Machine Learning to Generate a Multivariate Model of Brain Injury in HIV Patients

此R21应用程序响应RFA-MH-18-611“改变的神经元电路、受体和 艾滋病毒引起的中枢神经系统功能障碍中的网络(R21)“。 尽管联合抗逆转录病毒疗法有效地抑制了病毒，但多达50%的 患者继续患有艾滋病毒相关的神经认知障碍(HAND)。认知性 HIV患者的缺陷或损害可能是由于早期阶段的遗留影响 感染、持续神经炎的残留病毒库以及潜在的神经毒性 来自一些抗逆转录病毒药物。此外，与以下疾病相关的共病障碍 艾滋病毒+人口老龄化、药物使用的高流行率以及宿主特征，可能 进一步增加风险，加重手的严重程度。为患有以下疾病的患者提供最佳护理 HAND需要有效和适当的诊断，以指导有效的治疗。 然而，目前的手部诊断方法需要冗长和专门的认知 测试并涉及主观成分。我们的总体目标是开发一种无人监督的 机器学习(ML)算法评估大脑病理，使用客观测量，如 其他神经影像和临床增强的DTI上结构连接性的改变 变量。最终，这可能会导致一种稳健的方法来对子类型进行分类和量化 HIV感染者的脑损伤。该探索性项目有三个具体目标(SA)： SA1：使用自动无监督最大似然算法检测HIV感染者亚群 受试者，仅基于DTI示踪(结构连通性)。SA2：添加目标 人口统计学、遗传学、临床和非DTI磁共振变量与培训DTI声道成像数据的关系 设定，并确定它们对预测手感和认知能力的影响。SA3：评估 对欠抽样稳定性的优化模型，并判断其是否具有推广价值 到其他数据集，包括来自多中心研究(即多中心艾滋病队列)的数据集 研究)。我们优化的ML模型有可能提供高效、客观和 可重现的生物标志物，用于识别手部疾病或有手部疾病风险的个人，指导预防和 手部治疗，从而减轻艾滋病毒感染和痴呆症的负担。

This R21 application responds to RFA-MH-18-611 “Altered neuronal circuits, receptors and networks in HIV-induced Central Nervous System (CNS) dysfunction (R21)”. Despite effective viral suppression from combined antiretroviral therapy, up to 50% of the patients continue to have HIV-associated neurocognitive disorders (HAND). The cognitive deficits or impairment in HIV patients may be due to legacy effects from early stages of the infection, residual viral reservoirs with ongoing neuroinflammation, and potential neurotoxicity from some of the antiretroviral medications. Furthermore, co-morbid disorders associated with the aging HIV+ population, the high prevalence of substance use, and host characteristics, may further increase the risk and exacerbate the severity of HAND. Optimal care for patients with HAND requires efficient and appropriate diagnosis that can guide effective treatments. However, the current diagnostic approach for HAND requires lengthy and specialized cognitive tests and involves subjective components. Our overall goal is to develop an unsupervised machine learning (ML) algorithm to assess brain pathology, using objective measures such as alterations in structural connectivity on DTI, augmented with other neuroimaging and clinical variables. Ultimately, this may lead to a robust approach to classify subtypes and to quantify brain injury in HIV-infected individuals. This exploratory project has three specific aims (SA): SA1: Employ an automated unsupervised ML algorithm to detect subgroups of HIV-infected subjects, based solely on DTI tractography (structural connectivity). SA2: Add objective demographic, genetic, clinical, and non-DTI MR variables to the training DTI tractography data set, and determine their effects on predicting HAND and cognitive performance. SA3: Evaluate the optimized model for stability to undersampling, and determine whether it can be generalized to other data sets, including those from multi-center studies (i.e. Multicenter AIDS Cohort Study). Our optimized ML model has the potential to provide efficient, objective and reproducible biomarkers to identify individuals with or at risk for HAND, to guide prevention and treatment for HAND, and thereby ameliorate the burden of HIV infection and dementias.