Detectability index describes the information conveyed by sonographic images

Nghia Q. Nguyen, Craig K. Abbey, Michael F. Insana

Research output: Chapter in Book/Report/Conference proceedingConference contribution


We have been developing the ideal observer formalism for sonography, which is based on the best-possible diagnostic performance. The ideal performance was compared to that of trained human observers to estimate the visual efficiency for discriminating lesion features. We find that humans are generally less than 10% efficient at accessing visual information essential for breast cancer diagnosis. In seeking ways to improve this process, we must first establish a connection between standard ROC observer metrics and instrument properties used in system design. In radiography, that relationship is made through the lesion signal-to-noise ratio SNR I. SNR I 2, which describes task information, is simply related to contrast and spatial resolutions and noise power. Those relations break down for sonography due to the quadratic form of the ideal observer. Our goal in this paper is to establish a rigorous connection between ideal performance and engineering design metrics, which has directly applications for sonographic system design and optimization.

Original languageEnglish (US)
Title of host publication2011 IEEE International Ultrasonics Symposium, IUS 2011
Number of pages4
StatePublished - 2011
Event2011 IEEE International Ultrasonics Symposium, IUS 2011 - Orlando, FL, United States
Duration: Oct 18 2011Oct 21 2011

Publication series

NameIEEE International Ultrasonics Symposium, IUS
ISSN (Print)1948-5719
ISSN (Electronic)1948-5727


Other2011 IEEE International Ultrasonics Symposium, IUS 2011
Country/TerritoryUnited States
CityOrlando, FL


  • Breast sonography
  • ideal observer
  • image quality
  • Kullback-Leibler divergence
  • task-based design

ASJC Scopus subject areas

  • Acoustics and Ultrasonics


Dive into the research topics of 'Detectability index describes the information conveyed by sonographic images'. Together they form a unique fingerprint.

Cite this