基于人工智能青少年特发性脊柱侧凸进展预测系统的研发

Adolescent idiopathic scoliosis (AIS) is the most common type of scoliosis, which might cause adverse consequences to its patients. Three main factors are believed to be associated with its progressions: image characteristics, bone maturity and genetic characteristics. Image characteristics and bone maturity assessments primarily rely on judgment of image materials. However, due to limitations on differentiation of details through traditional way of eye-looking, mis-judgment might occur. In addition, experienced experts in AIS progression seem to be under-manpowered in face of large quantities of patients due to abundant information required for asssessments. In recent years, the implementation of computer aided diagnosis has significantly elevated diagnosis efficiency and accuracy in multiple kinds of diseases, but with few involvements regarding to image diagnosis of AIS. The current study proposes a novel way of progression prediction for AIS via assessments of patients’ materials in combination with characteristic evaluation through extraction of information from scoliosis and bone maturity images. Information is obtained by utilizing multi-dimensional characteristics of medical images in multiple information assessment at one time and in continuous time. The results of our study would aid promotion of medical efficiency, discovery of more reliable assessment methods, optimization of treatment strategies and improve overall prognosis. The current study aims at making breakthroughs in both theoratical and technical aspects in terms of data processing, characteristic extraction and validation and intelligent diagnosis analysis in realm of AIS medical imaging.

青少年特发性脊柱侧凸是最常见的脊柱侧凸类型，可能会对患者造成严重不良后果。目前与脊柱侧凸进展相关的主要因素包括影像学特征、骨骼成熟度及遗传学特征三大类。影像学特征及骨骼成熟度主要依赖对影像资料的阅读判断。传统的人工辨识由于人类细节辨识能力的局限，容易导致判断失误。另外，由于所涉及信息较多，经验丰富的专家有限，面对庞大且散在的患者群体明显人力不足。近年来，计算机辅助诊断技术的出现显著提高了医疗领域多种疾病的诊断效率及准确度，但在脊柱侧凸的医学影像诊疗方面涉及很少。本研究利用医学影像多重信息及连续时间的多时空特征，提取脊柱侧凸图像及骨骼成熟度相关信息，进行特征评估，结合患者临床资料，对患者预后进行预测评估。研究结果将有助于医生提高医疗效率，探索更合理的预测方法，优化的治疗方案，改善患者的预后。本研究旨在脊柱侧凸领域医学数据处理方法、特征提取与验证、智能诊断分析方面寻求理论和技术上的突破。

青少年特发性脊柱侧凸（Adolescent Idiopathic Scoliosis, AIS）是最常见的侧凸类型，AIS的预后转归不尽相同，治疗策略更是因人而异，总体上来说，治疗方法包括保守治疗（观察、锻炼或支具）及手术治疗。治疗方式的选择与侧弯是否可能进展及进展程度直接相关。但由于脊柱侧凸进展影响因素较多，相互间关系复杂，而且受到人工肉眼辨识及信息处理的局限，人工分析难以提取深层的图像特征，当面对连续的海量图像以及数据，依靠个体难以发现其中蕴藏的联系。如果能够利用计算机辅助诊断患者病情，普及AIS的筛查及科学指导，将极大地缓解医疗专家的压力，提高诊断效率，不但利于患者早期发现并及时控制病情，而且具有重要的经济价值和社会效益。.本课题的目的是在搜集患者临床数据和影像学资料的基础上，对患者画像进行构建。首先对脊柱侧凸图像数据进行预处理，之后构造卷积神经网络中基于流形学习的可变核，构建能够同时表示时序性和空间信息的深度学习网络，并与专家知识进行结合，利用医学图像进行脊柱侧凸进展的早期预测。.本研究中模型在测试集中预测的灵敏度最高达89.80%，特异度最高达81.42%，曲线下面积为0.9104，对患者脊柱侧凸最大Cobb角的年化进展率的预测值与实际值的相关系数R=0.84 (p<0.05)。