18
M. Tiwari and N. Anand
Step 3. The results of the data analysis conducted in step 2 showed that the error
of scanner extracted measurements were observed beyond the prescribed acceptable
limits of ISO 20685:2005 and ISO 8559:1989. However, it was observed that the
error in the scanner extracted measurements was unidirectional.
Intra-class correlation coefficients (ICC) was conducted to check for the consistency and repeatability between different scan measurements and it represented the
variance attributable to error [8, 9]. ICC reflects both degree of correlation and agreement between measurements. Values less than 0.5 are indicative of poor reliability,
values between 0.5 and 0.75 indicate moderate reliability, values between 0.75 and
0.9 indicate good reliability, and values greater than 0.90 indicate excellent reliability
[10]. The ICC between the different scan measurements (for a given dimension) were
calculated using IBM SPSS Statistics 23.0. Please refer Table 2 for ICC values of
scanner measurements between one to other.
As indicated in Table 2, the higher values of ICC (> = 0.90) for each of the
dimension except for Neck girth. The reason for low ICC values for neck girth may
be due to the disturbance in scanning caused by hair, especially with female subjects.
This might have caused an error while the scanner identifies landmark to measure
the neck girth. The ICC values for male subjects were observed good (0.912, 0.947,
and 0.903) and confirmed a good level of consistency and reliability.
Bias-shift. As the scanner measurements were observed consistent and repeatable, a
correction factor in the form of bias-shift was incorporated into the scanner extracted
measurements. The concept of bias-shift is like a tare function were going forward,
the average difference between scanner measurement and the manual measurement
(as a standard correction value associated with specific body dimension) was adjusted
to the scanner extracted measurement. It is suggested that the two measurements
would become comparable by reducing the bias (mean) and/or random fluctuation
(SD) of errors, however, reducing mean error (bias) is more effective than reducing
SD to make the two measurements comparable [11]. A bias is a noticeable systematic
error in the scanner measurements, which can be used as an offset to correct the
scanner measurements in order to achieve improved concordance [12, 13].
Impact of applying bias-shift on the accuracy level achieved against ISO
20685:2005 prescribed error levels was checked by repeating the validation analysis
(as explained in step 2) after applying bias-shift as a correction factor. It was observed
that after incorporating the bias-shift, all the differences between the scanner and
manual measurements were observed within the ISO 20685:2005 prescribed acceptable limits (refer Table 3 for Summary results after incorporating bias-shift), hence the
scanner measurements could be considered comparable against the manual measurements, which were used as prescribed gold standards for checking the accuracy.
Hence, the scanner extracted measurements were considered as “PASS”.
Following the same procedure, the average difference between scanner measurements and the manual measurements for each dimension was also compared to
the ISO 8559:1989 prescribed limits. As indicated in Table 3, it was observed
that the error between scanner measurements and the manual measurements for all
M. Tiwari and N. Anand
Step 3. The results of the data analysis conducted in step 2 showed that the error
of scanner extracted measurements were observed beyond the prescribed acceptable
limits of ISO 20685:2005 and ISO 8559:1989. However, it was observed that the
error in the scanner extracted measurements was unidirectional.
Intra-class correlation coefficients (ICC) was conducted to check for the consistency and repeatability between different scan measurements and it represented the
variance attributable to error [8, 9]. ICC reflects both degree of correlation and agreement between measurements. Values less than 0.5 are indicative of poor reliability,
values between 0.5 and 0.75 indicate moderate reliability, values between 0.75 and
0.9 indicate good reliability, and values greater than 0.90 indicate excellent reliability
[10]. The ICC between the different scan measurements (for a given dimension) were
calculated using IBM SPSS Statistics 23.0. Please refer Table 2 for ICC values of
scanner measurements between one to other.
As indicated in Table 2, the higher values of ICC (> = 0.90) for each of the
dimension except for Neck girth. The reason for low ICC values for neck girth may
be due to the disturbance in scanning caused by hair, especially with female subjects.
This might have caused an error while the scanner identifies landmark to measure
the neck girth. The ICC values for male subjects were observed good (0.912, 0.947,
and 0.903) and confirmed a good level of consistency and reliability.
Bias-shift. As the scanner measurements were observed consistent and repeatable, a
correction factor in the form of bias-shift was incorporated into the scanner extracted
measurements. The concept of bias-shift is like a tare function were going forward,
the average difference between scanner measurement and the manual measurement
(as a standard correction value associated with specific body dimension) was adjusted
to the scanner extracted measurement. It is suggested that the two measurements
would become comparable by reducing the bias (mean) and/or random fluctuation
(SD) of errors, however, reducing mean error (bias) is more effective than reducing
SD to make the two measurements comparable [11]. A bias is a noticeable systematic
error in the scanner measurements, which can be used as an offset to correct the
scanner measurements in order to achieve improved concordance [12, 13].
Impact of applying bias-shift on the accuracy level achieved against ISO
20685:2005 prescribed error levels was checked by repeating the validation analysis
(as explained in step 2) after applying bias-shift as a correction factor. It was observed
that after incorporating the bias-shift, all the differences between the scanner and
manual measurements were observed within the ISO 20685:2005 prescribed acceptable limits (refer Table 3 for Summary results after incorporating bias-shift), hence the
scanner measurements could be considered comparable against the manual measurements, which were used as prescribed gold standards for checking the accuracy.
Hence, the scanner extracted measurements were considered as “PASS”.
Following the same procedure, the average difference between scanner measurements and the manual measurements for each dimension was also compared to
the ISO 8559:1989 prescribed limits. As indicated in Table 3, it was observed
that the error between scanner measurements and the manual measurements for all