ABSTRACT
Objectives:
This paper presents the design and construction of a viable pupillometer system and demonstrates its merits with extensive validation tests.
Materials and Methods:
A web camera was modified by removing its infrared filter and mounted on a chin rest. Light emitting diodes (LEDs) operating at infrared and visible spectra were integrated to provide background and light stimulus, respectively. The LEDs were controlled by a microprocessor board. Stimulation was presented using a periodic paradigm with variable period and duty cycle. Videos of both pupils were recorded at 30 frames/second and processed offline using software developed in-house. The overall system was validated with data gathered from individuals with healthy vision under different stimulation paradigms. Temporal variations in pupil size were determined and analyzed statistically.
Results:
The analysis revealed that the pupil sizes were accurately measured from the video frames provided that reflections from both infrared and visible lights remain outside the pupil. The system achieved moderate to excellent repeatability scores (87.8 and 86.8% for short 1 second and long 2 second pulses, respectively), which demonstrated its effectiveness and confirmed that it can be used reliably as a pupillometer.
Conclusion:
The proposed pupillometer system produces useful, quantitative data characterizing pupillary light response. However, further development and implementation are needed to potentially turn it into a low-cost alternative for other studies involving the autonomic nervous system, cognitive function, drug metabolism, pain response, psychology, fatigue, and sleep disorders.
Introduction
The pupil is an important functional structure that balances the amount of illumination entering the eye to enable clear vision. The sphincter pupilla and dilator pupilla muscles act together under the control of the parasympathetic and sympathetic nervous systems. The muscles are respectively governed by the oculomotor nerve, sympathetic nervous system tracts and fibers that are located in the mesencephalon and the cervical spinal cord. The functional status of these structures (optic nerve, mesencephalon, spinal cord, oculomotor nerve, cervical sympathetic fibers, and pupillary muscles) is evaluated by the pupillary light response (PLR).1 Although physical examination of PLR is the usual method of evaluation, use of a device called a pupillometer offers more diagnostic sensitivity than physical examination alone.2 Commercial pupillometers are both costly and unavailable in clinical practice, especially in developing countries. Also, they lack flexibility and versatility for research purposes where light stimulation under different paradigms may be required. This paper addresses this issue. We have developed a viable pupillometer system for real-time video recording of pupil response to light stimulation and video analysis software for characterizing PLR. The overall system was validated and its merits were investigated with the data acquired from healthy individuals subjected to periodic white light stimulations with short and long pulse durations.
Materials and Methods
Results
Thirty-seven volunteers, 16 females (age: mean 34; min 20 and max 61 years old) and 21 males (age: mean 36; min 20 and max 60 years old), were included in this study. The test runs with the stimulation paradigms producing light intensities listed in Table 1. The intensity values represent the average over 12 periods of dark and light durations. The light levels achieved with the paradigms were sufficiently high to induce strong pupil responses. Figure 3 shows the representative graphs of NPAs as obtained from an examination session with the procedures described above. In both eyes, NPAs exhibited identical behaviors, indicating the capability of the system to promptly follow changes in pupil size in response to the light stimulus.
Intra-class correlation coefficient calculations from the first and second trials under both paradigms are summarized in Table 2. Most cases had moderate to excellent repeatability scores (87.8% for the 1 second and 86.8% for the 2 second stimulation paradigms). These findings confirmed the quality of the match between the signal pairs obtained with repetitions.6,7
Figure 4 shows representative mean traces (average over 12 cycles) of the pupil response in Figure 3. The corresponding plots for each eye also exhibit very close traces.
Discussion
In spite of commercial availability, there are still attempts to build custom-made pupillometers to address specific concerns.7 Parameters such as initial pupil size and duration/velocity/latency of pupillary contraction and dilatation are of clinical interest as they reflect the functional state of the eye. Classically, the V-shaped response was observed with the light stimulus. When the light was on, the pupil first contracts and then dilates after the light turns off (Figures 3, 4). The temporal appearances of PLR traces for the 1 second and 2 second stimulus paradigms were similar to those produced by the commercial pupillometers.7 However, while commercial devices typically produce data from a single stimulus with a fixed period and duty cycle, our system is capable of handling periodic stimulations with different periods and duty cycles. The system is also flexible in the sense that white diodes can be replaced by those with different colors of interest to facilitate PLR studies concerning color dependence. Moreover, the continuous stimulation does not lead to habituation.
Eye movements (saccades) affect video-based estimates of pupil size, especially when a computer screen is used for stimulation purposes.8 In our setup, the patient was asked to focus on the camera’s shutter during the examination. The software monitored the stability of the eye by tracking the center of mass of the pupil segment in each frame. Positional shifts of more than 10 pixels in length were considered as indicative of eye movement. These efforts ensured increased accuracy in the pupil capturing area. However, in a very small number of cases, data gathering and analysis were limited due to low eyelid position, frequent blinking, insufficient data capturing of pupil area because of interference with shadow due to light-dark cycle, and also interference related to make-up in female subjects. This is why the total numbers in the frequency column of Table 2 do not add up to 74 (37 participants with 2 eyes), meaning that it was not feasible to estimate the pupil size even after filling the missing data points with zero. Nevertheless, in the absence of such issues, the custom-made pupillometer and video analysis software platform developed in this study revealed moderate to excellent repeatability scores (greater than 85% of the people investigated) and its performance was comparable to the test-retest repeatability of the previous pupillometer scores.2,7 Lei et al.7 evaluated the test-retest reliability of hemifield, central-field, and full-field chromatic pupillometry. For the post-illumination pupil response, they determined intra-class correlation coefficients of 0.84 (0.69-0.95) and 0.94 (0.83-0.98) at full field stimulation with blue light. Unlike our study, which did not assess interobserver variability, Couret et al.2 investigated the interobserver variability and reported intra-class correlation coefficients of 0.95 and 0.87 for pupil size at both resting and after light stimulation, respectively. Our intra-class correlation coefficient values were similar, confirming that the pupillometer system can be used reliably to evaluate PLR.6
The potential value of PLR evaluations in normal and disease conditions have been investigated in many studies in areas such as the autonomic nervous system, cognitive function, drug metabolism, pain response, psychology, fatigue, and sleep disorders.9,10,11,12,13,14,15,16,17,18,19,20 There is a growing interest in diagnosis in these areas, and our custom-made pupillometer may be useful as an easily-applicable, non-invasive diagnostic tool.
Study Limitations
As mentioned earlier, in a few cases there were limitations in data gathering and analysis due to low eyelid position, frequent blinking, insufficient data capturing due to shadowing, and interference related to cosmetics worn by female subjects. Extracting data from the videos was also time-consuming.
Çalışmanın Kısıtlılıkları
Daha önce de belirtildiği gibi, az sayıda olguda, göz kapağı pozisyonunun düşük olması, sık göz kırpma, gölgelenmeye bağlı yeterli veri yakalanamaması ve kadın olgularda kullanılan kozmetiklerden kaynaklanan enterferans yeterli veri toplanamadı ve analizde dahil edilemedi. Ayrıca videolardan verilerin elde edilmesi zaman alıcıydı.
Conclusion
Our study demonstrates that our custom-made pupillometer and video analysis software platform can be used to reliably evaluate PLR. However, further development and implementation are needed to potentially turn it into a low-cost alternative.
