Detection Performance in Ultrasound Super-Resolution Imaging

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review


Many thresholding and localization methods have been developed in the field of image processing, but their effect has not been investigated and compared with each other for microbubble (MB) detection from ultrasound data. This work compares the detection performance in two sets of simulated data for the ULTRA-SR challenge with 15 thresholds and 2 localization methods. The best method based on the test dataset was Yen's adaptive threshold followed by localization using local maxima with parabolic interpolation. The true positive (TP), false positive (FP), and false negative (FN) detections using the Yens threshold compared with optimal fixed threshold were (TP:2638/2634, FP: 3508/3512, FN: 705/708) in high frequency data set and (TP:130/112, FP: 658/676, FN: 2380/180) in low frequency data set. The fixed threshold was optimal in Jaccard similarity coefficient (JSC) sense. This means that the threshold was adjusted to have the maximum JSC = TP/(TP+FP+FN) based on the ground truth. The Yens threshold with parabolic interpolation had the closest metrics to the optimal fixed threshold for both high and low frequency test dataset. The challenge's final synthetic and in vivo dataset were processed based on the selected method and their results are also shown and discussed in this article.
Original languageEnglish
Title of host publicationProceedings of 2022 IEEE International Ultrasonics Symposium (IUS)
Number of pages4
Publication date2022
ISBN (Print)978-1-6654-7813-7
ISBN (Electronic)978-1-6654-6657-8
Publication statusPublished - 2022
Event2022 IEEE International Ultrasonics Symposium - Venice, Italy
Duration: 10 Oct 202213 Oct 2022


Conference2022 IEEE International Ultrasonics Symposium


  • Adaptive threshold
  • Detection
  • Interpolation
  • Localization
  • Microscopy
  • Super-resolution
  • Tracking
  • Ultrasound


Dive into the research topics of 'Detection Performance in Ultrasound Super-Resolution Imaging'. Together they form a unique fingerprint.

Cite this