Page 438 - 8th European Congress of Mathematics ∙ 20-26 June 2021 ∙ Portorož, Slovenia ∙ Book of Abstracts
P. 438
EU-MATHS-IN: MATHEMATICS FOR INDUSTRY IN EUROPE (MS-66)
Quantitative Comparison of Deep Learning-Based Image Reconstruction
Methods for Low-Dose and Sparse-Angle CT Applications
Johannes Leuschner, jleuschn@uni-bremen.de
University of Bremen, Germany
Coauthors: Maximilian Schmidt, Poulami Somanya Ganguly, Vladyslav Andriiashen,
Sophia Bethany Coban, Alexander Denker, Daniel Otero Baguer, Dominik Bauer,
Amir Hadjifaradji, Kees Joost Batenburg, Peter Maass, Maureen van Eijnatten
Over the last years, deep learning methods have significantly pushed the state-of-the-art results
in applications like imaging, speech recognition and time series forecasting. This development
also starts to apply to the field of computed tomography (CT). In medical CT, one of the main
goals lies in the reduction of the potentially harmful radiation dose a patient is exposed to during
the scan. Depending on the reduction strategy, such low-dose measurements can be more noisy
or starkly under-sampled. Hence, achieving high quality reconstructions with classical methods
like filtered back-projection (FBP) can be challenging, which motivated the investigation of a
number of deep learning approaches for this task.
With the purpose of comparing such approaches fairly, we evaluate their performance on
fixed benchmark setups. In particular, two CT applications are considered, for both of which
large datasets of 2D training images and corresponding simulated projection data are publicly
available: a) reconstruction of human chest CT images from low-intensity data, and b) recon-
struction of apple CT images from sparse-angle data.
In order to include a large variety of methods in the comparison, we organized open chal-
lenges for either task. Our current study comprises results obtained with popular deep learning
approaches from various categories, like post-processing, learned iterative schemes and fully
learned inversion. The test covers image quality of the reconstructions, but also aspects such as
the required data or model knowledge and generalizability to other setups are considered. For
reproducibility, both source code and reconstructed images are made publicly available.
MACSI and industrial mathematics in Ireland
Kevin Moroney, kevin.moroney@ul.ie
University of Limerick, Ireland
MACSI was established on foot of a Science Foundation Ireland grant so-called mathematics
initiative grant in 2006. As a result, a fulltime business manager was employed, an industrial
consultancy was established, and European study groups were brought to Ireland for the first
time. Building on this, we have recently established the first Centre for Research Training in
mathematics in the country (also supported by Science Foundation Ireland).
436
Quantitative Comparison of Deep Learning-Based Image Reconstruction
Methods for Low-Dose and Sparse-Angle CT Applications
Johannes Leuschner, jleuschn@uni-bremen.de
University of Bremen, Germany
Coauthors: Maximilian Schmidt, Poulami Somanya Ganguly, Vladyslav Andriiashen,
Sophia Bethany Coban, Alexander Denker, Daniel Otero Baguer, Dominik Bauer,
Amir Hadjifaradji, Kees Joost Batenburg, Peter Maass, Maureen van Eijnatten
Over the last years, deep learning methods have significantly pushed the state-of-the-art results
in applications like imaging, speech recognition and time series forecasting. This development
also starts to apply to the field of computed tomography (CT). In medical CT, one of the main
goals lies in the reduction of the potentially harmful radiation dose a patient is exposed to during
the scan. Depending on the reduction strategy, such low-dose measurements can be more noisy
or starkly under-sampled. Hence, achieving high quality reconstructions with classical methods
like filtered back-projection (FBP) can be challenging, which motivated the investigation of a
number of deep learning approaches for this task.
With the purpose of comparing such approaches fairly, we evaluate their performance on
fixed benchmark setups. In particular, two CT applications are considered, for both of which
large datasets of 2D training images and corresponding simulated projection data are publicly
available: a) reconstruction of human chest CT images from low-intensity data, and b) recon-
struction of apple CT images from sparse-angle data.
In order to include a large variety of methods in the comparison, we organized open chal-
lenges for either task. Our current study comprises results obtained with popular deep learning
approaches from various categories, like post-processing, learned iterative schemes and fully
learned inversion. The test covers image quality of the reconstructions, but also aspects such as
the required data or model knowledge and generalizability to other setups are considered. For
reproducibility, both source code and reconstructed images are made publicly available.
MACSI and industrial mathematics in Ireland
Kevin Moroney, kevin.moroney@ul.ie
University of Limerick, Ireland
MACSI was established on foot of a Science Foundation Ireland grant so-called mathematics
initiative grant in 2006. As a result, a fulltime business manager was employed, an industrial
consultancy was established, and European study groups were brought to Ireland for the first
time. Building on this, we have recently established the first Centre for Research Training in
mathematics in the country (also supported by Science Foundation Ireland).
436
