Computer-aided detection of non-calcified plaques in coronary CT angiograms

心血管疾病是美国男性和女性死亡的主要原因。 超过1600万美国人患有冠心病（CHD），每年造成约50万人死亡。 最常见的冠心病是冠状动脉疾病，主要由动脉粥样硬化引起。临床 近年来的证据表明，非钙化斑块（NCP）比非钙化斑块更容易破裂。 钙化斑块斑块破裂和血栓形成是急性心肌梗死的主要原因 梗塞多排螺旋CT冠状动脉血管造影（cCTA）有可能帮助临床医生在早期 检测和定量NCP。因此，cCTA可用于CHD检测，风险分层， 监测和评估风险降低治疗的有效性。然而，许多潜在的 这些应用尚未在临床上使用。 本项目的目标是开发一个计算机辅助检测（CADe）系统， 阅读器，用于协助临床医生在cCTA检查中检测和量化NCP。我们的具体目标是 （1）开发用于检测导致狭窄和/或正性重塑的NCP的机器学习方法 沿着冠状动脉，（2）通过以下方法评估CADe对放射科医生在cCTA上检测NCP的影响： ROC研究。为了实现这些目标，我们将收集cCTA病例数据库用于培训， 测试CADe系统，通过设计3D多尺度冠状动脉响应来定义搜索空间 增强，分割和动态气球血管跟踪方法，开发一个独特的血管- 拼接方法，以自动识别所有动脉段中每个动脉段的最佳质量相位 前瞻性或回顾性门控cCTA检查的可用阶段，开发创新的血管部门- 轮廓分析和血管管腔分析，以检测导致狭窄或正性重塑的NCP， 估计NCP总体积，并探索通过体模定量斑块密度的校准方法 问题研究为了证明CADe的有用性，将进行临床前读片员研究，以比较 放射科医师在使用和不使用CADe的情况下对NCP的检测准确性。 本课题的主要创新点包括：（1）首次实现了自动检测 非钙化斑块，包括cCTA中导致正性重塑或狭窄的斑块，（2）新的 机器学习技术，包括血管缝合方法，血管扇区轮廓分析，多尺度 增强响应和动态球囊跟踪，特别适用于冠状动脉树，和（3） 进行第一次ROC研究，以评估CADe对放射科医生检测NCP的影响。

Cardiovascular disease is the leading cause of death in both men and women in the United States. Over 16 million Americans have coronary heart disease (CHD), causing about 0.5 million deaths each year. The most common CHD is coronary artery disease which is mainly caused by atherosclerosis. Clinical evidence in recent years shows that noncalcified plaques (NCPs) are more vulnerable to rupture than calcified plaques. Plaque rupture and the thrombosis that follows is the main cause of acute myocardial infarction. Multidetector coronary CT angiography (cCTA) has the potential to help clinicians in early detection and in quantification of NCPs. cCTA may thus be useful for CHD detection, risk stratification, monitoring, and evaluation of the effectiveness of risk reduction treatment. However, many of these potential applications have not been utilized clinically. The goal of this project is to develop a computer-aided detection (CADe) system to serve as a second reader for assisting clinicians in detection and quantification of NCPs in cCTA exams. Our specific aims are to (1) develop machine learning methods for detection of NCPs causing stenosis and/or positive remodeling along coronary arteries, and (2) evaluate the effect of CADe on radiologists' detection of NCPs on cCTA by observer ROC study. To achieve these aims, we will collect a database of cCTA cases for training and testing the CADe system, define the search space by designing 3D multiscale coronary artery response enhancement, segmentation, and dynamic balloon vessel tracking methods, develop a unique vessel- stitching method to automatically identify the best-quality phase for each individual artery segment from all available phases in prospectively or retrospectively gated cCTA exams, develop innovative vessel-sector- profile analysis and vessel lumen analysis to detect NCPs that cause stenosis or positive remodeling, estimate the total NCP volume, and explore calibration method to quantify plaque density by phantom studies. To demonstrate the usefulness of CADe, a preclinical reader study will be conducted to compare radiologists' detection accuracy of NCPs with and without CADe. The major innovations of this project include (1) being the first CADe system to automatically detect non-calcified plaques including those cause positive remodeling or stenosis in cCTA, (2) development of new machine learning techniques including the vessel-stitching method, vessel-sector-profile analysis, multiscale enhancement response, and dynamic balloon tracking specifically suited for coronary arterial trees, and (3) conducting the first ROC study to evaluate the effect of CADe on radiologists' detection of NCPs.