Zhu Yanjie, Fahmy Ahmed S, Duan Chong, Nakamori Shiro, Nezafat Reza
Department of Medicine, Cardiovascular Division, Beth Israel Deaconess Medical Center and Harvard Medical School, 330 Brookline Ave, Boston, MA 02215 (Y.Z., A.S.F., C.D., S.N., R.N.).
Radiol Artif Intell. 2020 Jan 29;2(1):e190034. doi: 10.1148/ryai.2019190034.
To assess the performance of an automated myocardial T2 and extracellular volume (ECV) quantification method using transfer learning of a fully convolutional neural network (CNN) pretrained to segment the myocardium on T1 mapping images.
A single CNN previously trained and tested using 11 550 manually segmented native T1-weighted images was used to segment the myocardium for automated myocardial T2 and ECV quantification. Reference measurements from 1525 manually processed T2 maps and 1525 ECV maps (from 305 patients) were used to evaluate the performance of the pretrained network. Correlation coefficient and Bland-Altman analysis were used to assess agreement between automated and reference values on per-patient, per-slice, and per-segment analyses. Furthermore, transfer learning effectiveness in the CNN was evaluated by comparing its performance to four CNNs trained using manually segmented T2-weighted and postcontrast T1-weighted images and initialized using random-weights or weights of the pretrained CNN.
T2 and ECV measurements using the pretrained CNN strongly correlated with reference values in per-patient (T2: = 0.88, 95% confidence interval [CI]: 0.85, 0.91; ECV: = 0.91, 95% CI: 0.89, 0.93), per-slice (T2: = 0.83, 95% CI: 0.81, 0.85; ECV: = 0.84, 95% CI: 0.82, 0.86), and per-segment (T2: = 0.75, 95% CI: 0.74, 0.77; ECV: = 0.76, 95% CI: 0.75, 0.77) analyses. In Bland-Altman analysis, the automatic and reference values were in good agreement in per-patient (T2: 0.3 msec ± 2.9; ECV: -0.3% ± 1.7), per-slice (T2: 0.1 msec ± 4.6; ECV: -0.3% ± 2.5), and per-segment (T2: 0.0 msec ± 6.5; ECV: -0.4% ± 3.5) analyses. The performance of the pretrained network was comparable to networks refined or trained from scratch using additional manually segmented images.
Transfer learning extends the utility of pretrained CNN-based automated native T1 mapping analysis to T2 and ECV mapping without compromising performance. © RSNA, 2020.
评估一种基于完全卷积神经网络（CNN）迁移学习的自动化心肌T2和细胞外容积（ECV）定量方法的性能，该网络已在T1映射图像上进行预训练以分割心肌。
使用一个先前使用11550幅手动分割的原始T1加权图像进行训练和测试的单一CNN来分割心肌，以进行自动化心肌T2和ECV定量。来自1525幅手动处理的T2图和1525幅ECV图（来自305名患者）的参考测量值用于评估预训练网络的性能。在每位患者、每层和每段分析中，使用相关系数和Bland-Altman分析来评估自动化值与参考值之间的一致性。此外，通过将其性能与使用手动分割的T2加权和对比剂后T1加权图像训练并使用随机权重或预训练CNN的权重初始化的四个CNN进行比较，来评估CNN中的迁移学习效果。
在每位患者（T2：r = 0.88，95%置信区间[CI]：0.85，0.91；ECV：r = 0.91，95% CI：0.
