从三维磁共振血管造影（MRA）数据集中自动和精确地分割出AVM结构，对于优化手术计划和降低手术风险至关重要。该领域面临的主要问题是单一MRA成像对AVM复杂结构显示不全、缺少解剖定位信息、存在较低的信噪比和较高的背景噪声；AVM图像分割时，也面临多类目标和非目标之间较近的类间距和较大的类间方差等挑战性问题。对此，我们拟利用各类MRA成像数据所反映的脑血管、血流、皮层沟回等结构形态信息，建立混合优化的统计学模型开展定性和定量分析。本项目已实现了脑动脉的精确分割，后续研究对策是：（1）解决噪声条件下目标与非目标结构的灰度和形状特性相似问题；（2）结合结构成像与血流信息，综合利用各种核提取算法和后处理技术进行复杂AVM的量化描述；（3）建立多类多层次的脑皮层沟回分割模型和三维网格模型，进行解剖结构的精确定位、三维重建、逼真显示。最终为可视化AVM辅助诊疗和临床应用建立有效的研究分析平台。

Automatic and accurate segmentation of the AVM internal structure from three dimensional magnetic resonance angiography (MRA) datasets is important to optimize surgical planning and relieve surgical risk. In this field, the main problems consist in imperfect display of AVM structure, absence of anatomical position information, low SNR and high background noise. In addition, the AVM segmentation is also confronted with close class separation distance and great intra-class variance between or among various objects and non-objects.Therefore, we should utilize the morphology structure such as cerebral vessels, blood flow, cortex gyrus and sulus to construct mixture-optimal statistic model for the qualitative and quantitative analysises. The project has realized accurate segmentation of cerebral arteries. The subsequent research measures are: (1) to solve the problem of luminance and shape similarities presented between the objects and non-objects within MRA data; (2) to quantitatively characterize the complicate AVM structures, integrating with structures and blood flows, synthetically utilizing various algorithms w.r.t nucleus extraction and postprocess; (3) to construct multi-class and multi-degree segmentation model and 3D grid model of cerebral cortex for accurate position，3D reconstruction， and living brain display. Finally, we will establish a effective platform for researching and analysing w.r.t. computer-assisted and visualized diagnosis & treat of AVM surgery.