Application of Novel Nonlinear Data Modelling and Analysis to the Study of Cervical Impedance Spectroscopy for Preterm Birth Prediction

每年，全球约有1500万婴儿在37周前出生。早产（PTB）并发症是5岁以下儿童死亡的主要原因，全球每年造成100万人死亡。PTB每年花费英国NHS超过10亿英镑，比足月婴儿高出10倍。由于结核病是一个重大的公共卫生问题，对社会有深远的影响，能够识别妇女在怀孕期间有结核病的风险是至关重要的，以便采取护理措施，以推迟出生，以减少潜在的长期残疾和损害。然而，为预防和减轻风险而准确查明妇女的工作仍然是一种幻想。目前筛查方法的根本问题是，它们无法评估和量化宫颈组织成分，也无法区分与PTB相关的各种临床状况。 对于PTB的发生，宫颈必须通过一系列分子水平的重塑事件软化和扩张。基于这一观察结果，由谢菲尔德的Dilly OC Anumba教授领导的一个医生和科学家小组发现，PTB高危女性在怀孕中期的子宫颈阻力低于足月分娩的女性。这一发现激发了MRC ECCLIPPxTM项目，该项目成功研究了谢菲尔德Mark V电阻抗谱（EIS）设备，以量化出生后的宫颈重塑，从而预测约500名妊娠患者的PTB。研究表明，在20-22周测量的EIS预测PTB（妊娠37周前出生）的灵敏度约为70%，特异性约为80%。这种有前途的性能仅基于单个导出的参数-在一些离散频率上的组织电阻-并且采用线性逻辑和常规统计分析。然而，子宫颈组织电特性实际上由在宽频率范围内的组织阻抗表示，其具有分别称为电阻和电抗的真实的实部和虚部，并且还可以涉及非线性的影响。因此，将所有EIS筛选数据参数与PTB综合关联，并考虑可能更复杂的非线性关联，可以进一步提高PTB预测的准确性。 出于这些考虑，Dilly OC Anumba教授和ZQ Lang教授建立了一个合作项目，初步采用了ZQ Lang教授团队开发的数据建模和模型分析技术，以分析更广泛频率范围内的组织阻抗数据。然后，他们产生了一个非线性逻辑回归模型，该模型使用EIS阻抗的振幅和相位的非线性组合来预测PTB。该模型应用于33例患者的EIS数据子集，实现了更好的PTB预测，灵敏度为82%，特异性为85%。这一早期范围界定研究的令人鼓舞的结果为这一提议提供了依据。 在这个项目中，我们建议采用先进的数据处理，建模和分析，独特的自动控制和系统工程系在谢菲尔德大学，以提高提取PTB相关功能，使开拓性的谢菲尔德宫颈阻抗谱为基础的PTB筛查设备。我们将开发一种新的非线性逻辑分析，以整合宫颈阻抗谱特征、人口统计学数据和其他临床可用的观察结果，从而获得更知情、临床可解释和显著改善的PTB预测。这些成就将通过预测700名妇女的PTB来证明，这些妇女将在谢菲尔德教学医院（STH）NHS基金会信托基金会进行基于宫颈阻抗谱的PTB筛查。

Every year, globally, about 15 million babies are born before 37 weeks. Preterm birth (PTB) complications are the leading cause of death of children under 5 years, causing 1 million deaths annually globally. PTB costs the UK NHS more than £1 billion annually, 10-fold higher than for term babies. As PTB is a major public health problem with profound implications on society, being able to identify women at risk of PTB during the course of their pregnancy is crucially important, so that care measures can be employed to delay birth to reduce potential long-term disability and impairment. However, accurate identification of women at risk for prevention and mitigation remain illusory. The fundamental problem with current screening approaches is that they are unable to assess and quantify cervical tissue composition, neither are they able to discriminate the various clinical conditions that are associated with PTB. For PTB to occur, the cervix must soften and dilate through a series of remodelling events at a molecular level. Based on this observation, a team of doctors and scientists led by Prof Dilly OC Anumba at Sheffield have revealed that women who are at high risk of PTB have lower resistance in their cervix in mid-pregnancy than women who deliver at term. This discovery motivated the MRC ECCLIPPxTM project where the Sheffield Mark V Electrical Impedance Spectroscopy (EIS) device was successfully investigated to quantify cervical remodelling remote from birth to predict PTB in a group of about 500 pregnant patients. The study has shown that EIS measured at 20-22 weeks predicts PTB (birth before 37 weeks gestation) with a sensitivity of ~70% and a specificity of ~80%. This promising performance was based only on a single derived parameter - tissue resistance over some discrete frequencies - and employed linear logistic and conventional statistical analysis. However, cervical tissue electrical properties are literally represented by the tissue impedance over a wide range of frequencies, which has both a real and an imaginary part known as resistance and reactance, respectively, and can also involve the effects of nonlinearities. Therefore, comprehensively correlating all EIS screening data parameters with PTB and taking into account possibly more complicated nonlinear associations could further improve the accuracy of PTB prediction. Motivated by these considerations, Profs Dilly OC Anumba and ZQ Lang established a collaboration which has preliminarily employed data modelling and model analysis techniques developed by Prof ZQ Lang's team to analyse the tissue impedance data over a wider range of frequencies. They have then produced a nonlinear logistic regression model which uses a nonlinear combination of the amplitude and phase of EIS impedance to predict PTB. The application of this model to a subset of the EIS data from 33 patients has achieved better PTB prediction with a sensitivity of 82% and a specificity of 85%. The encouraging outcome of this early scoping study informs this proposal. In this project, we propose to employ advanced data processing, modelling, and analysis uniquely developed in the Department of Automatic Control and Systems Engineering at the University of Sheffield to enhance the extraction of PTB related features enabled by the pioneering Sheffield cervical impedance spectroscopy-based PTB screening devices. We will develop a novel nonlinear logistic analysis to integrate the cervical impedance spectroscopy features, demographic data, and other clinically available observations for a more informed, clinically explainable, and significantly improved PTB prediction. The achievements will be demonstrated by prediction of PTB for 700 women who will have been studied by cervical impedance spectroscopy-based PTB screening at the Sheffield Teaching Hospitals (STH) NHS Foundation Trust.