To evaluate deep learning-based calcium segmentation and quantification on ECG-gated cardiac CT scans compared with manual evaluation. Automated calcium quantification was performed using a neural network based on mask regions with convolutional neural networks (R-CNNs) for multi-organ segmentation. Manual evaluation of calcium was carried out using proprietary software. This is a retrospective study of archived data. This study used 40 patients to train the segmentation model and 110 patients were used for the validation of the algorithm. The Pearson correlation coefficient between the reference actual and the computed predictive scores shows high level of correlation (0.84; P < .001) and high limits of agreement (±1.96 SD; −2000, 2000) in Bland–Altman plot analysis. The proposed method correctly classifies the risk group in 75.2% and classifies the subjects in the same group. In total, 81% of the predictive scores lie in the same categories and only seven patients out of 110 were more than one category off. For the presence/absence of coronary artery calcifications, the deep learning model achieved a sensitivity of 90% and a specificity of 94%. Fully automated model shows good correlation compared with reference standards. Automating process reduces evaluation time and optimizes clinical calcium scoring without additional resources.

中文翻译：

---

使用深度学习和多器官分割从心电图门控心脏计算机断层扫描全自动计算 Agatston 评分：一项验证研究

与手动评估相比，评估基于深度学习的心电门控心脏 CT 扫描的钙分割和量化。使用基于掩模区域的神经网络和用于多器官分割的卷积神经网络（R-CNN）进行自动钙定量。使用专有软件对钙进行手动评估。这是对存档数据的回顾性研究。本研究使用 40 名患者来训练分割模型，并使用 110 名患者来验证算法。在 Bland-Altman 图分析中，参考实际分数和计算预测分数之间的 Pearson 相关系数显示出高相关性 (0.84; P < .001) 和高一致性限度 (±1.96 SD; -2000, 2000)。所提出的方法对风险组的正确分类率为 75.2%，并将受试者分类为同一组。总的来说，81% 的预测分数属于同一类别，110 名患者中只有 7 名患者偏离了一个以上类别。对于是否存在冠状动脉钙化，深度学习模型的敏感性为 90%，特异性为 94%。与参考标准相比，全自动模型显示出良好的相关性。自动化流程可缩短评估时间并优化临床钙评分，无需额外资源。