[1]刘成林,陈浩,付佳豪,等.深度学习模型在计算机断层扫描中对股骨头坏死进行自动诊断和分割坏死区域的研究[J].医学研究与战创伤救治(原医学研究生学报),2026,39(01):45-53.[doi:10.16571/j.cnki.2097-2768.2026.01.007]
　LIU Chenglin,CHEN Hao,FU Jiahao,et al.A study on automatic diagnosis and segmentation of necrotic areas of femoral head necrosis using deep learning models in computed tomography[J].JOURNAL OF MEDICALRESEARCH —COMBAT TRAUMA CARE,2026,39(01):45-53.[doi:10.16571/j.cnki.2097-2768.2026.01.007]
点击复制

深度学习模型在计算机断层扫描中对股骨头坏死进行自动诊断和分割坏死区域的研究()

《医学研究与战创伤救治》（原医学研究生学报）[ISSN:1672-271X/CN:32-1713/R]

卷:
第39卷
期数:
2026年01期
页码:
45-53
栏目:
临床研究
出版日期:
2026-01-20

文章信息/Info

Title:
A study on automatic diagnosis and segmentation of necrotic areas of femoral head necrosis using deep learning models in computed tomography
作者:
刘成林; 陈浩; 付佳豪; 吴尧昆; 刘利坤; 兰海; 宦想; 刘锌; 孙光权
南京中医药大学附属医院（江苏省中医院）骨伤科,江苏南京 210029
Author(s):
LIU Chenglin;  CHEN Hao;  FU Jiahao;  WU Yaokun;  LIU Likun;  LAN Hai;  HUAN Xiang;  LIU Xin;  SUN Guangquan
（Department of Orthopedics and Traumatology, Affiliated Hospital of Nanjing University of Chinese Medicine/Jiangsu Province Hospital of Chinese Medicine, Nanjing 210029, Jiangsu, China）
关键词:
股骨头坏死诊断; 计算机断层扫描; 坏死区域分割; 深度学习
Keywords:
osteonecrosis of the femoral head diagnosis; computed tomography; necrotic regions segmentation; deep learning
分类号:
R681.8
DOI:
10.16571/j.cnki.2097-2768.2026.01.007
文献标志码:
A
摘要:
目的旨在开发一种基于计算机断层扫描（CT）的两阶段深度学习（DL）模型，用于股骨头坏死（ONFH）的诊断及坏死区域的分割。方法回顾性分析2023年2月至2024年6月就诊于江苏省中医院、江苏省第二中医院、南京市中医院及南昌市洪都中医院行髋关节CT检查的2312例患者。由江苏省中医院、江苏省第二中医院、南京市中医院的髋关节CT数据集按8∶2的比例随机分为内部训练集和内部测试集用于模型的训练和性能测试，南昌市洪都中医院的髋关节CT数据集被作为外部测试集用于评估模型的性能。3位医师在同一标准下对ONFH进行注释和标注并将其作为参考标准。开发并验证了基于ViT和SwinT的DL模型，并将模型的性能与住院医师规范化培训医师、5年临床工作经验的住院医师、10年临床工作经验的主治医师进行比较，以评估其在临床场景中的稳健性和有效性。结果2312例患者中1189例（24 534张CT冠状位图像）被诊断为ONFH，1123例被诊断（29 294张冠状CT图像）为非ONFH。在内部测试集中，对于ONFH诊断，DL模型的受试者工作特征曲线下面积（AUC）为0.99，敏感度为99.97%，精度为91.04%，F1分数为95.31%；对于坏死病灶的分割，DL模型的戴斯系数（Dice）值为72.76%，杰卡德系数（Jaccard）值为67.19%。在外部测试集中，对于ONFH诊断，DL模型的AUC为0.97，敏感度99.94%，精度为90.04%，F1分数为94.73%；对于坏死病灶的分割，DL模型的Dice值和Jaccard值分别为69.14%和62.75%。与不同年资的骨科医师相比，DL模型对于ONFH的诊断性能最为优秀，对于坏死病灶的分割性能优于住院医师规范化培训医师和5年临床工作经验的住院医师，接近10年临床工作经验的主治医师水平。结论DL模型在ONFH诊断上展现出优异性能，可以媲美或超越骨科主治医师，实现了ONFH的精准诊断与病灶量化分割，可以为ONFH的病程监测与病情评估提供客观依据。
Abstract:
ObjectiveThe study aims to develop an automatic, computed tomography (CT)based, twostage deep learning (DL) model for the diagnosis of osteonecrosis of the femoral head (ONFH) and necrotic regions segmentation.MethodsA retrospective analysis was conducted on patients who underwent hip CT examinations between February 2023 and June 2024 at Jiangsu Province Hospital of Chinese Medicine, Jiangsu Second Provincial Hospital of Chinese Medicine, Nanjing Hospital of Chinese Medicine, and Hongdu Hospital of Chinese Medicine of Nanchang. CT datasets from the first three hospitals were randomly divided into an internal training set and an internal test set in an 8∶2 ratio for model development and performance evaluation. The dataset from Hongdu Hospital of Traditional Chinese Medicine of Nanchang served as an external test set to evaluate model performance. Annotations of ONFH lesions were performed by three physicians, following consistent criteria, and the annotations were served as the reference standard. DL models based on Vision Transformer (ViT) and Swin Transformer (SwinT) were developed and validated. Their performance was compared with that of physicians on standardized residency training, junior residents with 5 years of clinical experience, and attending physicians with 10 years of clinical experience to evaluate robustness and clinical effectiveness.ResultsAmong 2312 patients, 1189 were diagnosed with ONFH (24 534 coronal CT images), and 1123 were nonONFH cases (29 294 coronal CT images). On the internal test set, for ONFH diagnosis, the DL model achieved an area under the receiver operating characteristic curve (AUC) of 0.99, a sensitivity of 99.97%, a precision of 91.04%, and an F1score of 95.31%. For necrotic lesion segmentation, the model attained a Dice similarity coefficient of 72.76% and a Jaccard index of 67.19%. On the external test set, the DL model achieved an AUC of 0.97, a sensitivity of 99.94%, a precision of 90.04%, an F1score of 94.73%, a Dice coefficient of 69.14%, and a Jaccard index of 62.75%. Compared to orthopedic physicians of different experience levels, the DL model demonstrated superior diagnostic performance for ONFH. Its segmentation performance exceeded that of training residents and junior residents, and was comparable to that of attending physicians with 10 years of experience.ConclusionThe DL model demonstrated excellent performance in diagnosing ONFH, comparable or even surpassing that of attending orthopedic physicians. It enables accurate diagnosis and quantitative segmentation of ONFH lesions, providing an objective tool for disease monitoring and clinical assessment.

参考文献/References:

[1]Powell C, Chang C, Gershwin ME. Current concepts on the pathogenesis and natural history of steroidinduced osteonecrosis[J]. Clin Rev Allergy Immunol, 2011, 41(1): 102113.[2]Chen Y, Miao Y, Liu K, et al. Evolutionary course of the femoral head osteonecrosis: Histopathologicalradiologic characteristics and clinical staging systems[J]. J Orthop Translat， 2022, 32: 2840.[3]陈浩，杜斌，陈阳，等. 头颈开窗打压植骨联合缝匠肌骨瓣移植术治疗ARCO Ⅱ期股骨头坏死的疗效分析[J]. 医学研究生学报, 2021, 34(12): 12671272.[4]Cooper C, Steinbuch M, Stevenson R, et al. The epidemiology of osteonecrosis: findings from the GPRD and THIN databases in the UK[J]. Osteoporos Int， 2010, 21(4): 569577.[5]Kang JS, Park S, Song JH, et al. Prevalence of osteonecrosis of the femoral head: a nationwide epidemiologic analysis in Korea[J]. J Arthroplasty， 2009, 24(8): 11781183.[6]Zhao D W, Yu M, Hu K, et al. Prevalence of Nontraumatic Osteonecrosis of the Femoral Head and its Associated Risk Factors in the Chinese Population: Results from a Nationally Representative Survey[J]. Chin Med J (Engl)， 2015, 128(21): 28432850.[7]Sato R, Ando W, Fukushima W, et al. Epidemiological study of osteonecrosis of the femoral head using the national registry of designated intractable diseases in Japan[J]. Mod Rheumatol， 2022, 32(4): 808814.[8]Zhu Z, Li S, Yu H, et al. Correlation between continuation of glucocorticoid treatment and risk of femoral head collapse : A retrospective cohort study of patients with glucocorticoidinduced osteonecrosis of femoral head after hippreserving interventions[J]. Orthopade， 2021, 50(2): 143149.[9]Yoon BH, Kim TY, Shin IS, et al. Alcohol intake and the risk of osteonecrosis of the femoral head in Japanese populations: a doseresponse metaanalysis of casecontrol studies[J]. Clin Rheumatol， 2017, 36(11): 25172524.[10]Zalavras CG, Lieberman JR. Osteonecrosis of the femoral head: evaluation and treatment[J]. J Am Acad Orthop Surg， 2014, 22(7): 455464.[11]GoslingGardeniers AC, Rijnen WHC, Gardeniers JWM. The Prevalence of Osteonecrosis in Different Parts of the World[M].Berlin Heidelberg：Springer, 2014:3537.[12]Hungerford DS. Osteonecrosis: avoiding total hip arthroplasty[J]. J Arthroplasty， 2002, 17(4 Suppl 1): 121124.[13]Zhao DW, Hu YC. Chinese experts′ consensus on the diagnosis and treatment of osteonecrosis of the femoral head in adults[J]. Orthop Surg， 2012, 4(3): 125130.[14]Banerjee S, Issa K, Pivec R, et al. Osteonecrosis of the hip: treatment options and outcomes[J]. Orthop Clin North Am，2013, 44(4): 463476.[15]Garvin KL, Miller RE, Gilbert TM, et al. Late Reinfection May Recur More Than 5 Years After Reimplantation of THA and TKA: Analysis of Pathogen Factors[J]. Clin Orthop Relat Res， 2018, 476(2): 345352.[16]Zhao D, Zhang F, Wang B, et al. Guidelines for clinical diagnosis and treatment of osteonecrosis of the femoral head in adults (2019 version)[J]. J Orthop Translat， 2020, 21: 100110.[17]Migliorini F, Maffulli N, Baroncini A, et al. Failure and progression to total hip arthroplasty among the treatments for femoral head osteonecrosis: a Bayesian network metaanalysis[J]. Br Med Bull， 2021, 138(1): 112125.[18]Kuroda Y, Tanaka T, Miyagawa T, et al. Classification of osteonecrosis of the femoral head: Who should have surgery?[J]Bone Joint Res， 2019, 8(10): 451458.[19]Shen X, He Z, Shi Y, et al. Automatic detection of early osteonecrosis of the femoral head from various hip pathologies using deep convolutional neural network: a multicentre study[J]. Int Orthop， 2023, 47(9): 22352244.[20]Shen X, Luo J, Tang X, et al. Deep Learning Approach for Diagnosing Early Osteonecrosis of the Femoral Head Based on Magnetic Resonance Imaging[J]. J Arthroplasty， 2023, 38(10): 20442050.[21]Ruckli AC, Nanavati AK, Meier MK, et al. A Deep Learning Method for Quantification of Femoral Head Necrosis Based on Routine Hip MRI for Improved Surgical Decision Making[J]. J Pers Med， 2023, 13(1)：153.[22]Zhu L, Han J, Guo R, et al. An Automatic Classification of the Early Osteonecrosis of Femoral Head with Deep Learning[J]. Curr Med Imaging， 2020, 16(10): 13231331.[23]Kim B, Lee GY, Park SH. Attention fusion network with selfsupervised learning for staging of osteonecrosis of the femoral head (ONFH) using multiple MR protocols[J]. Med Phys， 2023, 50(9): 55285540.[24]Shen X, He Z, Shi Y, et al. Development and Validation of an Automated Classification System for Osteonecrosis of the Femoral Head Using Deep Learning Approach: A Multicenter Study[J]. J Arthroplasty， 2024, 39(2): 379386.[25]Sustersic T, Rankovic V, Milovanovic V, et al. A Deep Learning Model for Automatic Detection and Classification of Disc Herniation in Magnetic Resonance Images[J]. IEEE J Biomed Health Inform， 2022, 26(12): 60366046.[26]Wang P, Liu X, Xu J, et al. Deep learning for diagnosing osteonecrosis of the femoral head based on magnetic resonance imaging[J]. Comput Methods Programs Biomed， 2021, 208: 106229.[27]Liu Y, She GR, Chen SX. Magnetic resonance image diagnosis of femoral head necrosis based on ResNet18 network[J]. Comput Methods Programs Biomed， 2021, 208: 106254.[28]Li X, Lv S, Tong C, et al. MsgeCNN: Multiscale geometric embedded convolutional neural network for ONFH segmentation and grading[J]. Med Phys， 2023, 50(6): 37883800.[29]Zhang J, Guo S, Yu D, et al. Subtractingadding strategy for necrotic lesion segmentation in osteonecrosis of the femoral head[J]. Int J Comput Assist Radiol Surg， 2024, 19(5): 961970.[30]Liu M, MartinGomez A, Oni JK, et al. Towards Visualizing Earlystage Osteonecrosis using Intraoperative Imaging Modalities[J]. Comput Methods Biomech Biomed Eng Imaging Vis， 2023, 11(4): 12341242.[31]袁鑫玮，黄艺轩，席洪钟，等. 坏死及支撑区三维空间分布对腓骨支撑保髋结局的影响[J]. 中国组织工程研究， 2024, 28(18): 27892794.[32]Huang Y, Gu C, Xi H, et al. Sclerosis rim volume ratio and Japanese Investigation Committee classification as predictive factors for femoral head collapse progression after nonvascularized fibular grafting in osteonecrosis patients[J]. Asian J Surg， 2024, 47(1): 250255.[33]Yixuan H, Xinwei Y, Feifei G, et al. Effect of Sclerosis Bands in Femoral Head Necrosis on NonVascularized Fibular GraftingA Finite Element Study[J]. Orthop Surg， 2024, 16(10): 25262538.[34]Xue P, Xi H, Chen H, et al. Predictive value of clinical features and CT radiomics in the efficacy of hip preservation surgery with fibula allograft[J]. J Orthop Surg Res， 2023, 18(1): 940.[35]Chen H, Xue P, Xi H, et al. Predicting efficacy and guiding procedure choice in nonvascularized bone grafting: a CT Radiomics and clinical predictor approach[J]. BMC Musculoskelet Disord， 2023, 24(1): 959.[36]卢光明，张志强. 人工智能医学影像[J]. 医学研究生学报, 2018, 31(7): 683687.[37]Dosovitskiy A, Beyer L, Kolesnikov A, et al. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale[C]. Proceedings of the International Conference on Learning Representations (ICLR), Virtual, 20210603.[38]Liu Z, Lin Y, Cao Y, et al. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows[C]. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), MONTREAL, 20211017.[39]沈先月. 基于深度学习技术的股骨头缺血性坏死智能诊断和图像分割研究[D]. 长春：吉林大学, 2023.[40]Han K, Wang Y, Chen H, et al. A Survey on Vision Transformer[J]. IEEE Trans Pattern Anal Mach Intell， 2023, 45(1): 87110.[41]Chen H, Cao X, Zhang X, et al. Automatic segmentation framework of XRay tomography data for multiphase rock using Swin Transformer approach[J]. Sci Data， 2023, 10(1): 812.[42]Zhang D, Wang C, Chen T, et al. Scalable Swin Transformer network for brain tumor segmentation from incomplete MRI modalities[J]. Artif Intell Med， 2024, 149: 102788.

相似文献/References:

[1]刘怡男,范学武,田龙.两种耐高压型经外周静脉穿刺置入中心静脉导管尖端定位技术临床对照研究[J].医学研究与战创伤救治(原医学研究生学报),2022,24(6):586.[doi:10.3969/j.issn.1672-271X.2022.06.006]
　LIU Yi-nan,FAN Xue-wu,TIAN Long.Clinical comparative study of two high pressure resistant peripherally inserted central catheter tip positioning techniques[J].JOURNAL OF MEDICALRESEARCH —COMBAT TRAUMA CARE,2022,24(01):586.[doi:10.3969/j.issn.1672-271X.2022.06.006]

备注/Memo

备注/Memo:
基金项目：国家自然科学基金（82474342）；江苏省中医药管理局面上项目（MS2023023）
更新日期/Last Update: 2026-01-20