【摘要】 针对冠脉血管堵塞诊断依赖人工观察计算机断层扫描造影（CTA）二维切片图像，存在主观性强、专业要求高等问题，提出了基于nnUNet模型的冠脉血管自动分割方法与改进的三维重建方法。该方法依据冠脉CTA的血管占比少的图像特点，使用Foacl Loss损失函数替换交叉熵损失函数，并通过引入全连接的密集条件随机场（Dense CRF）解决分割后特征信息损失问题。针对原有的三维重建方法无法展示患者冠脉具体病灶区域与实时性差的问题，提出了一种组合式冠脉二维切片图像三维重建方法。最后，以医院提供的患者临床冠脉二维切片影像为数据集进行实验，证明了改进后的nnUNet模型相较于其他图像分割模型对冠脉血管的分割精度得到了有效提升。更多还原

【Abstract】 To address the limitations of subjective interpretation and high expertise dependency in diagnosing coronary artery stenosis through manual inspection of two-dimensional Computed Tomography Angiography(CTA)slice images,this study proposes an automatic coronary artery segmentation method based on the nnUNet framework,coupled with an enhanced 3D reconstruction approach. Leveraging the characteristic sparse vascular distribution in coronary CTA images,the conventional cross-entropy loss function is replaced with Focal Loss to mitigate class imbalance,while a fully connected Dense Conditional Random Field(Dense CRF)is incorporated to preserve post-segmentation feature details. To overcome the deficiencies of conventional 3D reconstruction methods—particularly their inability to visualize specific lesion regions and poor real-time performance—a composite 3D reconstruction algorithm for coronary 2D slices is introduced. Experimental validation using clinical coronary CTA datasets from hospital patients demonstrates that the optimized nnUNet model achieves superior segmentation accuracy compared to existing methods,effectively improving coronary vessel delineation.更多还原