FASEN2023 - Wall segmentation of aorta in IVUS images

Currently, abdominal aortic aneurysms (AAAs) are treated based on the diameter of the aorta measured with ultrasound (US), however, a better patient-specific marker is needed. The mean thickness of the wall is a major indicator for AAA rupture risk, which varies significantly within and between patients. So far, regional thickness has not been used in previous rupture risk analysis studies, since it cannot be measured with CT, MRI, nor with non-invasive US. This is the first study to map locally varying wall thickness of AAAs using intravascular ultrasound (IVUS).Since no ground truth of AAA wall thickness can be obtained in vivo, a novel ex vivo dataset was created of porcine, phantom and simulated aortas, including a ground truth. The porcine aortas introduce the image features found in real aortic tissue, the phantom aortas introduce the motion by pulsation of the aorta, and the simulated aortas introduce realistic aneurysm geometries. The ground truth wall geometry for the different experimental set-ups were obtained in various ways, i.e. for porcine aortas, micro-CT was used, for phantom aortas, manual segmentation and a known wall thickness were used, and for simulated AAAs, a given wall geometry was used as input.This ex vivo dataset can be used to train a neural network, and a trained model can successfully segment the aortic wall in IVUS images in AAAs in vivo. Regionally varying wall thickness and geometry of aortas can be obtained, leading to a patient-specific marker for more advanced rupture risk assessment of AAAs.

当前，腹主动脉瘤（abdominal aortic aneurysms, AAAs）的临床诊疗以超声（ultrasound, US）测得的主动脉直径为核心依据，但目前仍亟需更贴合患者个体特征的特异性评估标志物。管壁平均厚度是预测腹主动脉瘤破裂风险的关键指标，但该指标在单患者体内及不同患者群体间均存在显著差异。截至目前，既往破裂风险分析研究均未采用区域管壁厚度这一评估维度，原因在于该指标无法通过计算机断层扫描（computed tomography, CT）、磁共振成像（magnetic resonance imaging, MRI）及无创超声进行有效测量。本研究首次利用血管内超声（intravascular ultrasound, IVUS）技术，对腹主动脉瘤的局部管壁厚度变化进行精准测绘。由于无法在活体（in vivo）环境下获取腹主动脉瘤管壁厚度的真值参考，本研究构建了一套全新的离体（ex vivo）数据集，涵盖猪源主动脉、仿体主动脉及模拟主动脉三类样本，并附带对应的真值标签。其中，猪源主动脉可模拟真实主动脉组织的影像学特征，仿体主动脉可复现主动脉搏动引发的运动伪影，而模拟主动脉则可构建贴合临床实际的动脉瘤几何形态。针对不同实验模型的管壁几何真值，本研究通过多种适配方式获取：猪源主动脉采用显微计算机断层扫描（micro-CT）进行高精度测量；仿体主动脉通过人工分割结合预设已知管壁厚度获取真值；模拟腹主动脉瘤则以预先设定的管壁几何形态作为真值输入。该离体数据集可用于神经网络模型的训练，经训练得到的模型可成功在活体腹主动脉瘤的血管内超声图像中实现主动脉壁的精准分割。通过该数据集可获取主动脉局部管壁厚度及几何形态的个性化信息，进而得到贴合患者个体特征的评估标志物，用于更精准的腹主动脉瘤破裂风险分级评估。