Dynamic tear meniscus parameters in complete blinking:insights into dry eye assessment

Ying-Huai Zhang, Jun Feng, Chen-Yuan Yi, Xian-Yu Deng, Yong-Jin Zhou, Lei Tian,Ying Jie

1Guangdong Key Laboratory of Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering,Shenzhen University Medical School, Shenzhen 518060,Guangdong Province, China

2Marshall Laboratory of Biomedical Engineering, Shenzhen 518055, Guangdong Province, China

3Beijing Ophthalmology and Visual Sciences Key Laboratory,Beijing Tongren Eye Center, Beijing Tongren Hospital, Beijing Institute of Ophthalmology, Capital Medical University,Beijing 100730, China

Abstract

KEYWORDS: dry eye; dynamic; tear meniscus; blinking

INTRODUCTION

Dry eye disease (DED) is a prevalent chronic ocular surface disorder characterized by a loss of tear film homeostasis and the presence of associated ocular surface symptoms.As one of the most common causes for visiting the ophthalmology centers, the prevalence of DED in adults is estimated to be between 5% and 50%[1].According to the Tear Film & Ocular Surface Society international DED consensus[2], DED can be categorized into two main types based on its underlying pathophysiology mechanisms:aqueous-deficient and evaporative dry eye.Aqueous-deficient DED arises from either insufficient or abnormal secretion of the aqueous component of tear, potentially resulting in an abnormal distribution of tear across the ocular surface[3].The tear meniscus, constituting approximately 75% to 90%[4]of the total tear volume, plays a pivotal role by supplying the aqueous layer to the tear film following each blink as a part of the ocular surface ecosystem[5-6].Situated along the margins of the upper and lower eyelids, the tear meniscus height(TMH) is of paramount importance as it reflects tear volume and contributes to the aqueous layer of precorneal tear film[7-9]Consequently, the measurement of TMH serves as an effective parameter in the diagnosis of DED, particularly in cases of aqueous-deficient DED.

TMH measurement methods can be divided into invasive and non-invasive approaches.The invasive approach employs fluorescein to stain tear meniscus[10].On the other hand, noninvasive assessments encompass various methodologies such as utilizing a slit lamp with or without reticule, employing kinds of image capture devices, implementing tear interferometry[11],and using optical coherence tomography (OCT)[12-14].However,these approaches are operator-dependent and tend to exhibit inadequate repeatability, leading to potentially erroneous outcomes.To overcome these limitations, we utilize deep learning techniques as they have excellent repeatability and are a component of artificial intelligence that has been employed in ophthalmic disease diagnosis research[15-17].Stegmannet al[18]introduced a deep learning segmentation method for the lower tear meniscus based on OCT images.In a similar vein, our research team proposed an automatic segmentation technique for the lower tear meniscus based on deep learning utilizing images captured by the Keratograph 5M Similarly[19].These approaches demonstrated notable accuracy and effectiveness.Nevertheless, it should be noted that the values of TMH in the lower tear meniscus undergo temporal change after a blink[20].Studies have indicated that TMH increases over time following each blink[21-22].Therefore, the primary aim of this study is to investigate the post-blink variation in TMH and elucidate potential patterns that may reveal sensitive parameters associated with the physiological structure or pathological condition of DED.

SUBJECTS AND METHODS

Ethical ApprovalThe study and data collection were conducted with the approval of the Ethics Committee of Beijing Tongren Hospital (TRECKY2021-238).Informed consent for participation was obtained from all patients and controls.

SubjectsThe data utilized in this study were collected from Beijing Tongren Hospital and consisted of 27 participants(mean age: 40±12.94y), comprising 3 male and 24 female individuals.A professional ophthalmologist used the Keratograph 5M (K5M; OCULUS Optikgeräte GmbH,Wetzlar, Germany; working at 8 frames/s) to record videos of the lower tear meniscus of both the left and right eyes of each participant, with a duration of approximately 10s per recording[23].During the recording, participants were instructed to execute a full blink, followed by maintaining their eyes open for as long as possible.Data instances that could not be accurately located or segmented due to image blurriness,participants blinking during the recording, or displacement of the placido disc were excluded.A total of 38 videos were collected from eyes that sustained an open state for more than 5s after a complete blink were collected for this study.

Additionally, the first tear film break-up time (NIBUT.f)and the average tear film break-up time (NIBUT.a) were documented.Based on the NIBUT.f values, DED was diagnosed for each eye of the participants.To investigate the correlation between dynamic parameters of tear meniscus and the presence of DED, the samples were stratified into two groups: DED group (NIBUT.f＜10s) and those non-DED group(NIBUT.f≥10s).Similarly, to explore the connection between the meniscus dynamic parameters and the severity of DED, the samples were categorized into three groups: eyes without DED(NIBUT.f≥10s), eyes with mild DED (5s≤NIBUT.f＜10s),and eyes with moderate DED (NIBUT.f＜5s)[24].Descriptive statistics for the participants are presented in Table 1.

Parameter CalculationThe captured videos were processed into individual frames.The last clear frame indicating closed eyes was designated as the initial frame (frame 0), while the 41stframe, corresponding to a five-second interval after the start frame, was selected as the terminal frame for calculating the dynamic parameters of the tear meniscus for each eye(Figure 1).Subsequently, segmentation was performed on the tear meniscus in each frame, yielding a calculation of seven parameters.These parameters including the overall TMH,TMH values specifically for the central, nasal, and temporal regions, the average upper boundary curvature across the entire tear meniscus, the upper boundary curvature of the central tear meniscus, and the area of the tear meniscus.Additionally,to facilitate a comparative analysis of the impact of different lengths of the tear meniscus within the central, nasal, and temporal regions, this study also calculates these parameters for tear meniscus lengths of 1, 2, 3, 4 mm respectively.

The tear meniscus segmentation model previously proposed by our research team was employed for the purpose of tear meniscus segmentation.Frames with open eyes were input to the network to obtain both the pupil center and the tear meniscus mask.Subsequently, the upper and lower boundaries of the tear meniscus (eupandelow) were subjected to a polynomial fitting process, guided by Formula 1.This fitting procedure was undertaken to achieve a smoother representation of the tear meniscus.The polynomial order, donated asN, was empirically set to 6, a choice made for enhanced performance.The coefficients of the polynomial, represented asw0,w1,w2,…,wN, played a pivotal role in this fitting process.Furthermore,the calculation of TMH across distinct regions was facilitated using Formula 2.Within this formula,Mindicated the pixelmillimeter magnification ratio (86 in this data set) whileDdenoted the length of the measurement section.

Figure 1 A schematic diagram of image frame selection used to calculate tear meniscus dynamic parameters.

Table 1 Descriptive statistical information of subjects in groups

In addition, Formula 3, wherePis the number of pixels of the tear meniscus mask, can be used to calculate the area of the tear meniscus.The curvature of the upper boundary of the tear meniscus in different regions is also calculated using a formula proposed by Zhanget al[25].

Using the y-axis of the pupil center coordinates as the center,the tear meniscus was sequentially divided into central regions of 1, 2, 3, and 4 mm in length, based on the previous work of our team and a priori knowledge.For left-eye image frames,the leftmost point of the tear meniscus was considered the beginning of the nasal tear meniscus, and the rightmost point was considered the starting point of the temporal tear meniscus.Conversely, for the right-eye images frames, the rightmost point of the tear meniscus was taken as the beginning of the temporal tear meniscus.Different lengths of the nasal and temporal tear meniscus were then selected.

Abnormal image frames that failed to segment the tear meniscus due to movements of the subject’s head, eye, or other blurs were assigned a value of 0.We fitted the effective parameters including TMH, curvature, and area of the tear meniscus within the 0-5s using the least squares method, and calculated the slope of the linear fit.Lastly, we performed statistical analysis of the slopes among different DED groups.Statistical AnalysisStatistical analysis for this study was conducted with SPSS 24.0 software (SPSS Inc., Chicago, IL,USA).Because of the small sample size in some groups and the lack of normality in most of the data (as determined by Shapiro-Wilk test), the non-parametric Mann-Whitney test was applied to analyze the differences between participants with and without DED and the slope of the line fitted from the dynamic parameters of the tear meniscus.The Fisher Chisquare test was used to examine the between-group difference analysis between the two groups of samples with or without DED and the two groups of samples with or without positive slopes of their fitted dynamic parameters of the tear meniscus.Welch’s ANOVA test was used to analyze the differences among the three participant groups, classified based on having no DED, mild DED, or moderate DED, as well as the slope of the line fitted from dynamic parameters of the tear meniscus.We conducted a Pearson Chi-square test to analyze the between-group difference analysis among three groups of samples, which include normal, mild DED, and moderate DED groups and three groups with or without positive slopes of their fitted dynamic parameters of the tear meniscus.Moreover, we performed a Spearman correlation analysis to assess the relationship between NIBUT and the linear slope ofthe dynamic parameters fitted to the tear meniscus.AP-value below 0.05 was considered statistically significant.

Table 2 The dynamic parameters of the tear meniscus were calculated for each frame of the sample over a span of 5s

RESULTS

Figure 2 visually confirms the correct segmentation of the tear meniscus by visualizing the boundaries of the tear meniscus,along with the average TMH in the left, right, and central regions of each frame.Figure 3 presents line graphs were depicting the 7 dynamic parameters of the tear meniscus calculated for each frame and the time variation.In addition,the trend lines obtained by linear fitting to the parameters are plotted.

Table 2 depicts the results of the descriptive statistical analysis of the trend straight-line slope obtained by fitting the seven dynamic parameters of the tear meniscus calculated for the 38 case samples.It can be inferred that the TMH of most subjects’eyes increased to varying degrees within 5s after blinking.This increase was observed both in the partial tear meniscus of different lengths and in the whole tear meniscus, particularly in the TMH of the central region and the overall TMH.In addition, among the tear meniscus of different lengths, the slope of the trend straight line obtained by fitting the TMH of the central region was always greater than or equal to that of the TMH of the nasal side.The TMH of the temporal side, on the slowest average rising trend.

Figure 2 The visualization result of the segmentation of the tear meniscus The green origin represents the pupil center.The green dashed line represents the y-axis, which is the central axis of the central tear meniscus.The leftmost green and red solid lines show the upper and lower boundaries of the segmented lefttear meniscus.The middle blue and orange solid lines display the upper and lower boundaries of the central tear meniscus.The right purple and pink lines indicate the upper and lower boundaries of the right tear meniscus.The TMH between the left, central and right tear meniscus are shown inside the black box located the upper leftcorner.TMH:Tear meniscus height.

Figure 3 Trend plots of the dynamic parameters of the tear meniscus over the period of 0 to 5s The light blue dashes represent the dynamic parameter of the tear meniscus at each moment and the dark blue-green line shows the trend lines that were obtained by linearly fitting the effective parameters.TMH served as an example in this figure.TMH: Tear meniscus height.

To investigate whether the trend in the dynamic parameter of tear meniscus obtained from the above calculation correlated with the DED, we performed an analysis of the differences between groups for each of the two groups of subject eyes classified as having or not DED according to NIBUT.f (Table 3),and for each of the three groups classified as without DED group, the DED group, the mild DED group and the moderate DED group (Table 4).The results showed that the slope of the straight line fitted to the upper border curvature of the whole tear meniscus was significantly different between the groups with and without DED (P=0.024).In the three groupsclassified according to the degree of DED, the slope of the line fitted to the central tear meniscus of length 4 mm also displayed a significant difference between the groups with and without DED (P=0.044).

Table 3 The results of the between-group difference analysis of the dynamic parameters

We classified the slope k of the fitted straight lines for the dynamic parameters calculated above into positive and negative groups and performed Chi-square tests among them.One group consisted of subjects with DED, whereas the other group included subjects without DED.Additionally,we also conducted Chi-square tests among three groups of subjects classified according to the prevalence degree of DED.The results of the tests are demonstrated in Tables 5 and 6.Based on the results, we concluded that were no significant differences observed in the positive and negative k values fitted to the dynamic parameters, between the two groups of subjects with and without DED.In contrast, in all three groups of data classified according to the degree of DED prevalence,significant differences were observed in the nasal side of the TMH with lengths of 2, 3, and 4 mm, respectively (P=0.024).

Table 5 The results of the between-group difference analysis of the slope

DISCUSSION

This study is founded on the automated TMH segmentation method proposed by our team.We calculated several key parameters by processing video data collected within five seconds following a full blink from 38 subjects.The parameters we computed consisted of the average TMH at different lengths, the curvature of the upper boundary of the tear meniscus, and the total area of the tear meniscus in each frame.The effective values of these dynamic parameters were then linearly fitted to explore the relationship between their changing trends and DED.

Among all 38 samples, our analysis indicates that 36 cases(94.74%) demonstrated an increasing trend in the average height of the central tear meniscus and 37 cases (97.37%)indicated a growing trend in the overall average height and area of the tear meniscus from the nasal to the temporal side over time.It is noteworthy that the TMH in most of the designated nasal and temporal areas also displayed an increase over time.This result further confirms the previous conclusion that the TMH tends to increases after a complete blink.

In addition, observed a correlation where increasing length of the tear meniscus being observed coincided with an increased number of samples demonstrating an upward trend in the TMH

Table 6 The results of the between-group difference analysis of the slope

on the nasal and temporal sides.Notably, during the dynamic changes of TMH, it appeared that the central average TMH grew at a faster pace or at the same rate as the nasal TMH,while the temporal TMH showed the slowest increasing trend.The statistical results suggest that changes in the curvature of the upper boundary curvature of the entire tear meniscus during the first 5s after a full blink may be related to the presence of DED.Moreover, the trends of the TMH of the nasal side in 2,3, and 4 mm during the same five-second period may assist in identifying the severity of DED in our collected data.

To summarize, our research indicates that there are dynamic changes in TMH following a blink.It is possible that the physiological structure of the eye and/or the presence of DED correlate with these changes.This observation is consistent with previous research that has shown that after a blink, TMH increases over time[10-11].

However, it is crucial to acknowledge that while this study substantiates the dynamic changes in TMH after blinking,it has certain limitations.Firstly, the videos collected in this study were captured with K5M equipment, which can only show the dynamic changes in the flat tear meniscus, making it difficult to accurately quantify the changes in tear volume.Additionally, discomfort from DED may cause some subjects to blink incompletely during testing, which can lead to blurry tear meniscus and sudden decreases in height, which may impact the study’s results.Lastly, the study only included 5 samples of the subjects’ eyes without DED, while the remaining 33 samples had varying degrees of DED.Therefore,future research should aim to obtain a more balanced sample distribution.

In conclusion, based on our study, we have confirmed that TMH undergoes dynamic changes after a complete blink.Moreover, we have identified several parameters that are sensitive to the physiological structure or pathological state of DED.

Our findings demonstrate that the most of our sample (94.74%)experienced an upward trend in the average height of the central tear meniscus as well as the overall TMH and area in the nasal-temporal direction after a blink.We also observed a higher count of samples with a trend of increasing average TMH in both the nasal and temporal areas as the length of tear meniscus increase.This study provides new insights into the dynamic changes in TMH after a blink.Identifying of diagnostic parameters based on these changes could potentially have significant clinical implications for the diagnosis and management of DED.

ACKNOWLEDGEMENTS

Foundation:Supported by Medical-Engineering Interdisciplinary Research Foundation of Shenzhen University and Research Development Fund of Beijing Municipal Health Commission(No.2019-4).

Conflicts of Interest: Zhang YH,None;Feng J,None;Yi CY,None;Deng XY,None;Zhou YJ,None;Tian L,None;Jie Y,None.