Improving the data efficiency of machine (deep) learning methods through multi-modal self-supervised (unsupervised) algorithms. Aiming at significantly reducing the human annotation required for medical application
Research Projects
Multimodal Self-Supervised Learning for Medical Image Analysis: we propose a novel self-supervised multimodal puzzle-solving proxy task, which facilitates neural network representation learning from multiple image modalities.
3D Self-Supervised Methods for Medical Imaging: we propose five different methods, which facilitate feature learning from unlabeled 3D scans.
Strong multimodal alignment of image and text patient data samples from a large real-world clinical corpus with a variety of cases. Aiming at improving disease detection, disease reporting, and annotation efficiency.