spondylarthritis on MRI[J]. Chin J Magn Reson Imaging, 2024, 15(5): 154-161. DOI:10.12015/issn.1674-8034.2024.05.024.


[Abstract] Objective To investigate the clinical significance of deep learning (DL) model based on self-attention mechanisms in predicting the progression of new bone formation on coronal T1-weighted MR images of sacroiliac joints (SIJ) in patients with axial spondylarthritis (axSpA).Materials and Methods We conducted a retrospective analysis of MRI data (with one-year, two-year or three-year follow-up duration) for 351 axSpA patients who were diagnosed at the Third Affiliated Hospital of Southern Medical University from January 2010 to December 2022. The patients were randomly allocated into training, validation, and test sets in a 8∶1∶1 ratio. The Bifpn-YOLOv8 model based on self-attention mechanisms was developed. And another three baseline models (YOLOv8, YOLOv7, Faster-RCNN) were constructed to compare model performance with Bifpn-YOLOv8. We evaluated the predictive performance of each model using metrics such as mean average precision (mAP), F1 score, accuracy, recall, and Common Objects in Context (COCO) evaluation metrics. Among them, mAP50 and mAP50:95 indicates the mean average precision at different intersection over union thresholds, respectively. The average precision (AP) of COCO metrics, such as AP, AP50, AP75, follows the same principle.Results The Bifpn-YOLOv8 model exhibited good predictive performance on both validation and test sets. In comparison to the baseline models, Bifpn-YOLOv8 achieved the highest mAP50 and mAP50:95 on the test set, with values of 83.8% and 50.4%, respectively. The results were statistically significant (all P<0.05) compared to each baseline model. Similarly, the Bifpn-YOLOv8 model outperformed the baseline models on the test set with superior COCO evaluation metrics (AP: 50.5%, AP50: 82.3%, AP75: 58.6%).Conclusions The self-attention-based Bifpn-YOLOv8 model could effectively predicting the progression of new bone formation in the SIJ on MR images of axSpA patients. This model is poised to become a valuable clinical tool for evaluating the progression of new bone formation, providing assistance to physicians in clinical decision-making and management of axSpA patients.
[Keywords] axial spondylarthritis；sacroiliac joint；self-attention；magnetic resonance imaging；deep learning

LI Yi   SONG Liwen   ZHAO Yinghua*  

Department of Radiology, The Third Affiliated Hospital of Southern Medical University, Guangzhou 510630, China

Corresponding author: ZHAO Y H, E-mail: zhaoyh@smu.edu.cn

Conflicts of interest   None.

Received  2024-01-26
Accepted  2024-04-30
DOI: 10.12015/issn.1674-8034.2024.05.024
Cite this article as LI Y, SONG L W, ZHAO Y H. A self-attention-based deep learning model predicts the progression of new bone formation in the sacroiliac joints of patients with axial spondylarthritis on MRI[J]. Chin J Magn Reson Imaging, 2024, 15(5): 154-161. DOI:10.12015/issn.1674-8034.2024.05.024.

[1]
SIEPER J, PODDUBNYY D. Axial spondyloarthritis[J]. Lancet, 2017, 390(10089): 73-84. DOI: 10.1016/s0140-6736(16)31591-4.
[2]
PODDUBNYY D, SIEPER J. Mechanism of new bone formation in axial spondyloarthritis[J/OL]. Curr Rheumatol Rep, 2017, 19(9): 55 [2023-12-12]. https://pubmed.ncbi.nlm.nih.gov/28752489/. DOI: 10.1007/s11926-017-0681-5.
[3]
SHENG H Q, ZHAO B, GENG L. Abnormalities of sacroiliac joints in ankylosing spondylitis: comparative study of MR imaging and spiral CT[J]. J Clin Radiol, 2008, 27(8): 1091-1094. DOI: 10.3969/j.issn.1001-9324.2008.08.024.
[4]
LALOO F, HERREGODS N, JAREMKO J L, et al. MRI of the axial skeleton in spondyloarthritis: the many faces of new bone formation[J/OL]. Insights Imaging, 2019, 10(1): 67 [2023-12-12]. https://pubmed.ncbi.nlm.nih.gov/31338670/. DOI: 10.1186/s13244-019-0752-4.
[5]
YANG X X, TIAN X P, LI M T, et al. Interpretation on the consensus on targeted drug therapy for spondyloarthritis[J]. Med J Peking Union Med Coll Hosp, 2024, 15(1): 58-67. DOI: 10.12290/xhyxzz.2023-0391.
[6]
WARD M M, DEODHAR A, AKL E A, et al. American college of rheumatology/spondylitis association of america/spondyloarthritis research and treatment network 2015 recommendations for the treatment of ankylosing spondylitis and nonradiographic axial spondyloarthritis[J]. Arthritis Rheumatol, 2016, 68(2): 282-298. DOI: 10.1002/art.39298.
[7]
ZHANG J R, LIU X J, XU W D, et al. Effects of tumor necrosis factor-α inhibitors on new bone formation in ankylosing spondylitis[J]. Joint Bone Spine, 2016, 83(3): 257-264. DOI: 10.1016/j.jbspin.2015.06.013.
[8]
BRAUN J, BARALIAKOS X, HERMANN K G, et al. Golimumab reduces spinal inflammation in ankylosing spondylitis: MRI results of the randomised, placebo- controlled GO-RAISE study[J]. Ann Rheum Dis, 2012, 71(6): 878-884. DOI: 10.1136/annrheumdis-2011-200308.
[9]
BRAUN J, BARALIAKOS X, GOLDER W, et al. Magnetic resonance imaging examinations of the spine in patients with ankylosing spondylitis, before and after successful therapy with infliximab: evaluation of a new scoring system[J]. Arthritis Rheum, 2003, 48(4): 1126-1136. DOI: 10.1002/art.10883.
[10]
PEDERSEN S J, WICHUK S, CHIOWCHANWISAWAKIT P, et al. Tumor necrosis factor inhibitor therapy but not standard therapy is associated with resolution of erosion in the sacroiliac joints of patients with axial spondyloarthritis[J/OL]. Arthritis Res Ther, 2014, 16(2): R100 [2023-12-14]. https://pubmed.ncbi.nlm.nih.gov/24755322/. DOI: 10.1186/ar4548.
[11]
LAMBERT R G, SALONEN D, RAHMAN P, et al. Adalimumab significantly reduces both spinal and sacroiliac joint inflammation in patients with ankylosing spondylitis: a multicenter, randomized, double-blind, placebo-controlled study[J]. Arthritis Rheum, 2007, 56(12): 4005-4014. DOI: 10.1002/art.23044.
[12]
PEDERSEN S J, WEBER U, SAID-NAHAL R, et al. Structural progression rate decreases over time on serial radiography and magnetic resonance imaging of sacroiliac joints and spine in a five-year follow-up study of patients with ankylosing spondylitis treated with tumour necrosis factor inhibitor[J]. Scand J Rheumatol, 2019, 48(3): 185-197. DOI: 10.1080/03009742.2018.1506822.
[13]
VAN DER HEIJDE D, RAMIRO S, LANDEWÉ R, et al. 2016 update of the ASAS-EULAR management recommendations for axial spondyloarthritis[J]. Ann Rheum Dis, 2017, 76(6): 978-991. DOI: 10.1136/annrheumdis-2016-210770.
[14]
LI K J, JOIS R, LICHAUCO J J, et al. A review on the effect of tumor necrosis factor inhibitors on structural progression in early axial spondyloarthritis using magnetic resonance imaging[J]. Rheumatol Ther, 2019, 6(2): 139-163. DOI: 10.1007/s40744-019-0141-y.
[15]
MAKSYMOWYCH W P, WICHUK S, CHIOWCHANWISAWAKIT P, et al. Fat metaplasia and backfill are key intermediaries in the development of sacroiliac joint ankylosis in patients with ankylosing spondylitis[J]. Arthritis Rheumatol, 2014, 66(11): 2958-2967. DOI: 10.1002/art.38792.
[16]
ZHANG Y M, GUO Z K, WANG L S, et al. Baseline severity of sacroiliitis and extensive fat Metaplasia predicts the progression of backfill at the sacroiliac joint in patients with axial spondyloarthritis[J/OL]. Front Immunol, 2022, 13: 887470 [2023-12-14]. https://pubmed.ncbi.nlm.nih.gov/35833133/. DOI: 10.3389/fimmu.2022.887470.
[17]
MAKSYMOWYCH W P, WICHUK S, CHIOWCHANWISAWAKIT P, et al. Fat metaplasia on MRI of the sacroiliac joints increases the propensity for disease progression in the spine of patients with spondyloarthritis[J/OL]. RMD Open, 2017, 3(1): e000399 [2023-12-14]. https://pubmed.ncbi.nlm.nih.gov/28469937/. DOI: 10.1136/rmdopen-2016-000399.
[18]
VAN MECHELEN M, GULINO G R, DE VLAM K, et al. Bone disease in axial spondyloarthritis[J]. Calcif Tissue Int, 2018, 102(5): 547-558. DOI: 10.1007/s00223-017-0356-2.
[19]
BARALIAKOS X, LISTING J, VON DER RECKE A, et al. The natural course of radiographic progression in ankylosing spondylitis: evidence for major individual variations in a large proportion of patients[J]. J Rheumatol, 2009, 36(5): 997-1002. DOI: 10.3899/jrheum.080871.
[20]
BRESSEM K K, ADAMS L C, PROFT F, et al. Deep learning detects changes indicative of axial spondyloarthritis at MRI of sacroiliac joints[J]. Radiology, 2022, 305(3): 655-665. DOI: 10.1148/radiol.212526.
[21]
NICOARA A I, SAS L M, BITA C E, et al. Implementation of artificial intelligence models in magnetic resonance imaging with focus on diagnosis of rheumatoid arthritis and axial spondyloarthritis: narrative review[J/OL]. Front Med, 2023, 10: 1280266 [2023-12-14]. https://pubmed.ncbi.nlm.nih.gov/38173943/. DOI: 10.3389/fmed.2023.1280266.
[22]
SIEPER J, RUDWALEIT M, BARALIAKOS X, et al. The Assessment of SpondyloArthritis international Society (ASAS) handbook: a guide to assess spondyloarthritis[J]. Ann Rheum Dis, 2009, 68(Suppl 2): ii1-ii44. DOI: 10.1136/ard.2008.104018.
[23]
MAKSYMOWYCH W P, WICHUK S, CHIOWCHANWISAWAKIT P, et al. Development and preliminary validation of the spondyloarthritis research consortium of Canada magnetic resonance imaging sacroiliac joint structural score[J]. J Rheumatol, 2015, 42(1): 79-86. DOI: 10.3899/jrheum.140519.
[24]
ZHANG Y M, GUO Z K, ZHAN Y, et al. Evaluation of active inflammation, chronic structural damage, and response to treatment of sacroiliitis in axial spondyloarthritis using the Spondyloarthritis research consortium of Canada scoring system[J/OL]. BMC Musculoskelet Disord, 2022, 23(1): 654 [2023-12-17]. https://pubmed.ncbi.nlm.nih.gov/35804360/. DOI: 10.1186/s12891-022-05609-x.
[25]
ZHU L, WANG X J, KE Z H, et al. BiFormer: vision transformer with Bi-level routing attention[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, BC, Canada. IEEE, 2023: 10323-10333. DOI: 10.1109/CVPR52729.2023.00995.
[26]
TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, WA, USA. IEEE, 2020: 10778-10787. DOI: 10.1109/CVPR42600.2020.01079.
[27]
LI H L, LI J, WEI H B, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[EB/OL]. 2022 [2023-12-17]. https://arxiv.org/abs/2206.02424. DOI: 10.48550/arXiv.2206.02424.
[28]
MAKSYMOWYCH W P, MALLON C, MORROW S, et al. Development and validation of the Spondyloarthritis Research Consortium of Canada (SPARCC) Enthesitis Index[J]. Ann Rheum Dis, 2009, 68(6): 948-953. DOI: 10.1136/ard.2007.084244.
[29]
LUKAS C, BRAUN J, VAN DER HEIJDE D, et al. Scoring inflammatory activity of the spine by magnetic resonance imaging in ankylosing spondylitis: a multireader experiment[J]. J Rheumatol, 2007, 34(4): 862-870.
[30]
JONES A. Assessing disease activity and response to treatment in axial spondyloarthritis: The unmet clinical need and potential role for quantitative imaging[D]. University College London, 2022.
[31]
BORDNER A, AOUAD T, MEDINA C L, et al. A deep learning model for the diagnosis of sacroiliitis according to Assessment of SpondyloArthritis International Society classification criteria with magnetic resonance imaging[J]. Diagn Interv Imaging, 2023, 104(7/8): 373-383. DOI: 10.1016/j.diii.2023.03.008.
[32]
ZHANG N, DONAHUE J, GIRSHICK R, et al. Part-based R-CNNs for fine-grained category detection[C]//European Conference on Computer Vision. Cham: Springer, 2014: 834-849. DOI: 10.1007/978-3-319-10590-1_54
[33]
LEE K H, CHOI S T, LEE G Y, et al. Method for diagnosing the bone marrow edema of sacroiliac joint in patients with axial spondyloarthritis using magnetic resonance image analysis based on deep learning[J/OL]. Diagnostics, 2021, 11(7): 1156 [2023-12-17]. https://pubmed.ncbi.nlm.nih.gov/34202607/. DOI: 10.3390/diagnostics11071156.
[34]
SHETTY A K, SAHA I, SANGHVI R M, et al. A review: object detection models[C]//2021 6th International Conference for Convergence in Technology (I2CT). Maharashtra, India. IEEE, 2021: 1-8. DOI: 10.1109/I2CT51068.2021.9417895.
[35]
ZOU Z X, CHEN K Y, SHI Z W, et al. Object detection in 20 years: a survey[J]. Proc IEEE, 2023, 111(3): 257-276. DOI: 10.1109/JPROC.2023.3238524.
[36]
CHAUDHARI S, MITHAL V, POLATKAN G, et al. An attentive survey of attention models[J/OL]. ACM Trans Intell Syst Technol, 12(5): 53 [2023-12-17]. https://dl.acm.org/doi/10.1145/3465055. DOI: 10.1145/3465055.
[37]
LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C]//European Conference on Computer Vision. Cham: Springer, 2014: 740-755.10.1007/978-3-319-10602-1_48
[38]
MÅRTENSSON G, FERREIRA D, GRANBERG T, et al. The reliability of a deep learning model in clinical out-of-distribution MRI data: a multicohort study[J/OL]. Med Image Anal, 2020, 66: 101714 [2023-12-17]. https://pubmed.ncbi.nlm.nih.gov/33007638/. DOI: 10.1016/j.media.2020.101714.

PREV Value of synthetic MRI in differential diagnosis of benign and malignant ovarian adnexal lesions with O-RADS MRI score 3-5
NEXT A case of malignant melanotic nerve sheath tumors in the thoracic spinal canal and review of the literature
  


LINK


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