As part of CIFASD, Blue Resonance LLC. is tasked with developing mHealth applications that would take our research and make it more widely available via mobile platforms. One of these applications was the piloting of a mobile app that could assist in the measurement of the facial features assessed during the evaluation of FASD. In particular, the assessment of papebral fissure length (PFL) is perhaps the most difficult. It is typically measured with a plastic mm ruler, or using 2D photogrammetry. Both have drawbacks. When measuring with a ruler, it is hard to reach the extremities of the eye as the exocanthion is seated a little deeper than the endocanthion, and the curvature of the eye makes it hard to get a linear measurement. With the second approach, the reference object placed on the forehead is not seated in the same vertical plane as the eye and therefore will lead to a measurement that underestimates the PFL. It also ignores the curvature of the eye, further diminishing the estimate. Problems with both the ruler and the photographs are easily documented. In one study, an individual’s PFL was measured with calipers (accurate measurement but not safe to use clinically) and found to be 28.02 mm. Eleven clinicians then measured the same PFL with a ruler and obtained a mean near 32 mm with a SD of over 6 mm. The range was 25-48. [This mean, SD and range were derived from a graph of these data and therefore may not be exact]. In another example physicians using a ruler were compared to 2D photogrammetric measures. The physicians were more than 1 mm off in 55% of the subjects. As PFL is one of three cardinal features of FAS, and some diagnostic guidelines only require two of the three cardinal features to be consistent with a diagnosis of FAS, it is critical to get accurate measurements.
Blue Resonance developed a mobile app to measure PFL using the TrueDepth camera on the iPhone X to implement its Face ID technology. This technology was leveraged to measure the PFL through a third-party software development kit (SDK) provided by Bellus 3D to generate a highly detailed 3D model with 500,000 data points. It takes less than 20 seconds to scan a face and a PFL measure in millimeters rounded to two decimals (e.g., 26.45 mm) is computed. The Bellus 3D SDK uses the open-source library Stasm along with OpenCV to landmark the exo and endocanthion. However, the landmarks are not accurately placed every time, so a feature was added to the app to allow the user the ability to easily correct the landmark position Users can even zoom in for ultra-precise placement.
The validity of this technique was verified in several ways. First, an object of known size (19.05 mm) was placed on the forehead of an individual being scanned. The app reported a size of 19.1 mm. Next we scanned 3 people, 10 times each. The average SD was 0.47 mm. with a range of 0.41 to 0.62. Recall that the smallest unit on the mm ruler is 1 mm, and the SD from the one study mentioned above was over 6 mm. Finally, we used a mannequin and measured the PFL with calipers (23.85 mm). We then took 14 scans of the mannequin, and the mean PFL automatically calculated by the app was 23.28 mm with a SD of 0.49 mm. As mentioned above the endo and exocanthion landmarks can be corrected and when this was done on the 14 scans, the mean was 23.67 with a SD of 0.20 mm.
We then took 14 scans of the mannequin, and the mean PFL automatically calculated by the app was 23.28 mm with a SD of 0.49 mm (range 22.2-23.9). Less than 15% of measures were over a 1 mm difference compared to 55% of the measures comparing the ruler measure mentioned above). As mentioned above the endo and exocanthion landmarks can be corrected and when this was done on the 14 scans, the mean was 23.67 with a SD of 0.20 mm (range 23.2-23.9 – None of the measures differed by more than 1 mm).
In summary, this app for assessing PFL appears to be highly reliable and accurate at a submillimeter level. This is a tremendous increase in accuracy compared to the typical techniques. The measurement takes under 20 seconds, and gives the clinician the ability to correct any issues if incorrect auto landmarking occurs. Furthermore, the PFL can be computed by individuals with little or no experience in measuring facial features.
References
Astley, S.J., Palpebral fissure length measurement: accuracy of the FAS facial photographic analysis software and inaccuracy of the ruler. Journal Of Population Therapeutics And Clinical Pharmacology Journal De La Therapeutique Des Populations Et De La Pharamcologie Clinique, 2015. 22(1): p. e9-e26.