A comparison of signal detection techniques in the multifocal electroretinogram

A common task in the analysis of the multifocal electroretinogram (mfERG) is determining which retinal areas have preserved signal in recordings which are attenuated by the effects of disease. Several automated methods have been proposed for signal detection from multifocal recordings, but no systematic study has been published comparing the performance of each. This article compares the sensitivity and specificity of expert human scoring with three different automated methods of mfERG signal detection. Recordings from control subjects were artificially modified to simulate decrease in signal amplitudes (attenuation) as well as total signal loss. Human scorers were able to identify areas with preserved signal at both low and high attenuation levels with a high specificity (minimum 0.99), sensitivities ranged from 0.2 to 0.94. Automated methods based on template correlation performed better than chance at all attenuation levels, with a slide fit method having the best performance. Signal detection based on signal to noise ratio performed poorly. In conclusion automated methods of signal detection can be used to increase signal detection sensitivity in the mfERG.