Automated Machine Learning-Based Brain Artery Segmentation, Anatomical Prior Labeling, and Feature Extraction on MR Angiography

基于自动机器学习的脑动脉分割、解剖先验标记和 MR 血管造影特征提取

• 批准号: 10759721
• 负责人: Md Nasir Uddin
• 金额: $ 15.4万
• 依托单位: UNIVERSITY OF ROCHESTER
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-08 至 2025-08-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10759721
• 关键词: 3-Dimensional; Address; Affect; Anatomy; Angiography; Arteries; Attention; Biological Markers; Blood - brain barrier anatomy; Blood Vessels; Blood capillaries; Brain; Case Study; Cause of Death; Cerebral small vessel disease; Cerebrovascular Disorders; Cerebrum; Classification; Clinical; Clinical Data; Cognition; Cohort Studies; Collaborations; Consumption; Contrast Media; Data; Data Analytics; Data Engineering; Data Set; Databases; Diagnosis; Diameter; Disease Marker; Distal; Effectiveness; Functional disorder; Goals; HIV; HIV Infections; High Prevalence; Image; Individual; Intervention; Label; Lead; MRI Scans; Machine Learning; Magnetic Resonance Angiography; Magnetic Resonance Imaging; Manuals; Measures; Methods; Monitor; Morphology; Neurocognitive; Participant; Pathogenesis; Patient-Focused Outcomes; Penetration; Persons; Play; Protocols documentation; Quality of life; Reading; Resolution; Risk; Role; Speed; Subcortical Infarctions; Techniques; Time; Training; Translations; Vascular Cognitive Impairment; Vascular Dementia; Vascular Diseases; Visual; White Matter Hyperintensity; antiretroviral therapy; arteriole; automated segmentation; biomarker performance; brain based; cerebral atrophy; cerebral microbleeds; cerebrovascular; clinical diagnosis; clinical examination; cognitive function; cognitive performance; cohort; deep learning; disability; executive function; feature extraction; imaging biomarker; improved; in vivo; machine learning framework; neuroimaging; neuroimaging marker; neurovascular unit; novel strategies; processing speed; prospective; reconstruction; ultra high resolution; vascular abnormality; vascular risk factor; venule

Cerebrovascular disease is a major cause of death and disability globally. Time-of-flight magnetic resonance angiography (TOF-MRA) is a common noninvasive technique to evaluate vascular abnormalities. Accurate segmentation and feature extraction of cerebral vessels from TOF-MRA data is crucial for diagnosis and treatment of cerebrovascular diseases. However, manual annotation of vessels is a time-consuming task even for experts, and an automatic segmentation method can speed-up the task significantly. Although several machine learning-based post-processing approaches exist, these are limited to vessel segmentation only and require large manually segmented training datasets. Further, image resolution plays an important role to segment the vessels as well as extracting features (e.g., diameters, number of branches, tortuosity) accurately. Aim 1: To develop an automated anatomical prior based machine learning framework for brain artery segmentation with minimal training MR high resolution data followed by feature extractions on low resolution MRA from an existing database. Aim 2: To assess the relationship between targeted vessel features and neuroimaging biomarkers of cerebral small vessel disease (CSVD). Aim 3: To assess the relationship between targeted vessel features and cognitive performance in HIV CSVD cohort. To achieve this goal, we will implement our recently proposed automated vessel segmentation and feature extraction pipeline “BayesTract” in combination with a super-resolution approach. We will validate it using in-vivo MRI scans. We will then examine the vascular features and their associations with CSVD markers and cognitive performance in an existing dataset (101 HIV+ and 102 HIV- controls with CSVD). We hypothesize that this approach will provide more accurate and time-efficient measures compared to existing approaches. We also hypothesize that vascular features and their associations with CSVD markers and cognition will be significantly different between those with and without CSVD. This study helps advance the state-of-the-art in brain vessel segmentation and feature extractions from non- invasive TOF-MRA, which could hasten the translation of vessel related biomarkers into the clinical setting. This will be essential in evaluating promising interventions and ultimately, lead to ameliorating patient outcome and quality of life.

脑血管疾病是全球死亡和残疾的主要原因。飞行时间磁共振 血管造影术（TOF-MRA）是评价血管异常的常用非侵入性技术。准确 从TOF-MRA数据中分割和提取脑血管特征对于诊断和 治疗脑血管疾病。然而，血管的手动注释是一项耗时的任务， 而自动分割方法可以显著加快任务的执行速度。尽管几 存在基于机器学习的后处理方法，这些方法仅限于血管分割， 需要大量手动分割的训练数据集。此外，图像分辨率对分割起着重要作用 血管以及提取特征（例如，直径、分支数量、迂曲度）。 目标1：开发一个基于脑动脉解剖先验的自动化机器学习框架 用最少的训练MR高分辨率数据进行分割，然后在低分辨率上进行特征提取 来自现有数据库的MRA。 目的2：探讨脑梗死靶血管特征与脑梗死神经影像学标志物的关系。 小血管疾病（CSVD）。 目的3：评估HIV CSVD中靶血管特征与认知表现之间的关系 队列。 为了实现这一目标，我们将实现我们最近提出的自动血管分割和特征 提取管道“贝叶斯跟踪”与超分辨率方法相结合。我们将使用体内 核磁共振扫描。然后，我们将研究血管特征及其与CSVD标志物和认知功能的关系。 现有数据集（101例HIV+和102例HIV-对照CSVD）的性能。我们假设这 与现有方法相比，新方法将提供更准确和更省时的措施。我们也 假设血管特征及其与CSVD标记物和认知的关联将显著 有CSVD和没有CSVD的人之间的差异。 这项研究有助于推进脑血管分割和非脑血管特征提取的最新技术。 侵入性TOF-MRA，可以加速血管相关生物标志物转化为临床环境。这 在评估有希望的干预措施中至关重要，并最终改善患者的结局， 生活质量