Development of a Method to Evaluate the Dynamic Fit of Face Masks

Abstract

Evaluating designed objects in real-world use cases enables usability optimization. For functional objects such as face masks, the mask must fit the user initially and continue to fit during movements such as talking. This paper describes methodology development for dynamic fit analysis of face masks using 3D head scans. Participants were scanned while wearing Basic, Cup, and Petal model masks before and after reading a passage aloud and completed surveys across eight fit dimensions. Face and mask measurements were virtually extracted from the head scans for quantitative fit analysis, and mask overlays were inspected for qualitative fit analysis. Four of eleven facial measurements changed significantly from closed to open-mouth posture while the nasal dorsum was identified as a stable landmark and served as a reference to define a mask shift metric. The mask shift was compared to the survey results for the model masks, with the Cup design fitting best and the Petal design rated as most comfortable. Poor fit modes identified from mask overlays were fabric buckling, compressed nose and ears, and gapping between the mask and facial features. This methodology can be implemented during the analysis stage of the iterative design process and complements static fit analyses.

1. Introduction

Cloth face coverings and surgical masks have received more attention since the onset of the COVID-19 pandemic due to respirator supply shortages and the call for community protective measures to slow the spread of the SARS-CoV-2 virus [1]. The protection afforded by individual and community masking extends beyond this virus to other respiratory viruses, the common cold, annual flu, and nonpathogenic inhalable and respirable particles such as pollen and dust [2,3,4]. However, face masks that do not fit well provide poor performance and can cause physiological discomfort [5,6]. Several factors affect the face mask fit and performance variables such as dead space, donning ease, assigned protection factor, leak shape complexity, seal pressure, and material choice [7,8].

Extensive research has been conducted on and many standards have been developed for respirator fit. Reviews of these studies highlighted disparities between the population that respirator fit test panels are designed for and the population that uses them. Notably age, weight, ethnicity, and gender groups outside of the young, fit, white, male category had low success rates of fit testing [9,10,11,12,13]. Further, face dimensions and shape change with movements such as talking affect the mask fit [14]. However, research into systematically measuring and analyzing the interaction between face and mask during movement to assess dynamic mask fit is limited. Mask fit tests typically involve the user performing various activities such as heavy breathing, moving head side to side, and talking. For quantitative fit tests, the inward leakage of test aerosol is measured via a PortaCount particle counter [15]. For qualitative fit tests, a scented aerosol is released and the user signals whether they detect the scent [16]. While these fit tests are considered dynamic and determine the extent of poor fit, they do not capture how the mask does not fit. That is, no root cause analysis is performed to uncover why the mask is not fitting properly, the test is a simple pass/fail. A poor fit cannot necessarily be fixed simply by going up or down a size in the fit test panel. Face masks (cloth face coverings and surgical masks) are not classified as respirators and do not pass the respirator fit tests as they do not necessarily create a seal on the user’s face. Yet, comparative fit studies across mask categories (N95, KN95, surgical, and cloth) are more prevalent due to the COVID-19 pandemic and underscore the importance of harmonized face sizing systems and mask designs to achieve proper fit and protection [17].

Human faces, much like bodies, are complex sizes and shapes. Thus, they require sophisticated measurement techniques to accurately capture their geometry. To this end, rapid improvement in the quality and accessibility of 3D scanning over the last two decades for the digital representation of human features has improved the understanding of head anthropometry and head variability. Three-dimensional scanning has previously been incorporated into the product design process [18,19]. Applied examples are an electroencephalogram (EEG) headset [20], a Personalized Fitting Device (PFD) for oronasal masks [21], elastomeric half-mask facepieces [22,23,24], face masks [25,26,27,28,29], and oxygen mask for pilots [30]. Beyond the initial design, 3D head data can also be used for the virtual fit assessment [31] and estimation of facial contact pressure [32]. Although these examples are based on 3D head scan data of faces with closed mouths (i.e., static fit), this same ergonomic product design process can be applied to the scans of facial movements (i.e., dynamic fit) to design better-fitting face masks [14,29,33,34,35]. A key benefit to 3D scanning is that the obtained scans are snapshots of the interaction between mask and face at a single point in time.

Based on the initial findings of our pilot study [36], the present study details the methodology development for dynamic fit evaluation using a 3D scanner. In AATCC M14, face mask fit was defined as the “ability to cover the user’s nose and mouth while making consistent and snug contact with the user’s face at the perimeter” (page 522, [37]). Therefore, in our study, we defined dynamic fit as maintaining the coverage of the nose and mouth and contact with the face at the mask perimeter. The disruption of the coverage or contact was qualified as poor fit. Our study includes a description of the workflow, Figure 1, and a demonstration of the methods on three models of reusable, cloth face masks. The Basic mask represented a simple, shaped mask design, the Cup mask represented a respirator-style mask design, and the Petal mask was a hybrid shape-and-stretch novel mask design. Face measurements were extracted from participant scans in their closed- and open-mouth postures, and the change between the postures was analyzed to identify which dimensions changed significantly. The dynamic fit was evaluated based on the displacement of the mask along the nose bridge after a talking exercise (defined as mask shift) and participants’ fit rating scores. Differences between the three masks were identified from the quantitative and qualitative results, exemplifying the complementary techniques. The methodology described in the following sections explores how the coverage and contact were disrupted. Our methodology can be applied during the analysis phase of the iterative design process of face masks and other headgear to optimize for fit and comfort during actual use.

2. Materials and Methods

2.1. Face Mask Designs

Three cloth face mask models, Figure 2, were designed for this study: the Basic mask was based on a common commercially available cloth face mask [38], the Cup mask was the cloth adaptation of the N95 “cup style” respirator, and the Petal mask was a novel design inspired by the mobility of flower petals. Each mask was constructed with the same fabric, nose wire, and ear loops. The inner and outer layers are 118 g per square meter (gsm) cotton sateen weave fabric, and the middle layer is 81.4 gsm polypropylene spunbond nonwoven fabric. The Petal mask additionally required stretch panels consisting of 199 gsm nylon/LYCRA spandex tricot fabric.

The experimental masks were designed as one-size-fits-all (OSFA) based on the commercial facemasks’ dimensions. Our novel face mask design also reflected these measurements with additional seams to better follow the contours of the face. Five sets of each mask design were constructed. Masks were measured as sewn, laundered, and air-dried between participants, and measured again after each wash cycle. The 95% confidence intervals and interquartile ranges were calculated for each mask measurement based on the corresponding 30 mask samples. These two methods for determining outliers are discussed in more detail in the Supporting Information.

2.2. Three-Dimensional Head Scanning

After receiving a Cornell University Institutional Review Board approval (IRB0010618 (2110010618)), 30 participants were recruited for this study. Participants were scanned in a sitting position using a Vitus 3D head scanner (Humanetics, Farmington Hills, MI, USA). For each participant, scans were taken in the two postures (viz., closed mouth and open mouth) without a mask. The open-mouth posture was instructed as the pronunciation of the “ahh” sound. For each sample mask, the participant donned the mask and adjusted it to a comfortable position at which they would normally wear the mask and nose wire, coded here as neutral position. Scans were taken in the closed-mouth and open-mouth postures, then again in the closed-mouth posture after the participant read the “Rainbow Passage” from the 2004 OSHA quantitative fit test (see Supplementary Information) out loud at a normal volume and pace to simulate regular talking while wearing the mask [39]. All scans were exported as object (.obj) files. Each participant completed a questionnaire (see Supplementary Information) with demographic and mask preference questions before scanning, evaluation criteria questions during scanning, and mask ranking questions after scanning. Three mask designs were prepared, and three replicates per mask type were tested to capture the variation in neutral mask placement and fit after talking. Order was randomized for each participant to minimize repetitive use bias [40].

2.3. Extraction of Face and Mask Measurements

Each scan was imported into Geomagic Wrap 2021 (Oqton, San Francisco, CA, USA) to prepare the scans for measurement. Scans were not made watertight to ensure no facial features were obscured [27,28]. For “no mask” scans only, the Frankfurt plane (Figure S1) was added using the 3-point plane feature by locating the right tragus, the left tragus, and the left inferior orbital rim. The Frankfurt angle was taken in reference to the participant’s position, i.e., horizontal plane parallel to the ground. Each scan was trimmed at the neck and exported as an object (.obj) file.

For qualitative fit analysis, trimmed scans were overlaid in Geomagic Wrap with the mask scan transparency set to 70%, while the “no mask” scan remained opaque. Masks in the closed- and open-mouth postures were overlaid on the corresponding “no mask” posture.

2.3.1. Face Functional Measurements

Each “no mask” object (.obj) file was imported into CLO3D (Version 7.1, CLO Virtual Fashion LLC., New York, NY, USA) as an avatar. To improve the reliability of the landmark placements and face measurements and reduce memory bias, each scan was landmarked (Figure 3) and measured three times with a minimum of 12-h intervals between measurement replicates.

Virtual caliper measurements were made using the linear measure tool. Curve length measurements were made using the virtual tape measure tool. The observer errors, calculated as both standard error and mean absolute deviation (MAD) for all linear and curved measures, were below the allowable error of 3 mm and 5 mm, respectively [41]. Additionally, landmarks were placed on one participant and manually measured using calipers and tape measure for linear and curve lengths, respectively. Physical face measurements were taken three times and compared to the virtual measurements to establish consistency and reliability for landmarking and measuring. As a result, 11 measurements (

Table 1. Linear and curve facial measurements from the “no mask” scans based on the landmarks placed.

Linear (Distance Between Landmarks, Measured with a Virtual Caliper)	Curve (Arc Lengths, Measured with a Virtual Tape Measure)
Lip length	Bitragion subnasale arc, right
Menton sellion length	Bitragion subnasale arc, left
Nasal root breadth	Bitragion chin arc, right
Nose breadth	Bitragion chin arc, left
Nose protrusion
Subnasale sellion length
Pronasale menton length

) were extracted and recorded in a Microsoft Excel file for data analysis.

2.3.2. Mask Shift Measurement

Mask scans in closed-mouth posture before and after talking were overlaid and aligned in Geomagic Wrap. The boundary between mask and face at the nasal dorsum was marked manually on both scans. A reference on the nasal dorsum was marked in the same location on both scans. The displacement in the x-and y-coordinates between the mask boundary and reference marker was calculated for both scans. Mask shift was calculated in Equation (1) as the difference in these displacements between the before and after talking scans. Before talking was taken as time point t₁, and after talking was taken as time point t₂, where ref refers to the reference marker and mask refers to mask boundary.

$$maskshift=\left({\left[x_{ref},y_{ref}\right]}_{t_2}-{\left[x_{mask},y_{mask}\right]}_{t_2}\right)-\left({\left[x_{ref},y_{ref}\right]}_{t_1}-{\left[x_{mask},y_{mask}\right]}_{t_1}\right)$$

(1)

2.4. Data Analysis

All data are reported as mean (M) and standard deviation (SD) unless otherwise noted. Statistical analyses including descriptive statistics, analysis of variance (ANOVA), and post hoc tests were performed with Statistics and Machine Learning Toolbox in MATLAB (The MathWorks, Inc., Natick, MA, USA, https://www.mathworks.com/products/matlab.html (accessed on 10 December 2024)) to comprehensively examine the data and draw meaningful conclusions.

3. Results

3.1. Demographics

Participants (N = 30) aged 19 to 40 years and weighing 105 to 250 pounds with a mean age of 26 years and a mean weight of 153 pounds were recruited. Twelve male, sixteen female, and two nonbinary participants self-identified their race and/or ethnicity, grouped below in

Table 2. Participant demographics grouped by race and/or ethnicity based on the U.S. Census Bureau definitions.

Category	Count
White	13
Asian	10
Black or African American	3
American Indian or Alaska Native	2
Hispanic or Latino	1
Prefer not to answer	1

.

Previous studies have disputed whether face dimensions are correlated to age, gender, race, and ethnicity [42,43,44]. In addition to the static measures, the jaw range of motion has been shown to be dependent on age and gender [45]. Designing for good fit based on demographic factors is therefore challenging. No comparison of demographic variables was made here due to the small sample size. The effect of change in facial dimensions with movement and mask fit are explored further in subsequent sections.

3.2. Participant Anthropometrics

3.2.1. Frankfurt Plane

In standing postures, the Frankfurt plane of each participant is assumed to be parallel to the horizontal floor plane, i.e., the angle between these two planes—defined as the Frankfurt angle—is 0° [45]. The 3D head scanner used in this study required the participants to sit in a position that could affect the tilt of the Frankfurt plane. To measure whether sitting influenced the Frankfurt angle, the Frankfurt angle was measured for both the closed- and open-mouth postures. In the closed-mouth posture, participants’ Frankfurt angles were 4° ± 6°, meaning that the participants’ chins were generally tilted above the parallel plane. In the open-mouth posture, the participants’ Frankfurt angles were 7° ± 6°. A one-sample t-test with two-tailed hypothesis and significance level of p < 0.05 was conducted for the Frankfurt angles from the closed- and open-mouth postures. Both the closed- and open-mouth postures while sitting were significantly different from the assumed population mean of 0° for a standing posture. Further, a paired t-test was conducted for the closed versus open-mouth posture Frankfurt angles. The participants’ Frankfurt angles were significantly greater in the open-mouth posture compared to the closed-mouth posture, suggesting that participants tended to tilt their head back when opening their mouth.

While the Frankfurt angles for these participants were statistically significantly different from a standing posture, it is not expected that these small but significant differences would affect the viability of using a seated posture for the dynamic fit methodology. These angles were comparable to those in the study of Gordon and colleagues, where participants were specifically asked to lift their chins slightly above the parallel Frankfurt plane to ensure that the scanner completely captured the region beneath the chin [41].

3.2.2. Closed-Mouth Versus Open-Mouth Dimensions

The current and historic fit test panels for respirators are based on static measurements, i.e., closed-mouth posture. For face masks to accommodate jaw movement during talking and other minor movements, dynamic measurements must also be considered. A pairwise t-test was performed on the differences between the closed-mouth and open-mouth measures for each measurement type in

Table 3. Differences in face measurements in the closed- and open-mouth postures. p-values are for pairwise t-test where the null hypothesis is that the difference in the closed-mouth and open-mouth means is zero.

Measurements	Closed-Mouth		Open-Mouth
	M (mm)	SD (mm)	M (mm)	SD (mm)	p-Value
Linear (Caliper)
Lip length	45.72	4.39	47.03	5.27	0.19
Menton sellion length	116.35	7.28	133.33	11.46	0.00 **
Nasal root breadth	15.87	2.00	15.90	2.08	0.78
Nose breadth	34.56	3.96	34.77	4.25	0.39
Nose protrusion	22.55	2.94	22.84	3.06	0.06
Subnasale sellion length	54.76	4.45	55.29	4.41	0.11
Pronasale menton length	83.99	6.71	102.01	10.71	0.00 **
Curve (Surface Tape)
Bitragion subnasale arc, right	145.69	9.11	145.28	9.12	0.47
Bitragion subnasale arc, left	144.59	8.96	143.30	8.88	0.04 *
Bitragion chin arc, right	160.21	11.42	162.19	10.94	0.02 *
Bitragion chin arc, left	158.21	11.11	161.30	11.13	0.00 **

Note. * p < 0.05, ** p < 0.01.

.

Of the insignificant differences in closed- and open-mouth measures, the nasal root breadth, nose breadth, nose protrusion, and subnasale sellion length were all expected to be unaffected directly by the mouth opening, and these results confirm that they were also not affected indirectly by the jaw movement. Importantly, subnasale sellion length did not significantly change, so any shift in masks along the nasal dorsum (nose bridge) is attributed to the mask sliding up or down the dorsum. While lip length was expected to be correlated with mouth configuration, the insignificant difference between the closed mouth and “ahh” sound could be attributed to the differences in how much each participant naturally opened their mouth when speaking. It is plausible that those participants that opened their mouth and jaw wider would deform their mouth so that the height, indirectly measured by menton sellion length and pronasale menton length, increased significantly, and as a result, width, or lip length, was invariant. Additionally, the total bitraigon chin arc increased while the right bitraigon subnasale arc did not change, and the left bitraigon subnasale arc decreased, implying potential asymmetry in some or all the faces. It can also be expected that the extension at the hinge point of the jaw at the tragion accounted for the bitraigon chin arc increase. Thus, most of the deformation during talking was in the vertical direction, which constrained a well-fitting mask to accommodate this vertical change or else the mask would feel restrictive and cause the mask to slide down the face, as observed in the pilot study.

These results are in agreement with Morishima’s open-mouth study [34]. The difference in closed-mouth and open-mouth postures were similarly shown to have small changes in horizontal dimensions and larger changes in vertical dimensions, depending on the sound being made by the participant. “Ma” was found to have the largest open-mouth area of the sounds tested. Negligible change occurred in distance between the fiducial markers at the nasal dorsum, confirming that any movement in the mask position is a function of the mask, not the face.

3.3. Quantitative Fit Analysis

3.3.1. Mask Shift

Based on the pilot study, the main mode of poor fit observed was the mask sliding down the nasal dorsum. This is a desirable metric for evaluating dynamic fit as it meets the following criteria: a single metric with well-defined limits (length of the nose) on a mostly rigid, straight reference. A face is a complex shape with highly deformable areas and few rigid areas. The nose is one of the few features on the face that is mostly nondeformable, and most noses can be treated as approximately linear. Most of the masks slid down the nose, but 29 out of the 270 mask trials (10.7%) slid up the nose, of which 4 were Basic masks, 21 were Cup masks, and 4 were Petal masks. All masks that slid up during talking had shifts of <5 mm that would have been difficult to capture with lower-sensitivity manual measurements where shifts <2 mm could be lost as “negligible”.

Mask shift was converted to a dimensionless number across participants by dividing the mask shift length by the subnasale sellion length. Aggregated mask shift distance (Figure 4a) and dimensionless mask shift followed the same trend with significant differences between each mask type, so the mask shift was not necessarily proportional to the participant’s nose length.

Looking at mask shift for each participant in Figure 4b, some participants (e.g., participants 1 and 27) had relatively low shift (good dynamic fit) for all three masks, some (e.g., participant 22) had relatively high shift (poor dynamic fit) for all three masks, and some (e.g., participants 9 and 15) had a mixture of low, moderate, or high shifts for masks. Data in Figure 4b demonstrate how one mask that fits one face size and shape well does not necessarily fit other face size and shape variations.

3.3.2. Correlation Between Face Measurements and Mask Shift

To identify which facial dimensions were related to mask shift, the correlation coefficient for each face measurement (from closed- and open-mouth postures) was calculated across mask types. The arithmetic mean was calculated for each face measurement and mask shift across each participant and mask type. As expected, no single face measurement was highly correlated with mask shift. That is, goodness of fit cannot be easily predicted with a single face measurement. However, correlation coefficients within the same face measurement showed differences between mask types. For example, the Basic mask shift was moderately positively correlated with nasal root breadth in both closed- and open-mouth postures, shown in

Table 4. Correlation coefficients between nasal root breadth from the closed- and open-mouth postures and the mask shift lengths for Basic, Cup, and Petal masks.

Nasal Root Breadth	Correlation Coefficient
	Basic	Cup	Petal
Closed mouth	0.31	−0.42	−0.04
Open mouth	0.34	−0.36	−0.06

. Nasal root breadth change between closed- and open-mouth postures was not significant, so the closed- and open-mouth correlation coefficients were expected to be similar, i.e., the larger the nasal root breadth, the larger the mask shift. Therefore, the positive correlation with nasal root breadth means that the Basic mask did not suit users with wider noses. Conversely, the Cup mask exhibited a correlation coefficient of similar magnitude but opposite sign to the Basic mask. Thus, the Cup mask tended to suit users with wider noses better. The Petal mask shift did not correlate with nasal root breadth as some users with narrower noses experienced mask shifts similar to those with wider noses. The differences in correlation for different masks signify that masks can be designed to accommodate different facial features, and a proper fit test panel could be defined by these facial shapes and features.

3.4. Qualitative Fit Analysis

3.4.1. Mask Overlays

Fit factors such as ease amounts, dynamic adaptiveness or restrictiveness, dead space, and face seal were examined by overlaying mask scans on face scans for closed- and open-mouth postures (Figure 5). The overlays show how the mask interacted with the face underneath the mask in both the static and dynamic postures. An advantage of in vivo masks on participants compared to the virtual fit of digital masks is that human participants have compressible skin that will deform depending on the pressure from the mask design, materials, and fit. Compression can indicate potential discomfort, such as at the nose tip and ears. This deformation was observed in the overlays where the “no mask” scan is visible through the “mask” scan. For example, in Figure 5b, the tip of the nose was protruding through the mask, and the ears were bent forward from the ear loops pulling on them.

Other common observations were the sides of the mask being flush with the skin as in Figure 5a or showing gapping at the sides and fabric buckling as in Figure 5b, especially in the open-mouth postures, and gapping under the chin appearing or disappearing in the closed versus open-mouth postures as in Figure 5c. These differences in mask configurations with the closed- and open-mouth postures indicate which mask features (shape and size) contribute to good dynamic fit and which impede the dynamic fit.

3.4.2. Participant Questionnaire

Participants’ perceptions of fit and comfort were rated across eight dimensions on a 5-point Likert-type scale, from 1 (Very Difficult/Dissatisfied/Uncomfortable/Restrictive) to 5 (Very Easy/Satisfied/Comfortable/Adaptive), with 3 being Neutral. These ratings were averaged (arithmetic mean) for each mask type within each participant (color map in Figure S2) and aggregated over all participants (Figure 6). Unless otherwise specified, significance is established at p < 0.05.

All masks were constructed with the same ear loops, nose wire, and fabric; however, functionality, fit, and comfort were rated differently across masks, indicating that the mask design had a significant effect on these dimensions of mask performance, as shown in the significance tests results in

Table 5. ANOVA and Tukey’s post hoc analysis of performance ratings for Basic, Cup, and Petal masks. Ratings with significant ANOVA results (p < 0.05) were compared pairwise with Tukey’s post hoc analysis. Pairs were found to have significantly different means at p < 0.05.

Rating	ANOVA p-Value	Tukey’s Post Hoc
		Basic to Cup	Basic to Petal	Cup to Petal
Functionality—ease of donning	<0.01 **	0.02 *	0.93	0.01 *
Functionality—ease of doffing	0.05	0.23	0.69	0.04 *
Functionality (breathing)	0.06	-	-	-
Functionality (talking)	0.26	-	-	-
Initial fit	<0.01 **	0.01 *	0.65	<0.01 **
Fit after reading passage	<0.01 **	<0.01 **	0.47	0.08
Comfort	<0.01 **	0.02 *	0.58	<0.01 **
Mobility	<0.01 **	0.88	<0.01 **	<0.01 **

Note. * p < 0.05, ** p < 0.01.

. The Cup mask was significantly more difficult to don compared to both Basic and Petal masks, most likely attributed to having to find the correct position for the mask to align with the contours of the face. While a significant difference was found for the Cup mask compared to the Petal mask, all three masks received mean doffing ratings of “Easy to doff”. Because all masks were constructed from the same fabric, any difference in the breathing and talking ratings was a function of the mask design. However, no difference was found between the three masks for breathing and talking.

Although the Petal mask was rated as having the highest mobility, it shifted more than the Cup mask. Shifting in the Basic mask was perceived as adaptive, while the lack of shifting in the Cup mask was perceived as restrictive. Thus, masks must not only be adaptive to the movement of the jaw during talking to stay in place, but they also must be comfortable not to feel restrictive. While both the Basic and Petal masks’ fit ratings decreased after reading the passage, the Cup mask’s fit ratings increased. The Basic mask was designed to be looser fitting than the Cup mask, yet both masks received an average rating of “neutral” mobility compared to the significantly higher rating of “adaptive” mobility for the Petal mask. As expected, the Cup mask was the least comfortable, most likely attributed to its tight fit around the perimeter of the mask. The tradeoff between fit and comfort is underscored here, and this method can be used to iteratively optimize the mask size and shape to achieve a better balance between the dimensions of fit and comfort.

The Petal mask performed the highest on all metrics except for “fit after reading passage”. While the “fit after reading passage” trend between masks (Cup > Petal > Basic) matched the mask shift measurement trends, masks generally were rated as either “minor shift of mask” or “negligible/no shift of mask”, whereas the quantitative shift shows that the Petal and Basic masks exhibited between minor and major shifts of mask depending on the participant. Only the Cup mask had a minor shift on average. This alignment of trends between quantitative (objective) and qualitative (subjective) measures affirms that this method of using 3D scans to evaluate the dynamic shift is effective. However, it showed that the users underestimated how much the mask shifts on their faces.

Upon comparing the participants’ perceived dynamic fit ratings with mask shift lengths for each individual mask (Figure 7), we found that the perceived and actual amounts were weakly correlated. The overall correlation coefficient between mask shift and dynamic fit rating was −0.64, with individual mask correlation coefficients of −0.66, −0.49, and −0.57 for Basic, Cup, and Petal masks, respectively. The weak correlation of mask shift and fit after reading passage rating may be associated with the limited ability of the linear rating scale to capture the nonlinear nature of perceived mask shift. For example, the Petal mask had average ratings of 3 (“Minor mask shift”), corresponding to mask shifts ranging from 5.0 mm to 20.4 mm. In the case of the Cup mask, the participants’ mean mask shifts were below 10.3 mm compared to 39.5 mm shift for the Basic mask and 20.4 mm shift for the Petal mask. Participants were generally good at identifying minor or extreme (e.g., fully sliding off nose) changes in their mask fit, but negligible, minor, and major changes were more difficult to differentiate. Therefore, the user’s perception of fit should not be the standalone factor when evaluating fit.

The dynamic fit rating, however, captured more than just the shift of the mask down the nose. As highlighted in the mask overlays, the masks displayed other types of poor fits outside of shifting down the nose. This gap between the perceived and actual fit might be attributed to other aspects of the mask shifting or perception of shifting, such as the seam line locations on the masks. For example, the Basic mask might be the most “linear” due to its vertical center seam and have the mask shift at the nose the major factor, whereas the Cup mask might change in multiple dimensions (e.g., buckling of the center seam). Despite the loss in total information from reducing the dynamic fit to a single metric, mask shift can be a simple but satisfactory estimate of dynamic fit.

3.4.3. Agreement Between Overall and Individual Ratings

Overall ratings (3-point scale: 1 = best, 3 = worst) at the end of the questionnaire were averaged over all participants for each mask type for each of the eight performance categories. The mean overall rankings (Figure 8) were compared to the mean individual ratings (Figure 6). The Petal mask was consistently the best performer across dimensions except for the “fit after reading passage”. All three masks in this category received mean overall rankings that were indistinguishable, shown in

Table 6. ANOVA and Tukey’s post hoc analysis of overall performance rankings for Basic, Cup, and Petal masks. Ratings with significant ANOVA results (p < 0.05) were compared pairwise with Tukey’s post hoc analysis. Pairs were found to have significantly different means at p < 0.05.

Rating	ANOVA p-Value	Tukey’s Post Hoc
		Basic to Cup	Basic to Petal	Cup to Petal
Functionality—ease of donning	<0.01 **	0.01 *	0.25	<0.01 **
Functionality—ease of doffing	<0.01 **	0.01 *	0.25	<0.01 **
Functionality (breathing)	<0.01 **	0.51	<0.01 **	<0.01 **
Functionality (talking)	<0.01 **	0.59	0.01 *	0.13
Initial fit	<0.01 **	<0.01 **	0.86	<0.01 **
Fit after reading passage	0.17	-	-	-
Comfort	<0.01 **	0.02 *	0.02 *	<0.01 **
Mobility	<0.01 **	0.86	<0.01 **	<0.01 **

Note. * p < 0.05, ** p < 0.01.

, whereas the Cup mask was rated significantly higher than the Basic mask for the mean individual ratings and the Petal mask was indistinguishable from the Cup and Basic mask ratings. This finding corroborates the claim that it is difficult for a user to perceive the difference in dynamic fit and speaks to the complexity of defining fit.

4. Conclusions

Designs for well-fitting masks must accommodate diverse facial anatomies. Additionally, masks that initially fit well when donned can shift during talking, causing discomfort and leakage. This study demonstrated a methodology to evaluate dynamic fit for face masks using 3D head scans and Computer-Aided Design (CAD) software. In this study, we looked at the dynamic fit of reusable cloth face masks by scanning participants in both stationary open- and closed-mouth positions and after completing a standard talking exercise. The findings included method development, demonstrated with the three mask design variations, showing quantitative and qualitative differences. All three masks shifted up and down the nose relative to their initial position, with the Basic mask shifting the most and the Cup mask shifting the least. Other modes of poor fit, such as gaps between the mask perimeter and face, fabric buckling, and mask pressure compressing facial features, were observed from the mask overlays. Participants tended to underestimate the dynamic fit of the masks, but the subjective feedback is especially important for the comfort dimension of mask fit. Therefore, this study serves as a demonstration of the feasibility of this method and can be applied to a larger participant study to evaluate new face masks that better accommodate the action of talking and maintain minimal gapping between the face and mask boundary. It is a complementary methodology to other static and dynamic fit assessments and can provide rich amounts of data in the analysis stage of the iterative design process. This method can be applied to other headgear outside face masks and can be useful in both product development and regulation/standards development. Incorporating biodesign principles highlights how the use of advanced technology such as 3D scanning elevates and optimizes the design process.

This study has two primary limitations. Firstly, the participants were a convenience sample. The number of participants was useful to demonstrate the viability of the methodology, but a larger sample size representative of a diverse population is needed to generalize these results and to build a more robust dynamic face measurement size and shape model. Secondly, this study defined the open-mouth posture as the participant’s natural “ahh” pronunciation, where the mouth opening height is dependent on the participant and is closer to real world conditions. Instead, the mouth opening could be held constant across participants using a rigid mouthpiece.

New research questions were generated from this work. Considering the mask fit evaluation, can we designate a threshold of acceptable mask shift that could be considered a good fit? Can mask shift alone be used to qualify a mask as having a good or poor dynamic fit? Are there other markers we can look at that evaluate fit on more dimensions that can then be correlated to mask shift such as the fit factor determined with the PortaCount used in Quantitative Fit (QNFT) assessments? Fit factor from QNFT has a threshold of 100 for a respirator to pass the test and can be used as a benchmark to determine a goodness of fit threshold for mask shift. Can mask shift be used to predict a dynamic fit factor? There is a need to conduct additional studies to follow up on these questions.