Individually-tailored clinical decision support for management of indeterminate pulmonary nodules

摘要（项目描述） 低剂量计算机断层扫描（LDCT）肺部筛查计划的推出正在美国加速。 美国，旨在早期发现肺癌，以提高长期生存率。然而， 这种成像程序是不确定的肺结节（IPN）的增加的发现。重要问题- 仍然围绕着屏幕和偶然检测到的IPN的有效管理：虽然许多是良性的， 一小部分会发展成癌症IPNs诊断模型和相关管理指南 已经在前面描述过，但是它们的真实世界验证是有限的。此外，大多数模型仅 在单个时间点使用IPN的“快照”，而没有考虑到渐进的变化。 现在有机会通过包含患者不断发展的医疗信息来推进这种预测模型。 病史，结合临床和影像学生物标志物，以改善预测和个性化定制管理 随着时间的推移。 该成像信息学提案的目标是为以下方面开发临床决策支持工具： 管理屏幕和意外检测到的IPN。我们应对两个关键挑战：1）发展 用于预测IPN将如何演变的连续时间模型;以及2）使用该预测来确定 随着时间的推移，一系列行动将优化（筛选）个人的结果。我们先来探索一下发展-- 连续时间信念网络（CTBN），一种时间概率模型，用于预测 患者患肺癌。与传统方法不同，CTBN不需要固定的采样频率， 随时间的数据（例如，每年进行的所有观察），因此更适合于现实世界的临床 设置和观测数据集。通过CTBN计算的概率随后被输入到 部分可观察马尔可夫决策过程（POMDP），以指导IPN管理决策。从 POMDP，可以选择策略（随时间的动作序列）以实现期望的目标（例如，尽量减少 诊断时间），给出个体的过去和当前观察/决定。对于CTBN和 POMDP，我们在设计和实现中探索新的方法，克服计算挑战， 把这些模式转化为实践。一个基于Web的界面实现，提供了一个临床的设计， 为医生提供决策工具，以了解模型的建议。评价侧重于评估 CTBN和POMDP相对于已知结果的性能，以及与其他常规方法相比， 方法（例如，逻辑回归，决策树，动态信念网络）;以及 制度影响决策。这一努力推进了我们过去在概率模型和资本方面的研究- 在肺癌筛查方面的专业知识，包括国家肺筛查试验（NLST）的过去领导。 这一努力的结果将是一套信息驱动的建模工具和新的时间预测模型 通知IPN管理层。

ABSTRACT (PROJECT DESCRIPTION) The rollout of low-dose computed tomography (LDCT) lung screening programs is accelerating in the United States, aiming for earlier detection of lung cancer to improve long-term survival. However, a consequence of such imaging programs is the increased discovery of indeterminate pulmonary nodules (IPNs). Significant ques- tions remain around the effective management of screen- and incidentally-detected IPNs: while many are benign, a fraction will go on to become cancerous. Diagnostic models for IPNs and associated management guidelines have been described previously, but their real-world validation is limited. Moreover, the majority of models only use a “snapshot” of the IPN at a single point in time and fail to take into consideration progressive changes. Opportunities now exist to advance such predictive models by encompassing the patient's evolving medical history, combining clinical and imaging biomarkers to improve prediction and individually-tailor the management of IPNs over time. The objective of this imaging informatics proposal is the development of a clinical decision support tool for the management of screen- and incidentally-detected IPNs. We address two key challenges: 1) the development of a continuous-time model for predicting how the IPN will evolve; and 2) the use of this prediction to determine a series of actions over time that will optimize (screening) outcomes for the individual. We first explore the devel- opment of a continuous time belief network (CTBN), a temporal probabilistic model to predict the likelihood of a patient to develop lung cancer. Unlike traditional approaches, CTBNs do not require fixed sampling frequency of the data over time (e.g., all observations made annually) and are thus more amenable to real-world clinical settings and observational datasets. The probabilities computed through the CTBN are subsequently input into a partially-observable Markov decision process (POMDP) to guide IPN management decisions. From the POMDP, policies (sequences of actions over time) can be chosen to achieve a desired goal (e.g., minimizing time to diagnosis), given past and current observations/decisions for an individual. For both the CTBN and POMDP, we explore novel methods in the design and implementation, overcoming computational challenges to realize translation of these models into practice. A web-based interface is implemented, providing a clinical de- cision making tool for physicians to understand the models' recommendations. Evaluation focuses on assessing the performance of the CTBN and POMDP relative to known outcomes and compared to other conventional methods (e.g., logistic regression, decision trees, dynamic belief networks); as well as the overall impact of the system to influence decision-making. This effort advances our past research in probabilistic models and capital- izes on expertise in lung cancer screening, including past leadership of the National Lung Screening Trial (NLST). The result of this effort will be a set of informatics-driven modeling tools and new temporal predictive models informing IPN management.