Spatiotemporal Independent Component Analysis for Retinal Images

This paper presents a new Independent Component Analysis algorithm (ICA-P) for modeling and measuring physiological responses to optical stimulation from a new, non-invasive optical imaging device. The ICA-P algorithm uses prior information on the visual stimulus to provide improved detection performance. In comparison with other methods that were considered (JADE, Infomax, SOBI, ESD, and Fast-ICA), ICA-P is shown to achieve significantly improved performance in detecting the retinal response to visual stimuli. Results from simulations show that we can estimate retinal reflectance changes as small as 0.01% (-40 dB, 10 dB lower than the other considered methods).