Laser in situ keratomileusis nomogram development.

BACKGROUND: There is no standard laser in situ keratomileusis (LASIK) nomogram for the Nidek EC-5000 excimer laser when used with current software versions 2.23 or 2.25. Nomogram tables developed by various EC-5000 users differ significantly from one another. METHODS: We analyzed the results of our first 138 eyes treated for myopia and myopic astigmatism with the Nidek EC-5000 excimer laser, model 2b, long working distance. In addition to analyzing standard clinical outcomes, the correction values entered into the laser were compared to the achieved change in refraction for these eyes. A global nomogram parameter was created, incorporating the effect of optical zone size and patient age. After observing initial overcorrections, we elected to treat most of the eyes using a 20% reduction of the attempted spherical correction. RESULTS: The mean value for the attempted change in spherical equivalent refraction was -6.85 D. This correction was achieved with a standard deviation of -0.04 D at 1 month, -0.08 D at 3 months, and -0.09 D at 6 months after LASIK. The postoperative values for the uncorrected and spectacle-corrected visual acuity were similar to other published studies. However, scatterplots comparing the laser settings to achieved postoperative refractions showed a clear 20% trend toward overcorrection. This trend increased with patient age and optical zone diameter. CONCLUSIONS: The basic calibration values (factory standard) for software versions 2.23 and 2.25 are too aggressive for LASIK treatments. When using optical zone diameters of 6 mm and above, the spherical refraction values entered into the laser must be reduced by at least 20%. LASIK nomograms for the Nidek EC-5000 excimer laser should be specific for optical zone diameter and patient age.

A neural network classifier in experimental particle physics.

A classification problem in high energy physics has been solved on simulated data using a simple multilayer perceptron comprising binary units which was trained with the CHIR algorithm. The unstable training of such a network on a nonseparable set has been overcome by selecting those weight vectors with good performance while providing a flexible choice of the two types of classification errors. Specific features of the problem have been exploited in order to simplify and optimize the solution which has been compared to the popular backpropagation algorithm and found to perform on a similar level. Additional aspects of this work are the use of the CHIR algorithm on continuous input and incorporating the classic idea of a phi-machine in a multilayer perceptron.