Pixel-based evaluation with 15% VGF DBT images.

MSE and GRMSE results of the DBT images reconstructed with FDK and TV-IR, deblurred images by DRCNN, deblurred images by the proposed method with MAE, AL-MAE, and PL-MAE in generalization testset. (mean±standard deviation).在泛化测试集（generalization testset）中，采用FDK重建算法（Feldkamp-Davis-Kress Algorithm, FDK）与总变分迭代重建（Total Variation Iterative Reconstruction, TV-IR）得到的数字乳腺断层摄影（Digital Breast Tomosynthesis, DBT）图像、经深度递归卷积神经网络（Deep Recurrent Convolutional Neural Network, DRCNN）去模糊后的图像，以及结合平均绝对误差（Mean Absolute Error, MAE）、自适应平均绝对误差（Adaptive Learning Mean Absolute Error, AL-MAE）与位置加权平均绝对误差（Position-Weighted Mean Absolute Error, PL-MAE）的本文所提方法得到的去模糊图像的均方误差（Mean Squared Error, MSE）与广义相对均方根误差（Generalized Relative Root Mean Squared Error, GRMSE）结果（均值±标准差）