Hugo J. Kuijf1, Manon Brundel2, Max A. Viergever1, Geert Jan Biessels2, Koen L. Vincken1
1Image Sciences Institute, University Medical Center Utrecht, Utrecht, Netherlands; 2Department of Neurology, Rudolf Magnus Institute of NeuroScience, UMC Utrecht
The current standard for cerebral microbleed detection is visual rating, which is laborious and has limited reproducibility. Semi-automated methods for detection of microbleeds on clinical images have been suggested before, but suffer from a large number of false positives. In this study, we present a method for microbleed detection based on the radial symmetry transform. This results in a high sensitivity of 75% and a limited number of false positives, requiring just two minutes of rater time to censor them. Furthermore, the method outperforms other known methods in terms of specificity.