This course covers the use of untrained neural networks for MR reconstruction through self-training and weak supervision techniques. The learning outcomes include understanding how untrained neural networks can be utilized for image reconstruction, particularly in MR imaging, and how to improve their inference time using weakly-labeled data. The course teaches skills such as implementing ConvDecoder for generating weakly-labeled data, training unrolled neural networks with limited supervised pairs, and comparing the performance of different methods in the limited data regime. The teaching method involves a lecture format with a focus on research findings and practical applications. The intended audience for this course includes individuals interested in medical imaging, machine learning, and data-efficient methods for solving inverse problems in imaging.
Self-Training - Weak Supervision Using Untrained Neural Nets for MR Reconstruction - Beliz Gunel
Stanford University via YouTube
Overview
Syllabus
Intro
Inverse Problems in Imaging
ML Methods for MR Reconstruction
Key Observations & Current Challenges
Motivation Can we significantly reduce the large paired training dataset requirement for
Self-Training in Natural Language Processing
Self-Training for MRI Reconstruction
Untrained Neural Networks (Deep Image Prior)
Untrained Neural Networks (ConvDecoder)
Key Observations & Ongoing Work
We know how to simulate motion
Standardization of ML pipelines matter
Self-supervised learning methods trained in-domain can learn good image-level representations for MR images
Taught by
Stanford MedAI
