Various sources of variability, such as speckle noise, depth dependence and inhomogeneous intervening tissue, are involved in B-mode images, even when using the same ultrasonic equipment with fixed settings. The behavior of these sources of variability was investigated by texture analysis of images obtained from simulations and from a tissue-mimicking phantom, a normal adult liver and a pediatric renal (Wilms') tumor. First-order statistics (MEAN and SNR) and second-order statistics from the co-occurrence matrix (ENT and COR) were calculated. In a phantom, the SNR and ENT show a clear depth dependence. In biological tissue, the variability is mainly caused by the speckle noise and inhomogeneous intervening tissue. In addition, almost the entire range of the COR feature is present in images of liver and tumor. Furthermore, all the features calculated in windows of 1 cm2 exhibit an overlap among the different media. With the second-order features, it is possible to discriminate 85% reliable (average) between the normal, adult, liver and the pediatric renal tumor above a window size of 9 cm2. The SNR can not discriminate between these tissues. The maximum resolution of 9 cm2 reveals a serious limitation of parametric imaging. Finally, the features reproduce well in the case of follow-up of an abdominal tumor during chemotherapy.

On-line learning in layered perceptrons is often hampered by plateaus in the time dependence of the performance. Studies on backpropagation in networks with a small number of input units have revealed that correlations between subsequently presented patterns shorten the length of such plateaus. We s h o w h o w to extend the statistical mechanics framework to quantitatively check the e ect of correlations on learning in networks with a large number of input units. The surprisingly compact description we obtain makes it possible to derive properties of on-learning with correlations directly from studies on on-line learning without correlations.

No abstract

No abstract