Computer Vision News 26 Congrats, Doctor Moucheng! Moucheng Xu has recently completed his PhD from the Centre for Medical Image Computing at the University College London. His research focuses on deep learning with limited supervisions in medical imaging. He has now moved to industry in MedTech to continue to pursue his passion for creating intelligent medical imaging systems. Congrats, Doctor Moucheng! Deep learning has now become the de facto data-driven approach for medical image analysis in the digital era. A well-executed deep learning model requires a massive amount of data and their corresponding labels. However, acquiring labels in medicine is extremely challenging due to the high costs in both time and money. Moucheng’s thesis seeks the answer to the following question: How can we train a deep learning model without sufficient labels for medical image analysis? In the first work of his thesis, he presents a new perspective on the visual attention mechanism as learnable morphological operations at the feature level. It also introduces a new consistency regularization on dilated feature perturbations and eroded feature perturbations for semi-supervised segmentation of medical images. Furthermore, his work reveals an empirical link between the consistency regularization and the model calibration for the first time. This work also led to publications at MIDL 2022, IEEE TMI, as well as a patent filing.
RkJQdWJsaXNoZXIy NTc3NzU=