This paper presents an efficient and accurate approach to myocardium extrac
tion in Positron Emission Tomography (PET) images based on a careful applic
ation of soft computing techniques. PET images present a noisy background,
making the automatic myocardium extraction and uptake quantification a diff
icult task. In this work a Self Organized Radial Basis Function Network (SR
BFN) is designed to focus on the myocardium in an iterative process until t
he total extraction of the myocardium from the noisy background is achieved
. Fuzzy sets and fuzziness measures are used to compute the error of the ne
twork. The method was tested on a set of nine images of different patients
and its effectiveness is illustrated in two patients showing tracer uptake
defects. (C) 2001 Elsevier Science Ltd. All rights reserved.