Advances in inverse problems of medical imaging
|
Organizer(s): |
Name:
|
Affiliation:
|
Country:
|
Samuli Siltanen
|
University of Helsinki
|
Finland
|
|
|
|
|
|
|
|
|
|
Introduction:
| | Inverse problems arise from many medical imaging modalities: X-ray tomography, nuclear magnetic resonance imaging, emission tomography, diffuse tomography and various hybrid approaches. The image reconstruction task is often both nonlinear and ill-posed, and even the linear imaging modalities become difficult when the data is severely limited. Pushing these methods to clinical practice involves both theoretical and computational challenges.
Recently, machine learning has enabled powerful new reconstruction methods and spectacular advances. However, medical imaging requires a high degree of interpretability and reliability. Furthermore, ill-posedness is a serious challenge for all algorithms, including neural network models.
Therefore, there is a fresh avenue of investigation: combining classical inversion mathematics with machine learning. Optimally, one could reap the benefits of AI while retaining the solid foundation of mathematical analysis. Targeted deep learning approaches minimize the black-box aspect of learning and maximize the provable steps of the method.
This session presents recent progress in medical imaging inverse problems, with various mixtures of machine learning and traditional mathematics.
|
|
|
|
|
|
