The Auto-eFACE: Machine Learning–Enhanced Program Yields Automated Facial Palsy Assessment Tool
Problem
Facial palsy assessment relied on subjective, nonstandardized clinician-graded scales prone to observer bias, making treatment comparison unreliable.
Why it matters
Subjective grading undermines consistent diagnosis and comparison of facial palsy treatments.
What the AI did
A machine learning Python script (auto-eFACE) automatically generated facial palsy scores from 160 photographs, which were compared against clinician eFACE grading.
Results
Auto-eFACE differentiated normal vs palsy faces; normal faces scored 93.83 ± 4.37 vs clinician 100.00 ± 1.58 (p = 0.01)
Impact
Evidence supports an automated tool that distinguishes palsy faces but diverges from clinician scores on normal faces.
Record
| Venue | Plastic & Reconstructive Surgery |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Matthew Q. Miller, Tessa A. Hadlock, Emily Fortier, Diego L. Guarín |
| Citations | 75 (OpenAlex, at import time) |
| References | 60 |
| Model or method | not stated in source |
| Organization | Massachusetts Eye and Ear Infirmary |
| Primary source | https://doi.org/10.1097/prs.0000000000007572 |
| Open access | closed |
| Keywords | facial palsy, eFACE, machine learning, automated grading, photographs |
Importance rating
real but narrower advance
Standardizing facial palsy grading could reduce observer bias and aid treatment comparison; meaningful clinical benefit but narrow scope, 160 photographs only.