No abstract

No abstract

-A 2-layer neural network was applied to determine the LV in radionuclide ventriculography. Alter learning by back-propagation, the correlations between computed pictures and learning data-set outputs, and between learning data-set outputs and other pictures were excellent (rz0.92 and r=0.83 respectively). I -INTRODUCI'IONNuclear medicine has always been precursory in the image processing field because of the intrinsic features of the pictures. Hence, we planned to apply a 2-layer perceptron to detect the left ventricle (LV) in radionuclide ventriculography.The goal of this paper is to present the results of our device, which final aim is to determine automatically the Left Ventricular Ejection Fraction (LVEF) and the time-activity curve being clinical parameters of cardiac function. -NEURAL NETWORK S T R U mThe neural network consists of two layers designed by an intermediate and an output layer. Each layer has 1024 neurons and each neuron has 1024 weights and a bias (which is not an output of the previous layer) as inputs. The transfer function of each neuron is a sigmoid. The output of each neuron is the sum of the weighed inputs added to the bias, modulated by the transfer function. Each neuron is connected with all neurons of the previous layer.