MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Deep Learning for Image Reconstruction and Processing

J Teuwen1*, A Hauptmann2*, M Kachelriess3*, (1) Radboud University Medical Center, Nijmegen, Netherlands, (2) University of Oulu, Oulu, Finland, (3) DKFZ Heidelberg




Presentations

(Thursday, 7/18/2019) 10:00 AM - 12:00 PM

Room: 221AB

The applications of deep learning to the fields of image reconstruction and processing are manifold and, potentially, of high impact in the near future. Image reconstruction describes the process for which the volumetric representation of an object is computed from the measured data. In a classical approach, one relies on domain specific knowledge contained in physical-analytical models to develop a reconstruction algorithm, which is often given by a certain iterative refinement procedure. Recent research seeks to develop deep learning based methods, in particular convolutional neural networks (CNN) and combine these with model-based reconstruction techniques. Next to this machine learning methods offer methods to impute missing data and speed up computations by fitting a neural network to computationally intensive tasks.
This lecture will illustrate these techniques with different applications: we will discuss how deep learning networks can be used to speed up reconstructions for real-time applications in radiation oncology and cardiovascular magnetic resonance imaging. Moreover, we will see how these models can be used for patient specific dosimetric models in digital breast tomosynthesis. Finally, we will review and critically discuss methods intended to replace missing data, to replace time-consuming computations and methods to incorporate a priori knowledge. Moreover the possibility to use deep learning to predict CT dose distributions in real-time is being highlighted and discussed, as it may be of importance for novel image-quality and dose-optimized CT scan protocols.

Learning objectives:
1) Learn about the state-of-the art deep learning methods to reconstruct medical images.
2) Understand the wide range of problems in image formation machine learning techniques can be applied to.
3) Learn to critically assess the possible impact and problems associated with the implementations of these techniques in daily practice.

Handouts

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email