Deep 2D-3D Modeling and Learning in Medical Image Computing
A MICCAI 2021 Tutorial
September 27th, 14:00-18:00 UTC
About
The goal of this half-day MICCAI2021 tutorial is to promote research of deep 2D-3D modeling and learning in medical image computing. Specifically, the participants will gain an understanding to effectively utilize 2D and 3D imaging data in deep learning systems on a variety of tasks, including image reconstruction, registration, segmentation, etc.
We will begin by introducing imaging physics on modalities of interests, and several tools/formulations that enable differentiable computation between 2D-3D images. We will then present a survey on methods on 2D-3D modelling both with and without differentiable operators. Finally, speakers will present more detailed spotlight talks on 2D-3D modelling in specific domains and tasks.
To best adapt to the online format, we will employ a combination of pre-recorded talks and live Q&A discussions. Please refer to the schedule section for video and live discussion links, which will be updated at the time of the tutorial.
Material and videos about the tutorial will be released on this website after the conclusion of MICCAI 2021.
Organizers
Cheng Peng
Johns Hopkins University
Amazon AWS AI
Stanford University
Johns Hopkins University
University of Science and Technology of China
Learning Objectives
To gain basic understanding of the relationship between observed acquisitions and reconstructed images for various modalities.
To survey recent literature that explore such relationships in a learning system, both implicitly and explicitly, on a variety of tasks
To understand the advantages of 2D-3D modelling with differentiable operators.
To look ahead for potential directions for improvement and take inspiration from the natural image counterparts.
Topics
Medical imaging physics
Medical image 2D-3D modelling
Modelling without differentiable connections
Modelling through differentiable ray casting
Modelling through projective transformation
Modelling through implicit representation
Relevant tasks: reconstruction, synthesis, segmentation, registration
Schedule
Opening introduction
S. Kevin Zhou, 14:00 -14:10 UTC
Survey on 2D-3D modelling
Cheng Peng, 14:10 - 14:50 UTC
Introduction on imaging physics
Non-differentiable 2D-3D modelling
Differentiable 2D-3D modelling
Exploiting medical imaging for deep medical image computing
Haofu Liao, 15:00 - 15:40 UTC
Exploiting medical image measurement
Leveraging 3D knowledge to solve problems in 2D
Ultrasound measurement - cardiac structure segmentation
X-ray measurement - novel view synthesis and bone suppression
Exploiting medical image reconstruction
Co-learning knowledge in different domains
CT reconstruction - metal artifact reduction
MRI reconstruction - undersampled MRI reconstruction
Live discussion with Cheng and Haofu
Cheng Peng, Haofu Liao, 15:40 - 16:00 UTC
Zoom Link: please refer to https://miccai2021.pathable.eu
Coffee break 16:00 - 16:10 UTC
Exploiting prior knowledge for medical image reconstruction
Liyue Shen, 16:10 - 16:50 UTC
Data-driven Reconstruction for Medical Imaging
Physics-informed Medical Image Reconstruction with DL
Prior-integrated Medical Image Reconstruction
Survey & Perspective on 2D-3D medical image registration
Mathias Unberath, 17:00 - 17:40 UTC
Latest methods on 2D-3D medical image registration
Live discussion with Liyue and Mathias
Liyue Shen, Mathias Unberath, 17:40 - 18:00 UTC
Zoom Link: please refer to https://miccai2021.pathable.eu