The Role of Facial Expression Analysis and Electrodermal Activity as an Objective Evaluation of Persistent Idiopathic Facial Pain.

ABSTRACT: Persistent idiopathic facial pain (PIFP) is an enigmatic condition presenting with variable features. Psychiatric comorbidities are speculated to influence PIFP.In this study, the authors evaluated patients with PIFP through the hospital anxiety and depression scale, facial expression analysis, and electrodermal activity.A total of 67 respondents enrolled as the experimental group and 28 participants as the control group. Pain scores were higher in the experimental group (5.24; SD 2.349) P < 0.001 depression (5.58 (SD 3.766) versus 3.07 (SD 2.418), P = 0.002) and anxiety scores (9.78 (SD 4.923) versus 6.75 (SD 4.097) P = 0.007) were higher in the experimental group. The experimental group expressed more negative episodes (P = 0.024); Electrodermal Activity data in terms of peaks/min (P = 0.872) and average peak amplitude (P = 0.168) were not significantly different between the groups.It may be concluded that pain levels may be influenced by psychiatric comorbidity as PIFP patients showed insignificant physiological response to pain.

Orofacial pain diagnosis is often complicated, especially when no organic pain or primary orofacial pain disorders are detected through multiple radiological and clinical examinations.[1] In the facial region, that type of pain is related to persistent idiopathic facial pain (PIFP). Diagnosis is also complicated by possible psychiatric comorbidity.[2,3] In this study, we used self-evaluating questionnaires to assess patients’ with PIFP tendency toward affective (mood) disorders and facial expression analysis (FEA) in combination with electrodermal activity (EDA) to objectify the questionnaire data and physiological pain response.

MATERIALS AND METHODS

This was a single-center, cross-sectional, controlled, parallel-group study conducted in the Hospital of LUHS Kaunas Clinics, Department of Oral and Maxillofacial surgery. All respondents signed an informed consent form.

Inclusion/Exclusion Criteria

Adult patients who visited the Hospital of LUHS Kaunas clinics during 2019–2021 were recalled for inclusion in this study. The primary inclusion criteria were that their last known diagnosis in the inner patient register system of the Hospital of LUHS detected as G50.1 (ICD-10) Atypical Facial Pain. After further examination and clarifying whether the condition could be accounted for PIFP (according to ICOP 6.2 PIFP diagnostic criteria), patients were included in the study.[1] We excluded patients with diagnosed psychiatric disorders, systemic diseases which may interfere with this study or head and neck oncology, patients with a history of addiction to drugs or alcohol, and pregnant women.

Control group patients were selected with chronic unilateral postfracture mandibular pain, (S02.6 according to ICD-10); 8 weeks after intermaxillary fixation. Exclusion criteria were the same as for the experimental group.

All respondents within this study were asked to discontinue their medication, which may affect pain perception if no negative consequences were expected, 24 hours prior to participation in the study.

Survey Data

Patient pain was assessed using a Visual Analogue Scale (0 = no pain, 10 = unbearable or unimaginable pain).[4] Patients’ tendency toward depression and anxiety was assessed using the Hospital Anxiety and Depression Scale.[5]

Physiological/Biometric data

During the interview, electrodermal activity (EDA) was collected from the left hand with electrodes placed on the palmar side of the medial part of the index and middle finger. The EDA signal was collected using a wireless EDA device (Shimmer3 GSR, Shimmer, Dublin, Ireland) sampling at 128 Hz. Moreover, a video (Mp4 format, 30 fps) of the respondent's face was stored for later postprocessing of facial expressions.

Data Analysis

Electrodermal Activity

The EDA signal from each respondent was analyzed using a peak detection analysis workflow with the purpose of extracting phasic events (peaks). In brief, lowpass Butterworth filters with a cutoff frequency of 5 Hz were applied to remove line noise. A moving average median filter with a window length of 8 seconds was then applied to the filtered signal to extract the phasic component. Lastly, a trough-to-peak analysis was applied to identify phasic events in the signal. A peak was identified as an event lasting for at least 500 ms with an onset threshold of 0.05 μS and a minimum amplitude of 0.01 μS. In this paper, the count of peaks/min was used for subsequent data analysis. This value is thought to represent the emotional arousal level of the respondent with a higher peak/min count signifying higher levels of arousal.[6]

Facial Expression Analysis

Analysis of the face recordings was post-processed in the iMotions software using the embedded facial expression analysis module (Affdex 4.0, Affectiva, Boston, MA).[7] The algorithm outputs likelihood scores for certain facial expressions being present for each video frame. In the present study, valence was used for further analysis. Valence is a probability score that ranges from 100 to −100 where facial expressions of smile and cheek raise increase the likelihood of a positive score, and the facial expression of inner brow raise, brow furrow, nose wrinkle, upper lip raise, lip corner depressor, chin raise, lip press, and lip suck increase the likelihood of negative score.[8] A valence score above 30% likelihood was considered as positive valence, a score between 30 and −30% likelihood as neutral valence, and a score below −30% likelihood was considered negative valence.

Statistical Analysis

The statistical analysis was performed using IBM SPSS Statistics 20 software with significance level was selected at P < 0.05. The interdependence of the qualitative evidence was evaluated by chi-square (χ2) criteria. We used a nonparametric Mann-Whitney test to evaluate questionnaire, GSR, and facial expression analysis differences between the groups.

RESULTS

Sixty-seven respondents enrolled as the experimental group and 28 enrolled as the control group. There were no differences between the groups in terms of age (P = 0.644) and gender (P = 0.087). Pain expression was more severe in the experimental group (mean - 5.24; SD 2.349) than in the control group (1.96; SD 2.027), P < 0.001. Regarding depression and anxiety scores, the experimental group expressed both higher depression (P = 0.002) and anxiety (P = 0.007) scores (Supplementary Digital Content, Table 1).

Based on the facial expression analysis, the experimental group patients on average expressed more negative emotions, which include anger, sadness, and fear compared to the control group patients (Supplementary Digital Content, Table 2).

Recordings of EDA revealed that there were no significant differences in EDA peaks/min and average peak amplitude throughout the recording, however, EDA peaks per minute were significantly higher in the experimental group during negative valence (Supplementary Digital Content, Table 3).

DISCUSSION

In this study, we investigated patients suffering from PIFP through self-evaluating questionnaires and recorded their emotional status using FEA and EDA. To our knowledge, this is the first study to evaluate patients with PIFP through subjective self-evaluating questionnaires and an objective emotional arousal evaluation system.

In this study, we found that patients with PIFP may have a tendency toward both depression and anxiety. It is speculated in the literature that the comorbidity between PIFP and affective (mood) disorders exists, although a direct causative relationship has yet to be explained.[4] As with other chronic pain conditions, it is more likely to be a shared vulnerability to both psychiatric comorbidities and chronic pain conditions.[4]

Considering facial expression analysis, patients with PIFP expressed more negative emotions than control group patients. Although it is difficult to estimate whether facial expression recognition may be a reliant marker for affective (mood) disorders, higher negative emotion outputs may be related to higher experienced pain levels.

It is known that EDA phasic events (peaks) may increase (higher peaks/min) when a patient is experiencing higher levels of pain.[9] When comparing our EDA data, we have noted that EDA peaks/min did not differ between the groups, nor did their average amplitude of EDA peaks. This evidence is important since the experimental group patients reported higher experienced pain levels compared to the control group, yet their physiological response to that pain was highly homogeneous with the lower-pain control group. It may be concluded that pain levels may be influenced by psychiatric comorbidity as PIFP patients showed insignificant physiological response to pain. A take-away message is that multidisciplinary treatment including psychiatrists may be indicated when considering PIFP diagnosis.