基于多模态影像构建强直性脊柱炎个体化疗效预测的融合模型研究

Ankylosing spondylitis (AS) is chronic progress disease with the high rate of the residual. AS is clinically treated with long-acting biologics, such as tumor necrosis factor-alpha (TNF-α) inhibitors. However, not only the cost is expensive, but also the incidence of potential side effects is high; whereas curative effects are low in some patients. Hence, accurate prediction of curative effects can effectively make clinical decision to choose a best therapeutic regimen before AS treatment. Currently, the method for the prediction of curative effects on AS is bad based on clinical indicators, laboratory and imaging methods. The research of my team confirmed that new technologies on magnetic resonance (MR) and computed tomography (CT) could make it possible to diagnose AS in the early stages, to detect AS activity with high sensitivity and specificity, and even to predict curative effects on AS. The biological characteristics of AS can be fully characterized by the multimodal imaging methods, such as MR, CT, and X-ray. The construction of fusion model can provide new ideas for the prediction of curative effects on AS based on these methods. This proposal aims to screen the multimodality imaging features having the high correlation with the efficacy on AS, quantify molecular markers linked to the efficacy on AS, integrate imaging features, biochemical molecular markers, and clinical indicators to construct an individual quantitative fusion model for predicting the efficacy of TNF-α inhibitors on AS. The results of this study will provide clinicians with decision-making basis.

强直性脊柱炎(AS)是慢性进展性疾病，其致残率高。临床多采取长效生物制剂-肿瘤坏死因子(TNF-α)抑制剂治疗，但费用昂贵、潜在副作用大，且部分病人疗效欠佳。AS治疗前准确预测药物疗效可指导其治疗方案的选择。目前，基于临床信息、实验室与影像检查的评估方法尚不能对AS疗效进行准确预测。申请人团队的前期研究证实：基于MR和CT新技术对AS进行早期诊断和活动性检测，不但灵敏度高、特异性强，而且可评估药物疗效。MR、CT和X线等多模态影像方法结合融合模型构建既可全面表征AS的生物学特征，又能预测AS的疗效。本项目拟：筛选与AS疗效相关联的MR、CT和X线多模态影像学特征；量化与AS疗效相关的分子标志物；融合影像学特征、临床和生化指标，构建量化的融合模型，更加准确地预测TNF-α抑制剂对AS的疗效，为临床医师提供决策依据。

强直性脊柱炎(AS)是以骶骼关节和脊柱附着点炎症为主的炎性疾病。研究发现以肿瘤坏死因子α （TNF-α）抑制剂为主的生物制剂可达到治疗AS的目的，然而目前临床上不能对AS治疗疗效准确预测。亟需发展新的预测方法，达到有效预测疗效、指导临床决策的目的。.围绕这一临床需求，本项目研发了疗效预测融合模型，成功筛选出与疗效相关的影像特征，并融合AS患者的临床和生化指标，可在治疗前实现对AS个体化疗效预测。项目组顺利完成了本项目的研究内容，取得了相应的研究成果并撰写及发表多篇论文。培养了医工交叉方向的博士研究生1名，硕士研究生4名，正高级职称人员1名。本项目主要从以下几个方面展开研究: .（1）为了提高强直性脊柱炎的诊断准确性，本项目构建深度学习模型自动化诊断强直性脊柱炎。此外，针对目前临床上HLA-B27阴性的中轴型脊柱关节炎（axSpA）患者诊断较为困难的问题，本项目构建基于迁移学习和数据增强的模型可以准确诊断此类患者。我们的模型比临床医生的诊断性能更好，有望为临床上axSpA诊断提供更有效的工具。.（2）为了筛选与疗效相关的影像特征用于疗效预测，项目组进行了大量的探索研究。前瞻性探讨axSpA患者骶髂关节炎的体素内不相干运动(IVIM)扩散加权成像(DWI)和动态对比增强MRI参数之间的关系。并且开发出自动化检测axSpA患者骨侵蚀及骶髂关节脂肪化生病变的深度学习模型，为医生观察axSpA患者骶髂关节的结构损伤提供了有力的工具。.（3）围绕axSpA患者的疗效预测，本项目开展医工交叉合作，确定基于MRI和临床指标的潜在预测因素，建立了一个稳健的模型可以预测1年内累及髋关节的axSpA患者的治疗反应。本项目同时评估了构建的模型在四种不同疗效评估标准下的预测性能。模型对axSpA患者的临床决策制定有较好的指导价值，为患者的个性化治疗提供了新依据。.综上所述，项目组顺利完成了预期研究任务，利用融合模型提高了预测强直性脊柱炎患者治疗疗效的效能，有望使广大强直性脊柱炎患者从中获益。.