报告摘要（ABSTRACT）：

Medical imaging is widely used in clinical decision making. However, medical image acquisition or its acquired image still suffers from an array of challenges such as metal artifacts, slow acquisition time, anisotropic resolution, strong noise, etc. In this talk, we present several learning approaches that attempt to recover the original images under these adverse conditions:

• a dual domain network (DuDoNet) for reducing metal artifacts in CT via joint learning in both sinogram and image domains;

• a dual domain recurrent network (DuDoRNet) for MRI image reconstruction from undersampled k-space data via joint and recurrent learning in both frequency and image domains;

• a spatially adaptive interpolation network (SAINT) for synthesizing slices to mitigate the anisotropic resolution issue; and

• an artifact disentanglement network (ADN) for removing artifacts or noises without paired data while preserving anatomical structures.

Our supervised and unsupervised approaches leverage deep neural networks as cores, integrate specific domain knowledge, and yield high quality recovery for both simulated data and clinical images.

报告人简介（BIO）：

周少华博士现任中科院计算所研究员，曾在西门子医疗研究院任职首席影像AI专家，致力于研发与图像相关的创新及产品。他已经编撰了五本学术专著，发表了200+篇学术期刊及会议论文，并拥有140+项批准专利。他多次因学术成就和创新贡献而获奖，包括发明奥斯卡奖、西门子年度发明家、马里兰大学ECE杰出校友等。他热心奉献于专业社区，是行业协会MICCAI财长和理事、专业期刊IEEE Trans. Medical Imaging和Medical Image Analysis编委会成员、专业会议CVPR，NeurIPS，AAAI和MICCAI领域主席、《视觉求索》公众号联席主编等。他担任了MICCAI2020的程序联席主席。Fellow of IEEE and AIMBE (美国医学与生物工程院)。

“论坛”主请人联系方式：

李亮  lliang@tsinghua.edu.cn