Abstract
Purpose:
Static Vessel Analysis is an accepted tool for risk evaluation of cardiac diseases based on retinal vessel diameters analyzed in fundus images. One reason for uncertainties between repetitions is the biological temporal variation of the vessels. Standard deviation (SD) between repetitions can be reduced by analyzing multiple images per measurement. We developed sampling paradigms using additional information about vessel activities and investigated their advantage in contrast to randomized acquired samples.
Methods:
Main signal components in the temporal variation of retinal vessel diameters were determined from literature. With respect to the Mayer waves as main components of the biological variation of the vessel diameters, paradigms were developed to sample the waves with phase shifts of 180 degree. For the first evaluation, a dataset of signals containing the main frequencies was simulated, considering the individual variation of these frequencies. The signal was sampled several times according to the developed paradigms and the mean value was calculated. SD between multiple measurements was determined and compared to SD of measurements with randomized sampling times. The influence of the precision of knowing the signal frequencies was observed. Following this, continuous vessel diameters were analyzed, recorded by Dynamic Vessel Analyzer under mydriatic conditions. The sampling paradigms were applied to these real vessel diameter data.
Results:
The acquisition and evaluation of multiple images improves the reproducibility of the determination of the vessel diameters. Our simulations reveal that SD between repetitions is reduced by 1/√n when randomly sampling n images, whereas sampling with the developed paradigms further reduces SD, in particular for small numbers of images. The benefit is high even with variations of the main signal frequencies. With an increasing number of samples, improvement decreases again. These results could be verified on measurement data.
Conclusions:
According to our investigations, uncertainties induced by temporal biological variations can be highly reduced by the acquisition of small numbers of images taken with dedicated sampling paradigms. By recording and evaluating of a small image series repeatability can be considerably increased. These images can be taken with a fundus camera under non-mydriatic conditions with a low strain on the patients.