New pediatric vision screener, part II: electronics, software, signal processing and validation

Abstract

Background: We have developed an improved pediatric vision screener (PVS) that can reliably detect central fixation, eye alignment and focus. The instrument identifies risk factors for amblyopia, namely eye misalignment and defocus.

Methods: The device uses the birefringence of the human fovea (the most sensitive part of the retina). The optics have been reported in more detail previously. The present article focuses on the electronics and the analysis algorithms used. The objective of this study was to optimize the analog design, data acquisition, noise suppression techniques, the classification algorithms and the decision making thresholds, as well as to validate the performance of the research instrument on an initial group of young test subjects-18 patients with known vision abnormalities (eight male and 10 female), ages 4-25 (only one above 18) and 19 controls with proven lack of vision issues. Four statistical methods were used to derive decision making thresholds that would best separate patients with abnormalities from controls. Sensitivity and specificity were calculated for each method, and the most suitable one was selected.

Results: Both the central fixation and the focus detection criteria worked robustly and allowed reliable separation between normal test subjects and symptomatic subjects. The sensitivity of the instrument was 100 % for both central fixation and focus detection. The specificity was 100 % for central fixation and 89.5 % for focus detection. The overall sensitivity was 100 % and the overall specificity was 94.7 %.

Conclusions: Despite the relatively small initial sample size, we believe that the PVS instrument design, the analysis methods employed, and the device as a whole, will prove valuable for mass screening of children.

Figures

Fig. 1

The human fovea (the small area of the retina responsible for sharp central vision) surrounded by uniquely arranged nerve fibers (Henle fibers). These fibers transmit the information to the optic disc and to the brain. The Henle fibers are birefringent, causing polarization change in the reflected light. The polarization change depends on the angle between the orientation of the fiber and the plane of polarization of the incoming light. Shown is the 3° scanning circle. When the center of the fovea coincides with the center of the scanning circle, central fixation is attained and detected. In the simplest case, during central fixation, the frequency of the returning signal would be twice the scanning frequency. Note that it is the fovea that is moving, and not the scanning circle. Retinal birefringence scanning (RBS) is a technique that uses the changes in the polarization of light returning from the eye to detect the projection into space of the array of Henle fibers surrounding the fovea

Fig. 2

A simplified diagram of the pediatric vision screener (shown is one eye only). Linearly polarized light emitted continuously by a 785-nm laser diode is transmitted by a plate polarizing beamsplitter (PBS) toward a half-wave-plate (HWP) that is spun by a motor using a pulley ratio to achieve a rotation 9/16ths as fast as the scan. After passage through the rotating HWP, the beam of continuously rotating linearly polarized light enters the scanning unit that consists of two gold-plated plane mirrors. The retina is scanned by the spot of laser light in a circle subtending a visual angle of 3° in diameter. A small percentage of light reflected from each ocular fundus is re-imaged back, following the same light path it originally came from, via the principle of conjugacy. The unchanged part of the returning light, in other words the part with the same polarization as the original light, is transmitted through the PBS, back toward the light source, thus never making it to the detection unit. The changed part of the returning light, on the other hand, is reflected by the PBS toward the photodetector assembly, consisting of a bull’s-eye photodetector (BEPDs). A band pass filter assures that only light in the desired wavelength range reaches the detectors. The graph in the right-hand bottom corner shows the generated frequencies for central and paracentral fixation

Fig. 3

Electronics of the PVS instrument. The four signals coming from the two BEPDs (center and annulus for each eye) are amplified and filtered by a 4-channel, 4-stage programmable-gain amplifier. The amplified signals are then fed to the analog-to-digital converter (ADC). The instrument was developed with two versions of computer support: a a luggable (“lunchbox”) PC, and b a 32-bit digital signal processing (DSP) system. The first (PC-based) configuration is easier to use for development, software debugging, algorithm validation, graphical user interface, PC graphics, etc. The second (DSP-based) configuration, was designed as a CPU support for an industrial prototype, with easily reconfigurable FPGA hardware

Fig. 4

The analog filter G(f) of each of the four analog channels. Shown are the frequencies to be detected. Note that the 4.5 fs frequency of relatively high amplitude needed less gain, to avoid saturation of the analog channels

Fig. 5

The phase-shift-subtraction method. The signal is subtracted from itself shifted by one period of the scanning system. This removes the background noise, while the useful odd multiples of half the scanning frequency (shown is only 2.5 fs) are doubled in amplitude

Fig. 6

Deriving the A–D sampling rate and the step-motor pacing pulses from the power line frequency. In order to eliminate the power line interference along with the instrumental noise in the PhSS procedure, the motor was paced with a frequency which is a precise multiple of the 60 Hz power line frequency: 6000 steps/sec (100 × 60 Hz). The power line frequency was obtained by means of a transformer with 4 kV isolation. Multiplication of the power line frequency by a factor of 100 is achieved by means of a phase-locked-loop (PLL) circuit. Thus, the motor is spinning at a speed of 6000/200 = 30 rotations per second. At each new step, one sample from each channel is acquired, i.e. every new data sample corresponds to 1.8° of rotation (1 step). Each of the 200 samples during one scanning rotation corresponds to one particular angle of rotation, with the angles spaced evenly at 1.8° intervals. At the same time, there are exactly two complete power line cycles (60 Hz) in one scanning cycle (fs = 30 rps), which means that power line noise is eliminated by the 360° PhSS technique

Fig. 7

The examination procedure. Without head restraint, the child is seated on a chair or in the parent’s lap, while the operator aims the instrument from a distance of 33 cm (± 1 cm) using built-in triangulating laser pointers converging on the bridge of the nose. During the exam, the child sees the flashing fixation target within the aperture of the instrument, (accompanied by synchronous sound), and the scanning circle

Fig. 8

The two methods used to distinguish between central- and paracentral fixation. The circles are measurement with central fixation, whereas the crosses stand for paracentral fixation. a Simple threshold for (P2.5 + P6.5)/P4.5 (Method 1). b Three-way discriminant analysis (Method 2 for 3D case). Shown is the discriminant plane. a shows the threshold for (P2.5 + P6.5)/P4.5, above which the eye is considered to fixate properly. b shows the discriminant plane according to Eqs. (4, 5 and 6) that best separates the same measurements in 3D parameter space

Fig. 9

The focus curve of the left eye (LE) of a 10-year old boy, to determine a “passing” threshold for focus. (C–A)/(C + A) is the focus goodness signal. The dashed line was obtained under cycloplegia. Shown is the 0.65 threshold used to discriminate between “passing” and “failing” focus