Thomas Küstner1,2, Martin Schwartz1,2, Annika Kaupp2, Petros Martirosian1, Sergios Gatidis1, Nina F. Schwenzer1, Fritz Schick1, Holger Schmidt1, and Bin Yang2
images are usually analyzed by a human observer (HO) according to a certain
diagnostic question. Flexible algorithm parametrization and the enormous amount
of data created per patient make this task time-demanding and expensive.
Furthermore, definition of objective quality criterion can be very challenging,
especially in the context of a missing reference image. In order to support the
HO in assessing image quality, we propose a non-reference MR image quality
assessment system based on a machine-learning approach with an Active Learning
loop to reduce the amount of necessary labeled training data. Labeling is
performed via an easy accessible website.