Page 132 - IJB-7-4

Mass Customization of Respiratory Protective Equipment
standard adult BMI classification : underweight (BMI of the participants were male (80.7%), or White (76.1%),
[76]
< 18.5), healthy (18.5 < BMI ≤ 24.9), overweight (25.0 or having BMI more than 25 (85.8%). The average RMSE
< BMI < 29.9), and obese (BMI > 30.0). For ethnicity, Euclidean distance between a mask surface and a face is
participants were grouped into the broad categories of 0.62 ± 0.20 mm (mean ± sd) for all 197 scans. This means
Asian, White, and Others (including Black, Arabs, and that the mask model was able to achieve an average sub-
any other ethnicity) as reported by participants. millimeter accuracy. The Maximum Euclidean distance is
2.03 ± 0.75 mm (mean ± sd), located on nasal sidewalls.
3. Results and discussion Lee et al. used a virtual fit testing method on 3D facial
[77]
Using our pipeline, 197 of the 205 included scans data of 336 Korean Air Force men and found that the
returned a valid CAD model and 8 scans were rejected gap between a pilot oxygen mask (similar to a reusable
by the alignment checking algorithms in the pipeline elastomeric half-face respirator) at the nasal bridge is
(misalignment between input scan and template mesh), approximately 6 mm when pressing the mask tightly onto
which gives an overall processing success rate of 96.1%. a face (10 mm into the cheek). A maximum 2 mm gap
Table 1 shows the respective average run time for the at the nasal sidewalls instead of the nasal bridge, which
processed 197 scans obtained from Bellus3D, ScandyPro has been reported as the most commonly injured site for
[1]
and Lightstage. Compared to a manual process the pipeline respirator users , is a significant improvement from the
has achieved significant time-savings. The average run mass-produced RPE. In addition, contrary to the design
time for processing a single scan was <2 min (104.4) bias in current commercially available RPE, where fit
that a significant portion of which time was consumed by was poorer among females and Asians, the Maximum
the template fitting sub-process. The underlying code for Euclidean distance was slightly lower for females
this sub-process has not been optimized for the graphics (1.82 mm) as compared to males (2.08 mm), and Asians
processing unit (GPU) computation; hence, further (1.80 mm) as compared to the others (White: 2.05 mm
time savings may be attained through GPU rather than and others: 2.33 mm).
CPU computation. This is a significant improvement To investigate if the proposed pipeline could
from earlier manual process studies. This time saving is produce respirator masks that fit equally well to people
crucial to scale up the production for mass-customizing with different demographic backgrounds, the following
respirator masks, which has a much higher demand than hypothesis was tested: there will be no difference between
conventionally customized orthotics or prosthetics. In the maximum or RMSE Euclidean distance across each
addition, the customisation process is now completely subcategory under age, gender, ethnicity or BMI. Choice
automated which eliminates labor cost. Cazon et al. of an appropriate statistical test is dependent on the nature
[61]
have reported that labor cost can take up to 80% of of the Euclidean distance results. Before deciding on
the entire AM design cost; therefore, such savings is an appropriate statistical test, Shapiro–Wilk tests were
significant for making AM a more viable production conducted to check for normality of the distributions
method for the MC of respirator masks. Notably, the for each grouped RMSE and maximum Euclidean
average run time per scan for ScandyPro scans was the distances. Most of the demographic groups rejected
longest (at almost 4 min/scan) compared to Bellus3D the null hypothesis (P < 0.05) that the data is normally
(<1 min/scan) and Light stage scans (<1.5 min/scan). distributed with unspecified mean and variance, except
This extended runtime is owed to the fact that ScandyPro for the Maximum Euclidean Distance in Female group
scans contained significantly greater number of vertices (P = 0.12), RMSE Euclidean distance in Other Ethnicity
than the other two. In addition, the ScandyPro scans group (P = 0.31), and RMSE Euclidean distance in
included noisier data, including large neck/shoulder parts Healthy BMI group (P = 0.23). Nonetheless, for these
and many disconnected regions scattered around the face. groups, the sample size is small (≤30). Therefore, the
Collectively, these resulted in longer time to perform the nonparametric Wilcoxon Rank Sum Test (confidence
template fitting process. level at 95%) was used to compare outcomes between
Table 2 shows participants’ profiles grouped into each pair of subcategories in age, gender, ethnicity and
the demographic categories, and their corresponding BMI. The underweight group was excluded as the sample
Maximum and RMSE Euclidean distances. The majority size is too small (n = 1).
Table 1. Average run time. Wilcoxon Rank Sum test results (Table 3) showed
Acquisition method Average run time, mean (sd) that there is no statistically significant difference for
All 104.4 (95.2) RMSE and maximum Euclidean distances between Male
and Female, Asian and White, White and others, healthy
Bellus3D 52.5 (25.7) and overweight, overweight and obese, Middle age, and
ScandyPro 235.5 (89.9) senior groups. This suggests that the pipeline produced
Lightstage 133.3 (56.8) mask models that fit equally well across these groups. This

128 International Journal of Bioprinting (2021)–Volume 7, Issue 4