Bin Liu

Automatic brain tumor segmentation on MRI is a prerequisite to provide a quantitative and intuitive assistance for clinical diagnosis and treatment. Meanwhile, 3D deep neural network related brain tumor segmentation models have demonstrated considerable accuracy improvement over corresponding 2D methodologies. However, 3D brain tumor segmentation models generally suffer from high computation cost. Motivated by a recently proposed 3D dilated multi‐fiber network (DMF‐Net) architecture that pays more attention to reduction of computation cost, we present in this work a novel encoder‐decoder neural network, ie a 3D asymmetric expectation‐maximization attention network (AEMA‐Net), to automatically segment brain tumors. We modify DMF‐Net by introducing an asymmetric convolution block into a multi‐fiber unit and a dilated multi‐fiber unit to capture more powerful deep features for the brain tumor segmentation. In addition, AEMA‐Net further incorporates an expectation‐maximization attention (EMA) module into the DMF‐Net by embedding the EMA block in the third stage of skip connection, which focuses on capturing the long‐range dependence of context. We extensively evaluate AEMA‐Net on three MRI brain tumor segmentation benchmarks of BraTS 2018, 2019 and 2020 datasets. Experimental results demonstrate that AEMA‐Net outperforms both 3D U‐Net and DMF‐Net, and it achieves competitive performance compared with the state‐of‐the‐art brain tumor segmentation methods.

Computer‐aided classification of pathological images is of the great significance for breast cancer diagnosis. In recent years, deep learning methods for breast cancer pathological image classification have made breakthrough progress, becoming the mainstream in this field. To capture more discriminant deep features for breast cancer pathological images, this work introduces a novel attention high‐order deep network (AHoNet) by simultaneously embedding attention mechanism and high‐order statistical representation into a residual convolutional network. AHoNet firstly employs an efficient channel attention module with non‐dimensionality reduction and local cross‐channel interaction to achieve local salient deep features of breast cancer pathological images. Then, their second‐order covariance statistics are further estimated through matrix power normalization, which provides a more robust global feature presentation of breast cancer pathological images. We extensively evaluate AHoNet on the public BreakHis and BACH breast cancer pathology datasets. Experimental results illustrate that AHoNet gains the optimal patient‐level classification accuracies of 99.29% and 85% on the BreakHis and BACH database, respectively, demonstrating the competitive performance with state‐of‐the‐art single models on this medical image application.