Premium
Automatic selection of arterial input function using cluster analysis
Author(s) -
Mouridsen Kim,
Christensen Søren,
Gyldensted Louise,
Østergaard Leif
Publication year - 2006
Publication title -
magnetic resonance in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.696
H-Index - 225
eISSN - 1522-2594
pISSN - 0740-3194
DOI - 10.1002/mrm.20759
Subject(s) - computer science , fully automatic , cerebral blood flow , reproducibility , pattern recognition (psychology) , artificial intelligence , algorithm , mathematics , statistics , medicine , cardiology , mechanical engineering , engineering
Quantification of cerebral blood flow (CBF) using dynamic susceptibility contrast MRI requires determination of the arterial input function (AIF) representing the delivery of intravascular tracer to tissue. This is typically accomplished manually by inspection of concentration time curves (CTCs) in regions containing the ICA, VA, and MCA. This is, however, a time consuming and operator dependent procedure. We suggest a completely automatic procedure for establishing the AIF based on a cluster analysis algorithm. In 20 normal subjects CBF maps calculated in 2 slices by the automatic procedure were compared to maps obtained with AIFs selected individually by 7 experienced operators. The average manual to automatic CBF ratio was 1.03 ± 0.15 in the lower slice and 1.05 ± 0.12 in the upper slice, demonstrating excellent agreement between the manual and automatic method. The algorithm provides means for objectively assessing AIF candidates in local AIF search algorithms designed to reduce bias due to delay and dispersion. Given the reproducibility and speed (10 s) of the automatic method, we speculate that it will greatly improve the accuracy of perfusion images and facilitate their use in clinical diagnosis and decision‐making, particularly in acute stroke but also in cerebrovascular disease in general. Magn Reson Med, 2006. © 2006 Wiley‐Liss, Inc.