Detecting fixation on a target using time-frequency distributions of a retinal birefringence scanning signal

Background: The fovea, which is the most sensitive part of the retina, is known to have birefringent properties, i.e. it changes the polarization state of light upon reflection. Existing devices use this property to obtain information on the orientation of the fovea and the direction of gaze. Such devices employ specific frequency components that appear during moments of fixation on a target. To detect them, previous methods have used solely the power spectrum of the Fast Fourier Transform (FFT), which, unfortunately, is an integral method, and does not give information as to where exactly the events of interest occur. With very young patients who are not cooperative enough, this presents a problem, because central fixation may be present only during very short-lasting episodes, and can easily be missed by the FFT. Method: This paper presents a method for detecting short-lasting moments of central fixation in existing devices for retinal birefringence scanning, with the goal of a reliable detection of eye alignment. Signal analysis is based on the Continuous Wavelet Transform (CWT), which reliably localizes such events in the time-frequency plane. Even though the characteristic frequencies are not always strongly expressed due to possible artifacts, simple topological analysis of the time-frequency distribution can detect fixation reliably. Results: In all six subjects tested, the CWT allowed precise identification of both frequency components. Moreover, in four of these subjects, episodes of intermittent but definitely present central fixation were detectable, similar to those in Figure 4. A simple FFT is likely to treat them as borderline cases, or entirely miss them, depending on the thresholds used. Conclusion: Joint time-frequency analysis is a powerful tool in the detection of eye alignment, even in a noisy environment. The method is applicable to similar situations, where short-lasting diagnostic events need to be detected in time series acquired by means of scanning some substrate along a specific path.