返回
Alignment-Aware 3D Point Cloud Anomaly Detection with Adversarial Normalizing Flows
DOI:10.3390/make8070206.png)
摘要
En 中文
检测三维点云中局部形态异常具有挑战性,因为几何偏差与刚性姿态变化、残留配准误差、采样噪声和个体间差异相互纠缠。这一难题在神经影像学转化研究中尤为突出,因为异常形状变化可能细微且异常标注数据稀少。我们提出了一种无监督框架AdvFlow3D-AD,将三维异常检测表述为两阶段分解问题。首先,通过快速全局配准结合多尺度迭代最近点优化建立共同几何参考框架,并减少刚性体无关变异。其次,采用对抗正则化的归一化流模型学习对齐法向坐标的残差分布,基于与学习所得正常隐式支撑集的距离生成局部异常分数。在正常数据上的百分位校准可定义可解释的点级和对象级操作点,无需训练时依赖异常样本。我们在Real3D-AD和Anomaly ShapeNet3D数据集上评估了AdvFlow3D-AD,在Real3D-AD上获得点级受试者工作特征曲线下面积(AUROC)0.747,在Anomaly ShapeNet3D上获得对象级AUROC 0.816。此外,我们针对儿科围产期窒息病例开展了一项探索性神经发育脑形态案例研究。结果生成的异常图在神经放射学评估下,在解剖学上合理的海马体和小脑区域表现出定性空间一致性。这些结果提示,当异常标签有限时,分离几何无关变异与残留形态有助于实现可解释的异常定位。
Keyword:
3D anomaly detection
adversarial training
point cloud alignment
neurodevelopmental disorders
normalizing flows
期刊
M
IF:
6
论文数:
832
被引数:
1.8K
机构
引用论文
A Multiparametric, Reliability-Weighted Fetal Brain Biometry Portrait for the Prediction of Multi-Domain Neurodevelopmental Outcomes Across 12 and 24 Months of Age多参数、可靠性加权的胎儿脑生物测量图,用于预测12个月和24个月年龄跨度的多领域神经发育结局
Unsupervised anomaly detection in brain MRI: Learning abstract distribution from massive healthy brains脑MRI中的无监督异常检测: 从大量健康大脑中学习抽象分布
Moon, A.S.; Lee, K.; Lee, J. Survey on 3D Anomaly Detection Regarding Defect Size: Tradeoff Between Accuracy and Efficiency with Future Research Direction. IEEE Access 2025, 13, 114595–114607. [Google Scholar] [CrossRef]Moon, A.S.; Lee, K.; Lee, J. 关于缺陷尺寸的3D异常检测综述:精度与效率的权衡及未来研究方向. IEEE Access 2025, 13, 114595–114607. [Google Scholar] [CrossRef]
An Iterative Closest Points Algorithm for Registration of 3D Laser Scanner Point Clouds with Geometric Features三维激光扫描仪点云几何特征配准的迭代最近点算法
SENSORS
IF3.5
Advances in magnetic resonance imaging of the developing brain and its applications in pediatrics发育中大脑的磁共振成像进展及其在儿科中的应用
Satheesan, A.P.; Chinnappa, A.R.; Goudar, G.; Raghoji, C. Correlation between early magnetic resonance imaging brain abnormalities in term infants with perinatal asphyxia and neuro developmental outcome at one year. Int. J. Contemp. Pediatr. 2020, 7, 1957–1961. [Google Scholar] [CrossRef]Satheesan, A.P.; Chinnappa, A.R.; Goudar, G.; Raghoji, C. 对足月新生儿围产期窒息早期脑磁共振成像异常与一年神经发育结局的相关性研究。Int. J. Contemp. Pediatr. 2020, 7, 1957–1961. [Google Scholar] [CrossRef]

