PhD defence by Jonathan Scharff Nielsen

On 9 May 2019 Jonathan Scharff Nielsen will defend his PhD thesis "Computed tomography metal artifact reduction  for radiotherapy using magnetic resonance imaging".

Time: 9 May 2019, at 12:30

Place: Bldg.308, auditorium 13

Supervisor: Professor Koen Van Leemput, DTU Health Tech


Professor Anders Bjorholm Dahl, DTU Compute

Professor Johan Nuyts, K.U. Leuven

Professor Per Rugaard, Aarhus University Hospital


Chairperson at defence:

Associate Professor Emil Boye Kromann, DTU Health Tech


When a medical patient contains metal implants, the computed tomography (CT) image may be corrupted by metal artifacts in the shape of potentially severe streaks and so-called cupping artifacts. The quantitative errors associated with these artifacts may be a major problem for external beam radiotherapy (RT) applications, where CT images are used to simulate the radioactive dose to the patient for treatment planning. This is especially true in the increasingly widespread but error-sensitive proton therapy, and as a consequence some patients that would otherwise benefit from proton RT may be disqualified from it.

Metal artifact reduction (MAR) algorithms are therefore employed, which seek to remove the artifacts either by improving the model used to reconstruct the data or through data pre- or post-processing steps. Despite at least 40 years of development, such methods are however still inadequate.

This thesis investigates a promising way to increase the efficacy of MAR: by capitalizing on the superior anatomical information in the corrupted region that may be found in a coacquired Magnetic Resonance Image (MRI), which are often acquired for tumor or soft tissue delineations and present far more localized metal artifacts, the uncorrupted CT values may be predicted using probabilistic modelling and machine learning. This is a challenging task due in particular to the weak correlation between CT values and MRI intensities, and so we develop a predictive model that is particularly suited to handling this problem; it in particular references both anatomical features in the MRI and the uncertain information in the artifact corrupted CT values.

We use this model to define several MAR algorithms and investigate their performance compared to standard, clinically used algorithms. We find promising results in terms of improving both the CT image quality and the dose calculation accuracy in, especially, proton therapy. 


tor 09 maj 19
12:30 - 15:30


DTU Sundhedsteknologi


Bldg. 3018, aud. 13