Large-Scale Data-Driven Lung Cancer Diagnostics using Time-Resolved Fluorescence Spectroscopy (TRFS)

肺部结节是胸部CT扫描中常见的偶然发现。然而，肺结节的调查是耗时的，并且经常导致持续的放射学监测的长期随访。目前，有一个迫切需要的临床计算器，可以评估结节的风险是malignant.Recent进展，在介入肺病，包括导航到结节和执行时间分辨荧光光谱（TRFS）的能力，可以使立即床旁诊断肺癌，并帮助患者分层。TRFS研究了样品在光照射下的荧光（光的发射）随时间的变化，研究小组已经收集了超过30 μ字节的数据，这些数据以每周3 μ字节的速度增长。我们假设TRFS可以用来评估结节的恶性程度。我们的目标是为TFRS数据开发最先进的信号处理和机器学习工具，以从原始信号中估计关键参数，并对从癌/非癌组织样本中获得的带注释的临床TRFS数据进行分类。我们感兴趣的是：1）从噪声测量中对荧光衰减率、峰值强度等进行稳健的统计估计，2）提取与良性和恶性组织相关的荧光光谱特征的多途径分析，3）包括深度神经网络在内的数据驱动方法对健康和恶性组织进行分类，4）处理可能错误标记的数据、重复测量和空间信息，5）从潜在的多个视图学习，例如，拉曼光谱，以补充TRFS中包含的信息6）建立真实的时间算法，用于床边快速决策。培训成果：该项目在医疗信息学，人工智能和机器学习领域培训申请人，并将他/她与工程师，科学家，临床医生和行业联系起来，旨在发展世界领先的跨学科研究组合。申请人将受益于与专门从事颠覆性光学技术和医疗设备创新（Dhaliwal）、医疗机器人和图像处理（Khadem）以及信号处理和机器学习（Seth）的临床合作者的合作，他们都处于支持的理想位置职业发展并促进该项目的临床路径和影响。此外，该项目为申请人提供了与领先的介入医疗公司（波士顿科学公司）合作进行临床产品开发和测试的机会。

Pulmonary nodules are common, often incidental, findings on chest CT scans. The investigation of pulmonary nodules is, however, time-consuming and often leads to protracted follow-up with ongoing radiological surveillance. Currently, there is a critical need for clinical calculators that can assess the risk of the nodule being malignant.Recent advances in interventional pulmonology including the ability to navigate to nodules and perform Time-Resolved Fluorescence Spectroscopy (TRFS) may enable the immediate bed-side diagnosis of lung cancer and help in the stratification of patients. TRFS investigates the fluorescence (the emission of light) of a sample as a function of time when irradiated with light, and the team have collected over 30 Gigabytes of data which is growing at the rate of 3 Gigabytes per week.We hypothesize that TRFS can be used to assess the malignancy of the nodule. We aim to develop state-of-the-art signal processing and machine learning tools for TFRS data to estimate key parameters from the raw signal and classify annotated clinical TRFS data obtained from cancerous/non-cancerous tissue samples.Aims:We are interested in1) robust statistical estimation of fluorescence decay rate, peak intensity etc. from noisy measurements,2) multi-way analysis to extract fluorescence spectrum signatures associated with benign and malignant tissues,3) data driven approaches including deep neural network to classify healthy and malignant tissues,4) tackling possibly mislabelled data, repeated measurements, and spatial information,5) learning from potentially multiple views, e.g., Raman spectroscopy, to complement information contained in TRFS6) building real time algorithm to be used bedside for fast decision making.Training Outcomes:The project trains the applicant in the field of medical informatics, AI and machine learning, and connects him/her to engineers, scientists, clinicians, and industry with the aim of growing a world-leading interdisciplinary research portfolio. The applicant will benefit from working with clinical collaborators specialised in disruptive optical technologies and medical device innovation (Dhaliwal), medical robotics and image processing (Khadem), and signal processing and machine learning (Seth), who are all ideally placed to support the career development and facilitate the project's clinical pathways and impact. Moreover, the project offers the applicant an opportunity to collaborate with a leading interventional medical company (Boston Scientific) on clinical product development and testing.