The spectrum of facial palsy: The MEEI facial palsy photo and video standard set

Reestablishment of the Smile after Hypoglossal-Facial Nerve Transfer: What Can We Learn?

Objective  The aim of this study was to assess the ability to smile following a hypoglossal-facial nerve transfer (N12-N7). Design  This is a retrospective chart review. Setting  National tertiary referral center for skull base pathology. Participants  Seventeen patients. Main Outcome Measures  The ability to smile following an N12-N7 transfer was assessed by five medical doctors on photographs of the whole face and frontal, orbital, and oral segments. The (segmented) photographs were scored for the symmetry, asymmetry, and correct or incorrect assessment of the affected side. Results  Seventeen patients were analyzed by 5 assessors providing 85 assessments. The whole face at rest was judged symmetrical in 26% of the cases and mildly asymmetrical in 56%. Frontal, orbital, and oral segments were symmetrical in 63, 20, and 35%, respectively. The affected side was correctly identified in 76%. When smiling, the whole face was symmetrical in 6% and mildly asymmetric in 59%. The affected side was correctly identified in 94%. The frontal, orbital, and oral segments during smiling were symmetrical in 67, 15, and 6%, respectively. The affected side of the frontal, orbital, and buccal facial segments during smiling was correctly identified in 89, 89, and 96%, respectively. Interobserver variability with Fleiss' kappa analysis showed that the strength of the agreement during smile of the total face was good (0.771) Conclusion  Following an N12-N7 transfer, a good facial symmetry at rest can be achieved. During smiling, almost all patients showed asymmetry of the face, which was predominantly determined by the orbital and oral segments. To improve the ability to smile after an N12-N7 transfer, additional procedures are needed.

Artificial Intelligence-Based Facial Palsy Evaluation: A Survey

Facial palsy evaluation (FPE) aims to assess facial palsy severity of patients, which plays a vital role in facial functional treatment and rehabilitation. The traditional manners of FPE are based on subjective judgment by clinicians, which may ultimately depend on individual experience. Compared with subjective and manual evaluation, objective and automated evaluation using artificial intelligence (AI) has shown great promise in improving traditional manners and recently received significant attention. The motivation of this survey paper is mainly to provide a systemic review that would guide researchers in conducting their future research work and thus make automatic FPE applicable in real-life situations. In this survey, we comprehensively review the state-of-the-art development of AI-based FPE. First, we summarize the general pipeline of FPE systems with the related background introduction. Following this pipeline, we introduce the existing public databases and give the widely used objective evaluation metrics of FPE. In addition, the preprocessing methods in FPE are described. Then, we provide an overview of selected key publications from 2008 and summarize the state-of-the-art methods of FPE that are designed based on AI techniques. Finally, we extensively discuss the current research challenges faced by FPE and provide insights about potential future directions for advancing state-of-the-art research in this field.

Assessing Facial Palsy: Does Feedback Improve Assessment Using the eFACE and Sunnybrook Facial Grading System?

OBJECTIVE(S)

To explore learning effects when applying the clinician-graded electronic facial function scale (eFACE) and the Sunnybrook Facial Grading System (Sunnybrook).

METHODS

Surgeons, facial rehabilitation therapists, and medical students were randomly allocated to the eFACE (n = 7) or Sunnybrook (n = 6) and graded 60 videos (Massachusetts Eye and Ear Infirmary open-source standard set); 10 persons with normal facial function and 50 patients with a wide variation of facial palsy severity. Participants received an introduction and individual feedback after each set of 10 videos. Scores were compared to the reference score provided with the set. Multilevel analysis was performed to analyze learning effect.

RESULTS

A learning effect was only found for the eFACE, with significant difference scores in set 1 and 2 compared to set 6, and no significant difference scores in the following sets. The difference score was associated with the reference score (severity of facial palsy) for eFACE (β = -0.19; SE = 0.04; p < 0.001) and Sunnybrook (β = -0.15; SE = 0.04; p < 0.001). Age of participants was also associated with the difference score in the eFACE group (β = 0.18; SE = 0.03; p < 0.001). No differences in scores were found between groups of participants.

CONCLUSION

The eFACE showed a learning effect of feedback while the Sunnybrook did not.

LEVEL OF EVIDENCE

NA Laryngoscope, 134:3105-3111, 2024.

Intelligent bell facial paralysis assessment: a facial recognition model using improved SSD network

With the continuous progress of technology, the subject of life science plays an increasingly important role, among which the application of artificial intelligence in the medical field has attracted more and more attention. Bell facial palsy, a neurological ailment characterized by facial muscle weakness or paralysis, exerts a profound impact on patients' facial expressions and masticatory abilities, thereby inflicting considerable distress upon their overall quality of life and mental well-being. In this study, we designed a facial attribute recognition model specifically for individuals with Bell's facial palsy. The model utilizes an enhanced SSD network and scientific computing to perform a graded assessment of the patients' condition. By replacing the VGG network with a more efficient backbone, we improved the model's accuracy and significantly reduced its computational burden. The results show that the improved SSD network has an average precision of 87.9% in the classification of light, middle and severe facial palsy, and effectively performs the classification of patients with facial palsy, where scientific calculations also increase the precision of the classification. This is also one of the most significant contributions of this article, which provides intelligent means and objective data for future research on intelligent diagnosis and treatment as well as progressive rehabilitation.

Comparison of Objective Facial Metrics on Both Sides of the Face Among Patients with Severe Bell's Palsy Treated with Mirror Effect Plus Protocol Rehabilitation Versus Controls

Objective: The extent to which the healthy hemiface dynamically contributes to facial synchronization during facial rehabilitation has been largely unstudied. This study compares the synchronization of both hemifaces in severe Bell's palsy patients who either received facial rehabilitation called "Mirror Effect Plus Protocol" (MEPP) or basic counseling. Methods: Baseline and 1-year postonset data from 39 patients (19 = MEPP and 20 = basic counseling) were retrospectively analyzed using Emotrics+, a software that generates facial metrics with artificial intelligence (AI) algorithms. Paired t-tests were used for intrasubject comparisons of hemifaces, and mixed model analysis were used to compare between groups. Results: For voluntary movements, a significant difference in favor of the MEPP group was only found for smiling (p = 0.025*). However, at 1-year postonset, the control group showed significant variability between hemifaces for most synkinesis measurements [nasolabial fold (p = 0.029*); eye area (p = 0.043*); palpebral fissure (p = 0.011*)]. Conclusion: In this study, a better synchronization of both hemifaces was found in the MEPP group. Interestingly, motor adaptation in movement amplitude of the healthy hemiface seemed to contribute to this synchronization in MEPP patients. Further studies are needed to standardize the procedure of AI measurements and to adapt it for clinical use.

Mid-Term Effects of Selective Denervation Surgery on Facial Symmetry in Patients with Nonflaccid Facial Paralysis: Intermediate Follow-Up of Rest and Smile Symmetry

Introduction: Although selective denervation (SD) improves clinician-graded, objective, and patient-reported smile symmetry, changes in resting symmetry have not been comprehensively studied. Objective: To assess mid-term changes in resting facial symmetry after SD, and to evaluate changes in perceived emotions in faces at rest and with smiling. Methods: Nonflaccid facial paralysis (NFFP) patients undergoing SD were studied from September 2021 to October 2022. Patient-reported, clinician-graded, and objective metrics of resting and smile symmetry were quantified. Results: Fourteen patients (median age 59.5 years; median paralysis duration 76.5 months; median follow-up 249.5 days (138-400); average of 5.7 branches sacrificed) underwent SD. Resting oral commissure (OC) symmetry (p = 0.028) and upper lip symmetry (p = 0.030) improved after surgery. In smiling faces, OC excursion (p = 0.004), smile angle (p = 0.016), and dental show (p = 0.012) improved. There were significant increases in perceived happiness at rest (p = 0.006) and with smile (p = 0.002). Clinician grading revealed improved nasolabial fold depth at rest (p = 0.087) after surgery, and patients reported improved facial function and happiness with the decision to undergo surgery. Conclusion: Intermediate follow-up suggests SD can improve objective resting and smile symmetry in patients with NFFP. After surgery, patients' faces convey more happiness in repose and with smile.

Effect of Filler Therapy on Psychosocial Distress in Facial Paralysis: Patients with Major Asymmetry Derive Greater Benefit

Background: Facial filler is an effective nonsurgical treatment option for improving facial symmetry in patients with facial paralysis (FP). Objective: To compare the effects of filler among patients with FP that is self-perceived as major or minor asymmetry, by measuring psychosocial distress. Methods: In this prospective cohort study of patients with FP undergoing filler at a tertiary academic center, patients were classified as having minor or major self-perceived asymmetry using a visual analog scale (VAS). FACE-Q Appearance-Related Psychosocial Distress was administered before and after filler. Descriptive statistics and a random-effects generalized linear model assessed the relationship between perceived facial asymmetry and change in psychosocial distress. Results: A total of 28 patients participated. Twenty-five (89%) patients were female with median age of 54 (interquartile range [IQR]: 49-66). Median VAS score was 2 (IQR: 1-3.5, 0 = completely asymmetric, 10 = no asymmetry). Psychosocial distress improved in all patients after filler. In multivariable modeling, patients with major asymmetry experienced 2.45 (confidence interval: 0.46-4.44, p = 0.016) points more improvement in psychosocial distress than patients with minor asymmetry. Age, gender, and FP duration were not associated with change in psychosocial distress. Conclusion: Facial filler treatment was seen to improve psychosocial distress in patients with FP, especially by those with more self-perceived deficit.

Measuring the Potential Effects of Mirror Therapy Added to the Gold Standard Facial Neuromuscular Retraining in Patients With Chronic Peripheral Facial Palsy: Protocol for a Randomized Controlled Trial

BACKGROUND

Facial neuromuscular retraining (fNMR) is a noninvasive physical therapy widely used to treat peripheral facial palsies. It consists of different intervention methods that aim to reduce the debilitating sequelae of the disease. Recently, the use of mirror therapy in the acute facial palsy and postsurgical rehabilitation contexts has shown promising results, suggesting its use as an adjunct to fNMR in treating patients with later stages of paralysis, such as the paretic, early, or chronic synkinetic.

OBJECTIVE

The main aim of this study is to compare the efficacy of an added mirror therapy component with fNMR in patients with peripheral facial palsy (PFP) sequelae in 3 different stages. The specific objectives of this study are to measure the effects of combined therapy compared to fNMR alone on (1) participants' facial symmetry and synkinesis, (2) quality of life and psychological aspects of the participants, (3) motivation and treatment adherence, and (4) different stages of facial palsies.

METHODS

This study is a randomized controlled trial that compares the effect of fNMR combined with mirror therapy (experimental group: n=45) with fNMR alone (control group: n=45) in 90 patients with peripheral facial palsy presenting with sequelae 3-12 months after onset. Both groups will receive 6 months of rehabilitation training. Facial symmetry and synkinesis; participants' quality of life; and their psychological factors, motivation, and compliance will be assessed at baseline (T0), 3 months (T1), 6 months (T2), and 12 months (T3) postintervention. Outcome measures are (1) changes in facial symmetry and synkinesis assessed with facial grading tools, (2) quality of life changes with patient questionnaires, and (3) therapy motivation with a standardized scale, as well as adherence to treatment with metadata. Changes in facial symmetry and synkinesis will be judged by 3 assessors blinded to group assignment. Mixed models and Kruskal-Wallis, chi-square, and multilevel analyses will be conducted according to the appropriate variable type.

RESULTS

Inclusion will start in 2024 and is anticipated to be completed in 2027. The 12-month follow-up will be completed with the last patient in 2028. We expect patients included in this study to experience improvement in facial symmetry, synkinesis, and quality of life, regardless of group assignments. A potential benefit of mirror therapy for facial symmetry and synkinesis could be noted for patients in the paretic phase. We hypothesize better motivation and adherence to treatment for the mirror therapy group.

CONCLUSIONS

The results of this trial may provide new guidelines for PFP rehabilitation with patients dealing with long-term sequelae. It also fills the need for robust evidence-based data in behavioral facial rehabilitation.

INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID)

PRR1-10.2196/47709.

Middle Fossa Decompression for Recurrent Facial Palsy: Prevalence and Surgical Outcomes

OBJECTIVE

To investigate the surgical outcomes in patients treated for recurrent facial nerve palsy (RFP) at a quaternary facial nerve referral center.

METHODS

A retrospective chart review was performed on 132 patients with RFP who presented to our institution's facial nerve clinic from 2001 to 2021. Records were analyzed for etiology of palsy, facial nerve function, and recurrence rates. Pre- and post-operative audiometric outcomes were also assessed in surgically managed patients.

RESULTS

6.8% of RFP patients underwent surgical decompression. For patients who did not undergo surgery, the House-Brackmann (HB) score was 2.9 ± 1.3 (SD) at the initial clinic visit, and 2.4 ± 1.3 (SD) at the last clinic visit. This difference was significantly different (p = 0.01, t-test). For surgical patients, the pre-operative HB score was 2.9 ± 0.9 (SD) and post-operative HB score was 1.8 ± 0.6 (SD), which were significantly different (p = 0.01, t-test). The number of facial palsy episodes also decreased pre- and post-operatively from 3.5 ± 0.8 (SD) to 0.2 ± 0.4 (SD) episodes, which were significantly different (p < 0,001, t-test). Audiometric outcomes were not significantly different pre- and post-surgery (p = 0.31, t-test for PTA; p = 0.34, t-test for WRS).

CONCLUSION

Facial nerve decompression for RFP patients with incomplete functional recovery may be an effective treatment for decreasing the frequency and severity of facial palsy episodes.

LEVEL OF EVIDENCE

4 Laryngoscope, 133:1222-1227, 2023.

Reducing Risk in Facial Reanimation Surgery

Facial reanimation surgery can greatly improve quality of life, but these procedures are not without risk. Important considerations for risk reduction in facial reanimation surgery include preoperative risk-stratification, protecting patients' clinical media, clearly and thoroughly setting expectations, and intraoperative strategies to maximize technical success and minimize operative time.

Facial Nerve Exploration for Known or Suspected Transection: The Massachusetts Eye and Ear Experience

Background: There is no ideal test to determine likelihood of spontaneous recovery after post-traumatic and postsurgical facial palsy (FP). Objective: Among patients with unexpected FP undergoing facial nerve (FN) exploration for suspected discontinuity, we endeavored to discern whether intraoperative findings, repair type, and time to exploration impact FN recovery, as measured by electronic Facial Clinimetric Evaluation (eFACE) and FaCE scales. Methods: Retrospective cohort study of 42 adult patients who underwent FN exploration. Results: FN injury resulted from either surgery (n = 29) or trauma (n = 13). Average time to repair was 68.4 (standard deviation 79.6) days. Postoperative improvements were observed in total eFACE (73.3-86.5; p < 0.0001) and FaCE (21.5-38.1; p = 0.0214) scores. Distal FN injuries were most common (n = 29) and had best recovery (percentage change in eFACE 57.2% vs. 34.3% main trunk, p = 0.0306). Discontinuity injuries (n = 33) repaired with primary coaptation (n = 18) had noninferior outcomes compared with cable graft repair (n = 16; percentage change in eFACE 49.6% vs. 39.2%, p = 0.3470). Denervation times <3 months yielded better recovery using percentage change in eFACE score (56.9% vs. 33.1%, p = 0.0270). Conclusions: Surgical exploration for unexpected FP allows for direct visualization of anatomical nerve status and timely repair.

Contribution and safety of the side-to-end hypoglossal-to-facial transfer in multidisciplinary facial reanimation

BACKGROUND

This study evaluates facial and tongue function in patients undergoing side-to-end hypoglossal-to-facial transfer (HFT) with additional techniques.

METHODS

Thirty-seven patients underwent a side-to-end HFT. Twelve had additional cross-face grafts, and 9 had an additional masseter-to-facial transfer. Facial was assessed with House-Brackmann (HB), Sunnybrook Facial Grading Scale (SFGS), and eFACE. Martins scale and the Oral-Pharyngeal Disability Index (OPDI) were used to assess tongue function.

RESULTS

Ninety-four percent of cases reached HB grades III-IV. Mean total SFGS score improved from 16 ± 15 to 59 ± 11, while total eFACE score from 52 ± 13 to 80 ± 5. Dual nerve transfers were a predictor for a better eFACE total score p = 0.034, β = 2.350 [95% CI, 0.184-4.516]), as well as for a higher SFGS total score (p = 0.036, β = 5.412 [95% CI, 0.375-10.449]). All patients had Martin's grade I. Mean postoperative OPDI scores were 84 ± 17 (local physical), 69 ± 16 (simple and sensory motor components), 82 ± 14 (complex functions), and 73 ± 22 (psychosocial).

CONCLUSIONS

The side-to-end HFT offers predictable facial function outcome and preserves tongue function in nearly all cases. Dual nerve transfers appear to improve the final outcome.

Towards Facial Gesture Recognition in Photographs of Patients with Facial Palsy

Humans express their emotions verbally and through actions, and hence emotions play a fundamental role in facial expressions and body gestures. Facial expression recognition is a popular topic in security, healthcare, entertainment, advertisement, education, and robotics. Detecting facial expressions via gesture recognition is a complex and challenging problem, especially in persons who suffer face impairments, such as patients with facial paralysis. Facial palsy or paralysis refers to the incapacity to move the facial muscles on one or both sides of the face. This work proposes a methodology based on neural networks and handcrafted features to recognize six gestures in patients with facial palsy. The proposed facial palsy gesture recognition system is designed and evaluated on a publicly available database with good results as a first attempt to perform this task in the medical field. We conclude that, to recognize facial gestures in patients with facial paralysis, the severity of the damage has to be considered because paralyzed organs exhibit different behavior than do healthy ones, and any recognition system must be capable of discerning these behaviors.

The Auto-eFACE: Machine Learning-Enhanced Program Yields Automated Facial Palsy Assessment Tool

BACKGROUND

Facial palsy assessment is nonstandardized. Clinician-graded scales are limited by subjectivity and observer bias. Computer-aided grading would be desirable to achieve conformity in facial palsy assessment and to compare the effectiveness of treatments. This research compares the clinician-graded eFACE scale to machine learning-derived automated assessments (auto-eFACE).

METHODS

The Massachusetts Eye and Ear Infirmary Standard Facial Palsy Dataset was employed. Clinician-graded eFACE assessment was performed on 160 photographs. A Python script was used to automatically generate auto-eFACE scores on the same photographs. eFACE and auto-eFACE scores were compared for normal, flaccidly paralyzed, and synkinetic faces.

RESULTS

Auto-eFACE and eFACE scores differentiated normal faces from those with facial palsy. Auto-eFACE produced significantly lower scores than eFACE for normal faces (93.83 ± 4.37 versus 100.00 ± 1.58; p = 0.01). Review of photographs revealed minor facial asymmetries in normal faces that clinicians tend to disregard. Auto-eFACE reported better facial symmetry in patients with flaccid paralysis (59.96 ± 5.80) and severe synkinesis (62.35 ± 9.35) than clinician-graded eFACE (52.20 ± 3.39 and 54.22 ± 5.35, respectively; p = 0.080 and p = 0.080, respectively); this result trended toward significance.

CONCLUSIONS

Auto-eFACE scores can be obtained automatically using a freely available machine learning-based computer software. Automated scores predicted more asymmetry in normal patients, and less asymmetry in patients with flaccid palsy and synkinesis, compared to clinician grading. Auto-eFACE is a quick and easy-to-use assessment tool that holds promise for standardization of facial palsy outcome measures and may eliminate observer bias seen in clinician-graded scales.

CLINICAL QUESTION/LEVEL OF EVIDENCE

Diagnostic, III.

A New Dataset for Facial Motion Analysis in Individuals With Neurological Disorders

We present the first public dataset with videos of oro-facial gestures performed by individuals with oro-facial impairment due to neurological disorders, such as amyotrophic lateral sclerosis (ALS) and stroke. Perceptual clinical scores from trained clinicians are provided as metadata. Manual annotation of facial landmarks is also provided for a subset of over 3300 frames. Through extensive experiments with multiple facial landmark detection algorithms, including state-of-the-art convolutional neural network (CNN) models, we demonstrated the presence of bias in the landmark localization accuracy of pre-trained face alignment approaches in our participant groups. The pre-trained models produced higher errors in the two clinical groups compared to age-matched healthy control subjects. We also investigated how this bias changes when the existing models are fine-tuned using data from the target population. The release of this dataset aims to propel the development of face alignment algorithms robust to the presence of oro-facial impairment, support the automatic analysis and recognition of oro-facial gestures, enhance the automatic identification of neurological diseases, as well as the estimation of disease severity from videos and images.

Facial Paralysis Detection on Images Using Key Point Analysis

The inability to move the muscles of the face on one or both sides is known as facial paralysis, which may affect the ability of the patient to speak, blink, swallow saliva, eat, or communicate through natural facial expressions. The well-being of the patient could also be negatively affected. Computer-based systems as a means to detect facial paralysis are important in the development of standardized tools for medical assessment, treatment, and monitoring; additionally, they are expected to provide user-friendly tools for patient monitoring at home. In this work, a methodology to detect facial paralysis in a face photograph is proposed. A system consisting of three modules—facial landmark extraction, facial measure computation, and facial paralysis classification—was designed. Our facial measures aim to identify asymmetry levels within the face elements using facial landmarks, and a binary classifier based on a multi-layer perceptron approach provides an output label. The Weka suite was selected to design the classifier and implement the learning algorithm. Tests on publicly available databases reveal outstanding classification results on images, showing that our methodology that was used to design a binary classifier can be expanded to other databases with great results, even if the participants do not execute similar facial expressions.

Reliability Between In-Person and Still Photograph Assessment of Facial Function in Facial Paralysis Using the eFACE Facial Grading System

Importance: Assessing facial function using high-quality photographs would improve long-term and objective outcomes tracking in facial palsy, facilitate automated facial grading using artificial intelligence algorithms, and allow for remote follow up. Objective: To determine agreement between in-person and photographic electronic facial function scale (eFACE) assessments, and evaluate inter-rater reliability of photographic eFACE evaluation. Design, Setting, and Participants: Retrospective review of eFACE scores from in-person interviews and standardized photographs using the Massachusetts Eye and Ear (MEE) Standard Facial Palsy Dataset. Main Outcomes and Measures: eFACE total scores and subset scores determined by two experienced facial reanimation surgeons in person and from photographs. Results: Intraclass correlation coefficients for eFACE scores were 0.96 (95% CI 0.94 to 0.97) for total scores, 0.99 (95% CI 0.989 to 0.995) for static scores, 0.82 (95% CI 0.74 to 0.88) for dynamic scores, and 0.98 (95% CI 0.97 to 0.99) for synkinesis scores. Photographic and in-person scores had a mean difference of -0.64 (95% CI -2.05 to 0.77; p = 0.37) for total score, -1.58 (95% CI -4.22 to 1.05; p = 0.24) for the static subset, 0.14 (95% CI -1.70 to 1.97; p = 0.88) for the dynamic subset, and -1.11 (95% CI -3.09 to 0.86; p = 0.26) for the synkinetic subset. Bland-Altman analysis showed no trend for increasing differences in total score or subset scores. Conclusions: eFACE assessment obtained via photographs exhibits excellent inter-rater reliability and strong agreement with in-person assessment, demonstrating facial symmetry in facial palsy patients can be monitored using standardized frontal photographs.

Toward an Automatic System for Computer-Aided Assessment in Facial Palsy

Importance: Quantitative assessment of facial function is challenging, and subjective grading scales such as House-Brackmann, Sunnybrook, and eFACE have well-recognized limitations. Machine learning (ML) approaches to facial landmark localization carry great clinical potential as they enable high-throughput automated quantification of relevant facial metrics from photographs and videos. However, the translation from research settings to clinical application still requires important improvements. Objective: To develop a novel ML algorithm for fast and accurate localization of facial landmarks in photographs of facial palsy patients and utilize this technology as part of an automated computer-aided diagnosis system. Design, Setting, and Participants: Portrait photographs of 8 expressions obtained from 200 facial palsy patients and 10 healthy participants were manually annotated by localizing 68 facial landmarks in each photograph and by 3 trained clinicians using a custom graphical user interface. A novel ML model for automated facial landmark localization was trained using this disease-specific database. Algorithm accuracy was compared with manual markings and the output of a model trained using a larger database consisting only of healthy subjects. Main Outcomes and Measurements: Root mean square error normalized by the interocular distance (NRMSE) of facial landmark localization between prediction of ML algorithm and manually localized landmarks. Results: Publicly available algorithms for facial landmark localization provide poor localization accuracy when applied to photographs of patients compared with photographs of healthy controls (NRMSE, 8.56 ± 2.16 vs. 7.09 ± 2.34, p ≪ 0.01). We found significant improvement in facial landmark localization accuracy for the facial palsy patient population when using a model trained with a relatively small number photographs (1440) of patients compared with a model trained using several thousand more images of healthy faces (NRMSE, 6.03 ± 2.43 vs. 8.56 ± 2.16, p ≪ 0.01). Conclusions and Relevance: Retraining a computer vision facial landmark detection model with fewer than 1600 annotated images of patients significantly improved landmark detection performance in frontal view photographs of this population. The new annotated database and facial landmark localization model represent the first steps toward an automatic system for computer-aided assessment in facial palsy. Level of Evidence: 4.