Learning procedure in a neural control model for the urinary bladder

A continuous neural network coupled to a dynamical model of the urinary bladder is defined. The neural network is trained to control the bladder model to track a prescribed volume fluctuation, by adjusting weights and time constants. The gradients of the error in the output neurons of the neural network are unknown. Therefore, the learning procedure discussed here minimizes the error functional without using gradient descent.     

Two Methods for Optimal MECG Elimination and FECG Detection from Skin Electrode Signals

Two signal processing techniques for the suppression of the maternal ECG and simultaneously optimal detection of fetal ECG with respect to noise are presented. Both techniques are based on the singular value decomposition of a measurement matrix. Criteria are given in order to evaluate, a priori, electrode locations and sampling schemes for both methods. A fundamental difference with other methods is that the number of linearly independent FECG signals is not constrained to one. One of the presented techniques is a typical offline method. It is well suited for a large number of electrodes and large number of samples, which results in a better signal to noise ratio. The second technique is a typical on-line method. It gives fetal ECG signals within about 1 s, and is adaptive to changes of the transfer (e.g., due to fetal movement). It can be applied with a small number of electrodes (e.g., eight). It is shown that if three of these signals are from thoracic electrodes, the MECG suppression is guaranteed.     

Scan optimization of gadolinium contrast-enhanced three-dimensional MRA of peripheral arteries with multiple bolus injections and in vitro validation of stenosis quantification

OBJECTIVE: Development of a method for semiautomated preparation of purified, representative and conventionally stained monolayer smears from bronchial secretions suitable for subjective and/or automated cytodiagnosis. STUDY DESIGN: Bronchial secretions from 50 patients with and 48 without carcinoma cells of different types were collected in Saccomanno's fixative. After routine pick-and-smear processing, residual material was subjected to a mucolytic agent (ammonium thioglycolate). Separation of cells was performed by differential centrifugation through aqueous sucrose. The pellet was automatically processed by the AutoCyte PREP system. RESULTS: Slides revealed well-preserved, slightly shrunken, homogeneously distributed cells devoid of mucus, cellular debris and bacteria in monolayer arrangement nearly without overlap. Granulocytes were eliminated to a large extent. Comparison with pick-and-smear specimens showed more tumor cells per square centimeter of slide surface in 100% of AutoCyte PREP slides. The number of tumor cells per AutoCyte PREP slide was higher in 46% and lower in 54%. Selecting slides at random and requiring at least 10 abnormal cells to establish a tumor diagnosis were achieved in 82.7% if only one, in 88.0% if two and 94.0% if seven or eight AutoCyte PREP slides were investigated. CONCLUSION: The semiautomated method yielded conventionally stained, purified monolayer smears from bronchial secretions with cellular morphology suitable for evaluation by cytologists and screening machines. Representativity of AutoCyte PREP monolayers was superior to that of pick-and-smear slides.