Is There a Difference in Facial Emotion Recognition after Stroke with vs. without Central Facial Paresis?

The Facial Feedback Hypothesis (FFH) states that facial emotion recognition is based on the imitation of facial emotional expressions and the processing of physiological feedback. In the light of limited and contradictory evidence, this hypothesis is still being debated. Therefore, in the present study, emotion recognition was tested in patients with central facial paresis after stroke. Performance in facial vs. auditory emotion recognition was assessed in patients with vs. without facial paresis. The accuracy of objective facial emotion recognition was significantly lower in patients with vs. without facial paresis and also in comparison to healthy controls. Moreover, for patients with facial paresis, the accuracy measure for facial emotion recognition was significantly worse than that for auditory emotion recognition. Finally, in patients with facial paresis, the subjective judgements of their own facial emotion recognition abilities differed strongly from their objective performances. This pattern of results demonstrates a specific deficit in facial emotion recognition in central facial paresis and thus provides support for the FFH and points out certain effects of stroke.


Introduction
Emotion recognition is omnipresent in social interactions [1] and represents an important social competence [2]. Faces provide relevant clues for the recognition of emotions [2,3]. One explanation of the facial recognition of emotions is provided by the Facial Feedback Hypothesis (FFH) [4]. The present study therefore compares stroke patients with vs. without unilateral central facial paresis, i.e., the partial inability to perform facial movements [5], in order to test the FFH prediction of a specific deficit of visual facial emotion recognition in individuals with central facial paresis.

Emotion Processing and the Role of Facial Feedback
Facial emotion expressions are part of nonverbal communication [3] and are regarded as some of the most important nonverbal features in the identification of emotions [6]. Facial expression can be highly variable due to the precise control of the different facial muscles [1] and their voluntary or affective control [7], although the basic emotions framework considers a set of emotions to be highly elementary, unique and independent of culture, time and place [8]. These basic emotions are: anger, disgust, fear, joy, sadness and surprise [9,10]. Each of the basic emotions is characterized by specific patterns of facial of a patient with bilateral facial paralysis in Guillain-Barré syndrome, nor did Bogart and Matsumoto [35] report facial emotion recognition deficits in patients with congenital bilateral facial paresis in Moebius syndrome. However, Calder et al. [36] did observe differences in the accuracy of emotion recognition with respect to at least one basic emotion in patients with Moebius syndrome.
A different way of investigating facial feedback in healthy participants is with an injection of botulinum toxin in the facial muscles for temporarily paralysis. Different studies using this method showed changed emotion recognition in terms of accuracy and time [22,32]. The results may point to a direct link between facial feedback and emotion processing [32].
Besides limited facial movements due to experimental induction and peripheral facial palsy, other disorders could also affect (1) facial movements and (2) facial emotion recognition-for instance, central facial palsy after stroke and Parkinson's disease. Stroke occurs suddenly due to disturbed blood flow and oxygen deficiency (ischemic stroke) or bleeding (hemorrhagic stroke) in the brain and leads to individual disabilities [37], whereas Parkinson's disease is a neurodegenerative disorder involving loss of dopamine in the substantia nigra, resulting in typical symptoms of rigor, tremor and bradykinesia [38]. Both central facial palsy after stroke [26,39] and Parkinson's disease [40][41][42] could result in similar effects, i.e., reduced facial expression and therefore reduced facial feedback. Following the FFH, facial feedback due to facial integrity is needed for facial emotion recognition [23]. Both in stroke [43] and in Parkinson's disease [41], facial emotion recognition could be impaired. However, there is not necessarily a direct correlation between limitations in facial expression and facial emotion recognition, at least in Parkinson's disease [41].
In summary, there is evidence that patients with limited facial feedback and facial mimicry abilities (e.g., in peripheral facial paresis) are potentially affected by limited facial emotion recognition. To date, to the best of our knowledge, patients with peripheral facial palsy have been studied, whereas patients with central facial palsy have been overlooked.
The care of patients with central facial palsy is insufficient and rehabilitation guidelines are required [44]. To improve treatment and establish guidelines, deficits or remaining abilities must be identified first. To this end, we designed a study to find proof of facial emotion recognition abilities in patients with central facial palsy.
Consequently, the aim of the study was to test facial emotion recognition in patients with central facial paresis after stroke in terms of accuracy and time with visually presented, i.e., facial, stimuli presented by healthy subjects. Testing different modalities (facial and auditory) in two patient groups (with or without facial paresis after stroke) allows assessment of whether there is a general deficit in emotion recognition-which is a possibility after stroke [43]-or whether only one particular modality is (more) affected. If there are no deficits in emotion recognition at all, i.e., if the performance is comparable to that of healthy control subjects, it can assume that emotion recognition may be intact. Accordingly, the primary research question was: Can patients with central facial paresis after stroke recognise facial emotions?

Participants
Three groups of participants were considered for this study: (1) patients with unilateral central facial paresis after stroke, (2) patients without facial paresis after stroke and (3) healthy subjects. The data for the patient groups (1) and (2) were collected within the study (data are available from the authors on request), whereas the reference values for the healthy subject group (3) were already available [45][46][47] and served for an additional comparison.
The inclusion and exclusion criteria are summarised in Table 1. The patients were referred by various cooperation partners, hospitals and local practices for speech-language therapy. Recruitment and data collection took place in the period from 22 February until 14 May 2019 in Germany. A total of 67 patients were recruited. Four of these were drop-out cases (one case: disorientation; one case: suspected bucco-facial apraxia with no possibility of assessing facial paresis; two cases: antidepressant medication with suspected altered emotional regulation). The remaining 63 patients were assigned to the study group (patients with central facial paresis, n = 34) or the control group (patients without facial paresis, n = 29) according to their diagnosis of facial paresis. Sociodemographic data and information on lesions, facial paresis, general mental capacities and aphasia for the study and control groups are given in Tables A1-A3, A5 and A6 (Appendix A).
The study was approved by the local ethics committee (key: EK 271/18) of the Medical Faculty at RWTH Aachen University, and all regulations of the ethics committee were implemented. All experiments were performed in accordance with the relevant guidelines and regulations. All participants signed an informed consent form after receiving detailed information.

Materials
For both facial emotion recognition and auditory emotion recognition, the same conditions were set, i.e., an item was presented (visually or auditorily) and the patients had ten seconds to respond. There were different options available as answers. The respective software systems recorded accuracy and time. For both modalities, a pre-test with ten items (initially randomized, later presented in the same order) was performed. The pre-test ensured that the task was understood [48] (see, also, Appendix B).

Visual Facial Emotion Recognition
In our study, we opted to use the Myfacetraining (MFT) Program (CRAFTA Cranio Facial Therapy Academy, Hamburg, Germany) [47,49], which consists of a standard test for accuracy and time taken for facial emotion recognition [47,49]. Forty-two subjects, each showing a basic emotion with their face, were presented on a screen. The person was first shown in a neutral position before changing to an emotional facial expression (basic emotion). Six additional answer options were displayed on the screen according to the basic emotion [47] (see, also, Appendix B).

Auditory Emotion Recognition
In addition to faces, voices (auditory) are the most important modalities in emotional communication [1]. A sub-portion of the Montreal Affective Voices (MAVs) [45] was used for the assessment. These are emotional, non-linguistic, vocal expressions of /a/ (to be compared with a as in apple, British English). Sixty items for the six basic emotions [45] were used. The Montreal Affective Voices were presented with a specially programmed experiment with the software PsychoPy, version 3.0.0b9 [50] (see, also, Appendix B). Coulson et al. [51] asked relatives of patients with facial paresis for their assessments of emotional recognition. Based on this, two standardized questionnaires were designed for the present study which enabled the systematic collection of subjective facial emotion recognition data. The Self-Assessment Questionnaires Emotion Recognition Accuracy and Time were used to document self-assessment of facial emotion recognition of the six basic emotions (anger, disgust, fear, joy, sadness and surprise) [51]. In order to be able to look at the evaluation in a differentiated way, one questionnaire was developed to assess accuracy and another was developed to assess time taken for facial emotion recognition. The questionnaires assess possible changes between pre-morbid and current abilities per basic emotion. The questions that featured in the questionnaires in each case were as follows: How well do you recognize the following feelings in other people's faces? One of three answer options could be selected for each questionnaire. For Accuracy, the patient evaluated whether the basic emotion in question was more difficult, just as well as or more easily recognised than before stroke. For Time, the patient indicated whether the basic emotion was detected slower, as fast as or faster than before stroke. For deteriorations (indicated by the response options more difficult or slower), a score of −1 was assigned. If the patient did not notice any changes (response options just as well as or just as fast as), zero points (0) were recorded. For improvements (answer options easier or faster), the patient achieved a score of +1, resulting in a score between −6 and +6 per questionnaire.

Sunnybrook Facial Grading System for Diagnosing Facial Paresis
In order to answer the main research question, all patients were examined in a standardised way to identify possible facial paresis. Only this allowed to divide the patients into the study group (participants with central facial paresis) or the control group (participants without central facial paresis). The Sunnybrook Facial Grading System [52,53] is used for the standardised assessment for diagnosing facial paresis or paralysis, respectively. This measurement method is explicitly recommended [54]. It is also considered the current standard in the evaluation of facial paresis [55] and has been used in various studies (e.g., [54,[56][57][58][59][60][61][62]). Ross et al. [52] published the original version of the Sunnybrook Facial Grading System in 1996, which was implemented in the present study (German version [53]). For this purpose, a video was made of each patient with an Apple iPod touch (camera at right angles, at the individual height of the chewing plane, 150 cm from the patient's chin), in which the patients were asked in a standardised manner to show their face at rest or to perform an arbitrary movement with their face (raise eyebrows, close eyes gently, smile with open mouth, show teeth, pucker lips). The videos were evaluated by a speech-language therapist (see, also, Appendix B).

Statistical Analysis
Two-factorial ANOVAs with post-hoc t-tests were performed with the factors group (with vs. without facial paresis) as between-subject factor and modality (facial vs. auditory emotion recognition) as within-subject factor. Accuracy and time taken for emotion recognition were considered as dependent variables. In order to compare the empirical data obtained in the present study with normative data for healthy controls (without stroke and without facial paresis) which were already available, a series of t-tests were subsequently performed separately both for accuracy and time. To compare facial emotion recognition and auditory emotion recognition in terms of accuracy and time in patients and healthy subjects, t-tests were performed for one sample. For the comparison between patients with and without facial paresis, two-factorial ANOVAs and (post-hoc) t-tests for independent samples were run. t-tests for dependent samples were performed to compare facial emotion recognition and auditory emotion recognition in patients with and without facial paresis. To analyse subjective emotion recognition in terms of accuracy and time, one-sample t-tests were conducted. To compare accuracy and time, t-tests for dependent samples were performed.
Benjamini-Hochberg correction was applied if more than one t-test was conducted.

Results
The results for objective (accuracy and time) and subjectively perceived success in emotion recognition are summarised in Figure 1, Figure 2, Figure 3, Figure 4 and Table A4 (Appendix A).

Statistical Analysis
Two-factorial ANOVAs with post-hoc t-tests were performed with the factors group (with vs. without facial paresis) as between-subject factor and modality (facial vs. auditory emotion recognition) as within-subject factor. Accuracy and time taken for emotion recognition were considered as dependent variables. In order to compare the empirical data obtained in the present study with normative data for healthy controls (without stroke and without facial paresis) which were already available, a series of t-tests were subsequently performed separately both for accuracy and time. To compare facial emotion recognition and auditory emotion recognition in terms of accuracy and time in patients and healthy subjects, t-tests were performed for one sample. For the comparison between patients with and without facial paresis, two-factorial ANOVAs and (post-hoc) t-tests for independent samples were run. t-tests for dependent samples were performed to compare facial emotion recognition and auditory emotion recognition in patients with and without facial paresis. To analyse subjective emotion recognition in terms of accuracy and time, one-sample t-tests were conducted. To compare accuracy and time, t-tests for dependent samples were performed.
Benjamini-Hochberg correction was applied if more than one t-test was conducted.

Results
The results for objective (accuracy and time) and subjectively perceived success in emotion recognition are summarised in Figures 1-4 and Table A4 (Appendix A).

Accuracy of Facial Emotion Recognition
The results of the ANOVA for accuracy were a main effect of group F(1;61) = 6.620; p = 0.013, a main effect of modality F(1;61) = 96.535; p < 0.001 and an interaction effect group x modality F(1;61) = 18.330; p < 0.001, which means that participants with central facial paresis recognised visually presented basic emotions significantly worse (reduced accuracy) compared to participants without facial paresis (t(49.425) = −3.767; p < 0.001; after correction p = 0.002) and compared to healthy controls (t(33) = −22.888; p < 0.001; after correction p = 0.002). Participants without facial paresis recognised visually presented basic emotions significantly worse (reduced accuracy) compared to healthy controls (t(28) = −10.476; p < 0.001; after correction p = 0.002), Figure 1. Accuracy of facial emotion recognition (mean, median, interquartile range). Participants after stroke with facial paresis performed significantly worse compared to healthy controls (p < 0.001) and compared to participants after stroke without facial paresis (p < 0.001). The data for healthy controls were not collected in this study but were taken from [46,47], so no information on the actual distribution of the data is available but only the mean as an indicator of the central tendency. Therefore, the figures only contain two box plots, not three. Figure 1. Accuracy of facial emotion recognition (mean, median, interquartile range). Participants after stroke with facial paresis performed significantly worse compared to healthy controls (p < 0.001) and compared to participants after stroke without facial paresis (p < 0.001). The data for healthy controls were not collected in this study but were taken from [46,47], so no information on the actual distribution of the data is available but only the mean as an indicator of the central tendency. Therefore, the figures only contain two box plots, not three.

Accuracy of Auditory Emotion Recognition
Participants with central facial paresis recognised auditorily presented basic emotions significantly worse (reduced accuracy) compared to healthy controls (t(33) = −13.258; p < 0.001; after correction p = 0.002). Participants without facial paresis recognised auditorily presented basic emotions significantly worse (reduced accuracy) compared to healthy controls (t(28) = −11.259; p < 0.001; after correction p = 0.002). Participants with vs. without central facial paresis did not differ significantly in auditory emotion recognition (accuracy) (t(61) = 0.616; p = 0.540; after correction p = 0.540), Figure 2. Accuracy of auditory emotion recognition (mean, median, interquartile range). Participants after stroke with facial paresis performed significantly worse compared to healthy controls (p < 0.001) but did not differ significantly compared to participants after stroke without facial paresis (p = 0.540). The data for healthy controls were not collected in this study but were taken from [45], so no information on the actual distribution of the data is available but only the mean as an indicator of the central tendency. Therefore, the figures only contain two box plots, not three.

Comparison of Accuracy of Facial and Auditory Emotion Recognition
Participants with central facial paresis recognised visually presented basic emotions significantly worse (reduced accuracy) than auditorily presented basic emotions (t(33) = −11.252; p < 0.001; after correction p = 0.002). Participants without facial paresis recognised visually presented basic emotions significantly worse (reduced accuracy) than auditorily presented basic emotions (t(28) = −3.485; p = 0.002; after correction p = 0.002). Figure 2. Accuracy of auditory emotion recognition (mean, median, interquartile range). Participants after stroke with facial paresis performed significantly worse compared to healthy controls (p < 0.001) but did not differ significantly compared to participants after stroke without facial paresis (p = 0.540). The data for healthy controls were not collected in this study but were taken from [45], so no information on the actual distribution of the data is available but only the mean as an indicator of the central tendency. Therefore, the figures only contain two box plots, not three.  Participants after stroke with facial paresis performed significantly faster compared to healthy controls (p = 0.02) but did not differ significantly compared to participants after stroke without facial paresis (p = 0.68). The data for healthy controls were not collected in this study but were taken from [46,47], so no information on the actual distribution of the data is available but only the mean as an indicator of the central tendency. Therefore, the figures only contain two box plots, not three.

Time Taken for Auditory Emotion Recognition
Participants with vs. without central facial paresis did not differ significantly with respect to the average time taken for auditory emotion recognition (t(61) = −1.851; p = 0.069), Figure 4. Participants after stroke with facial paresis did not differ significantly compared to participants after stroke without facial paresis (p = 0.069).

Comparison of Time Taken for Facial and Auditory Emotion Recognition
Participants with central facial paresis recognised visually presented basic emotions significantly (not significantly after correction) faster (increased time) than auditorily presented basic emotions (t(33) = −2.269; p = 0.030; after correction p = 0.060). Participants without facial paresis recognised visually presented basic emotions not significantly differently to auditorily presented basic emotions (t(28) = −0.041; p = 0.968; after correction p = 0.968).

Figure 3.
Average time of facial emotion recognition (mean, median, interquartile range). Participants after stroke with facial paresis performed significantly faster compared to healthy controls (p = 0.02) but did not differ significantly compared to participants after stroke without facial paresis (p = 0.68). The data for healthy controls were not collected in this study but were taken from [46,47], so no information on the actual distribution of the data is available but only the mean as an indicator of the central tendency. Therefore, the figures only contain two box plots, not three.  Participants after stroke with facial paresis performed significantly faster compared to healthy controls (p = 0.02) but did not differ significantly compared to participants after stroke without facial paresis (p = 0.68). The data for healthy controls were not collected in this study but were taken from [46,47], so no information on the actual distribution of the data is available but only the mean as an indicator of the central tendency. Therefore, the figures only contain two box plots, not three.

Time Taken for Auditory Emotion Recognition
Participants with vs. without central facial paresis did not differ significantly with respect to the average time taken for auditory emotion recognition (t (61)   Participants after stroke with facial paresis did not differ significantly compared to participants after stroke without facial paresis (p = 0.069).

Comparison of Time Taken for Facial and Auditory Emotion Recognition
Participants with central facial paresis recognised visually presented basic emotions significantly (not significantly after correction) faster (increased time) than auditorily presented basic emotions (t(33) = −2.269; p = 0.030; after correction p = 0.060). Participants without facial paresis recognised visually presented basic emotions not significantly differently to auditorily presented basic emotions (t(28) = −0.041; p = 0.968; after correction p = 0.968). Participants after stroke with facial paresis did not differ significantly compared to participants after stroke without facial paresis (p = 0.069).

Time Taken for Auditory Emotion Recognition
Participants with vs. without central facial paresis did not differ significantly with respect to the average time taken for auditory emotion recognition (t(61) = −1.851; p = 0.069), Figure 4.

Further Analysis
In order to verify that the identified pattern is reasonable on the basis of these results, the following further control calculations were made.
A correlation calculation (Pearson's product moment correlation) between objective accuracy and objective time taken for facial emotion recognition in patients with and without central facial paresis was performed. The accuracy of and the time taken for facial emotion recognition in patients with central facial paresis were positively correlated with

Further Analysis
In order to verify that the identified pattern is reasonable on the basis of these results, the following further control calculations were made.
A correlation calculation (Pearson's product moment correlation) between objective accuracy and objective time taken for facial emotion recognition in patients with and without central facial paresis was performed. The accuracy of and the time taken for facial emotion recognition in patients with central facial paresis were positively correlated with each other (r = 0.729; p < 0.001). The average accuracy and the average time taken for facial emotion recognition in patients without facial paresis were not significantly correlated with each other (r = 0.291; p = 0.126).
Furthermore, a correlation calculation (Pearson's product moment correlation) between objective facial emotion recognition, accuracy and severity of facial paresis using the Sunnybrook Facial Grading System across all patients (with and without facial paresis) was performed. The average accuracy of facial emotion recognition and the severity of facial paresis were significantly positively correlated with each other (r = 0.31; p = 0.014).
Furthermore, a chi-squared test to compare the number of patients with limitations in general mental capacity in both groups (Table A5, Appendix A) was performed. Both groups were comparable, with x 2 (1, n = 63) = 0.204; p = 0.651. Another chi-squared test to compare the number of patients with aphasia in both groups (Table A6, Appendix A) was also carried out. Both groups were comparable, with x 2 (1, n = 63) = 1.546; p = 0.214.
Additionally, univariate and multivariate regressions, with emotion recognition (facial and auditory, accuracy and time taken) as the dependent variable and predictors diagnosis of facial paresis, sex, age, subjective judgement, general mental capacity and time post-onset as independent variables, were conducted (Tables A7 and A8, Appendix A). Patients with facial paresis recognised visually presented basic emotions significantly worse (reduced accuracy) compared to patients without facial paresis, as calculated by means of univariate regression (beta = −0.444; p < 0.001) as well as by multivariate regression (beta = −0.353; 0.003).

Discussion
This study investigated visual facial emotion recognition (VFER) in patients with and without central facial paresis vs. healthy individuals. The results of our study showed that the participants with central facial paresis had significantly lower average accurate emotion recognition abilities with respect to the facial modality compared to the auditory modality. The less accurate VFER in cases of facial paresis but not in auditory emotion recognition may be due to changes in the facial feedback mechanism. Clinically, this means that VFER in persons with limited facial mimicry abilities, as in central facial paresis patients, does appear to be affected, in contrast to auditory recognition [36]. Taking into account that we did not test facial mimicry itself (i.e., facial muscle activity was not measured during the emotion-recognition task), but facial emotion recognition, facial paresis can be inferred to be one factor influencing the accuracy of objective facial emotion recognition, which may be affected by changes in the facial feedback mechanism. This may be an indication that the accuracy of objective facial emotion recognition is especially limited when facial feedback is altered by facial paresis. Auditory performance does not appear to be affected by facial paresis (for a similar finding, cf. [36]). Besides facial paresis, stroke, also, could be one factor influencing the accuracy of objective facial emotion recognition in our sample. All participants (with and without facial paresis) had had at least one stroke. Since stroke may also cause deficits in emotion recognition [43], our examined patient groups may be affected as well. These two potential factors (altered facial feedback and altered central processing due to stroke) indicate the relevance of and need to study patients without stroke but with limited facial feedback-for example, patients with peripheral facial palsy.
Our results reveal significant deficits in terms of accuracy of facial emotion recognition, in contrast with other studies that did not report any differences, e.g., [27,28,34]. This fact may be due to the large sample size (participants with facial paresis: n = 34; participants without facial paresis: n = 29) and the inclusion of different phases post-onset, with a wide range since the time of stroke (day 5 up to day 6361 post-onset). However, previous studies reported significant limitations in terms of average time taken for facial emotion recognition, e.g., [27,28], while the participants in the present study showed faster reaction times. This, in turn, could indicate that the participants after stroke replied quick and dirty [63], while they suffered from other impairments, such as deficits in attention, concentration and memory [64], in addition to the facial paresis after stroke. In order to investigate a possible systemic connection between the fast, inaccurate responses, the significant positive correlation between the objective accuracy and the objective time taken for facial emotion recognition in patients with facial paresis provides further insight: the faster a patient with facial paresis responded, the less accurate was the response, whereas no correlation was found in patients without facial paresis. This could indicate that the patients with facial paresis were themselves aware of their deficit in the time taken for facial emotion recognition (as reported in the Self-Assessment Questionnaires Emotion Recognition) but wanted to show their best performance in the test situation and therefore answered as quickly as possible.
The participants with facial paresis subjectively felt limited both in terms of parameter accuracy and time in VFER. They stated that they were more impaired with respect to time than accuracy. The participants felt that facial emotion recognition had slowed down considerably since the stroke and was somewhat less accurate. These results provide a new insight into subjective emotion recognition, as this was not considered in previous studies. However, the clinical measurement gave contradictory results and showed that the patients were clearly less accurate but faster. Thus, the measured performance appears to be controversial to the subjectively perceived performance.
In the present study, we considered the difference in facial and auditory emotion recognition shown in the results. This may support, for example, FFH, as mentioned before. Nevertheless, it should be noted that a large part of human emotion is communicated via the face and the voice, as discussed in the literature. To the best of our knowledge, this is the first clinical study which combines two different modalities in a clinical setting [65]. The mentioned factors (limitations such as deficits in attention, concentration and memory [64], besides facial paresis and emotion recognition) influenced both the study results and everyday communication in the patient groups. Although for stroke patients their survival is of primary importance [66], participation is also highly relevant, particularly in the post-acute and chronic phase [67]. Since both groups of patients showed a significant reduction in the accuracy of facial and auditory emotion recognition compared to healthy subjects, intervention recommendations for both groups and both modalities are required. Although there is limited evidence for FFH [68], FHH can be used as an explanation for assessment and rehabilitation [69].

The Relevance of Assessment of Emotion Recognition
The described results not only provide evidence for the FFH and certain effects of stroke but also have implications for the treatment of patients with central facial paresis after stroke. As early as 2013, Dobel et al. [69] called for the examination of facial emotion recognition in patients with facial paresis using basic emotions [69]. In summary, the present study supports this demand and once again advocates it.
Since the accuracy of facial emotion recognition can be impaired, especially in patients with facial paresis after stroke, appropriate assessment and therapy is recommended for this patient group. Deficits should be assessed because the performance limitations may have negative consequences on communication and may increase over time. If the performance of emotion recognition remains impaired, this can lead to the development of disorders such as alexithymia (the inability to recognise or describe one's own emotions) [11,70]. For example, if sadness is not adequately interpreted, a patient may react defensively and thus not appropriately to a situation [6]. The effects of facial emotion recognition are therefore far-reaching and decisive for adequate social contact. The somewhat controversial results for the objective measurement and subjective assessment of facial emotion recognition in participants with facial paresis require detailed and individual examination in clinical practice. It is not sufficient to either ask the patient for his or her opinion or carry out an objective diagnosis. Both options should be taken and the results should be compared.
In addition, the lack of disease insight to be expected according to the available results (comparison between clinical measurement and subjective assessment) must become a focus of treatment in order to show the patient the relevance of facial emotion recognition therapy. This should not underestimate the importance of considering the individual wishes and goals of the patient and including them in the sense of joint decision making [71]. The basis for this is the tripartite evidence-based practice [71,72]. This ensures not only the effectiveness and efficiency of therapy, but also therapy motivation and transfer into the patient's everyday life [71].

Limitations of the Study
The composition of the sample may be considered a limiting factor of the study. A larger and more representative, homogeneous sample tested at the same time post-onset after stroke and subdivided according to the subtypes of central facial paresis (voluntary and involuntary central facial paresis [73]) would therefore be desirable for future studies.
For a more precise observation of the lesion localization and comparability of patients, imaging with detailed description of affected brain areas would be useful. In addition, statistical adjustment for different stroke locations and lesion sizes would be beneficial, as differences in emotion recognition could depend on the hemisphere affected [43]. Despite the possibility of different lesion locations and lesion sizes, the results for facial emotion recognition showed significant differences between the patient groups. Since significant effects can already be observed in our sample, we expect similar or stronger effects to be observed with more carefully selected samples with stricter inclusion criteria in further studies. Furthermore, a strong and reliable test battery to assess cognitive capacity (see [74]) is needed to differentiate deficits in emotion recognition and limitations on general mental capacity after stroke. Since emotion perception depends on general mental capacity [74][75][76], any emotion perception test measures general mental capacity to some degree. In the present study, there were comparable numbers of patients with limitations in mental capacity and aphasia, as proven by chi-squared tests. In future studies, comparability should be extended and improved by standardised diagnostics.
However, the significant positive correlation observed between objective facial emotion recognition accuracy and severity of facial paresis, calculated using the Sunnybrook Facial Grading System across all patients, points to facial paresis as the main differentiator between the two patient groups. Thus, the higher the accuracy of facial emotion recognition, the higher the score on the Sunnybrook Facial Grading System. That is, facial competence correlates with accuracy in facial emotion recognition, or the lower the facial competence, the worse the accuracy in facial emotion recognition. Moreover, significant univariate and multivariate regressions documented the relation between facial emotion recognition accuracy and facial paresis. These results demonstrate the influence of facial paresis on facial emotion recognition once more, but only in terms of accuracy. No significant differences were detected with respect to objective facial emotion recognition accuracy and time taken between patients with left-or right-sided facial paresis. If one hemisphere is dominant in emotion processing [43], patients with lesions in this dominant hemisphere with contralateral facial paresis [25] could possibly be more affected. We cannot confirm this hypothesis and previous research on facial palsy that reported that patients with left-sided facial palsy showed lower performance in facial emotion recognition compared with patients with right-sided facial palsy [29]. However, our results are in line with findings for patients with Parkinson's disease, where facial asymmetry is not related to hemispheric dominance for emotion processing [77]. Further evidence is needed, then, to inspect possible differences in facial emotion recognition and expression depending on the side affected with facial palsy and on hemisphere.
Perfect comparability of the standard data with the sample data cannot be guaranteed without gaps-for instance, due to the age of the participants (e.g., the Montreal Affective Voices validation sample with an average age of 23.3 ± 3 years [45] vs. the patients with facial paresis with an average age of 62.6 ± 9.3 years and patients without facial paresis average with an average age of 58.4 ± 10.7 years). It must also be noted that only a small sample size of normative data (n = 29) was used for the auditory emotion recognition assessment (Montreal Affective Voices) [45]. Furthermore, the measurement of auditory and facial emotion recognition is not completely comparable. Especially with regard to the time taken for emotion recognition, it should be noted, for example, that the response modes differed (selecting an option on screen vs. pointing to a surface) and that the numbers of items and response options were not identical. As a consequence, for further research, normative data from healthy individuals should be freshly collected, with comparability extended to the patient groups. Moreover, measurement in facial and auditory emotion recognition tasks should be made even more comparable.
The separate presentation of facial and auditory items in emotion recognition assessments should also be critically questioned. Facial and auditory expressions are not necessarily independent as they can mutually influence their recognition. For example, a facial expression can be generated by moving the mouth while a vocal expression is also made [1]. However, a separation of the modalities, i.e., just visual or just auditory impressions, seemed to make sense in this study in order to differentiate and compare performances. In order to be able to answer the question reliably, this seems unavoidable. At the same time, however, this separate type of emotion recognition is far removed from everyday life and thus reduces the external validity. Equally adapted to optimal experimental conditions, static photographs instead of everyday situations were used [78]. A person is able to show up to 8000 different emotional facial expressions with his or her face [17]. However, it should be critically noted that our study only examined emotion recognition with respect to basic emotions and thus minimized the requirements compared to nonverbal communication in everyday life. It should be noted here that basic emotions can be regarded as the basis for far more complex emotions or emotional states [8]. However, since the recognition of the comparatively primitive basic emotions [8] was assessed as limited in the present study, an even worse performance can be expected for more complex emotions.

Conclusions
From this study, it may be concluded that: -After a stroke, participants with central facial paresis were significantly less accurate in visually recognising basic emotions compared with stroke patients without facial paresis and compared with a sample of healthy controls; -Auditory emotion recognition in both stroke groups was less accurate than in the control sample; - The facial emotion recognition accuracy of participants with central facial paresis was significantly worse than the auditory accuracy of emotion recognition; -Since visual emotion recognition was clearly worse than auditory emotion recognition in participants with facial paresis after stroke, facial mimicry probably plays an important role in communication with patients after stroke; - The results of our observational study may indicate the overall effects of stroke on emotion recognition and support the FFH, which is a practical and appropriate model implemented in clinical assessments and interventions; -Future research should investigate patients with facial palsy without stroke to further explore the impact of facial feedback on emotion recognition. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to their having been collected as part of a larger research project that has not yet been completed.     n.a. [45] Subjective facial emotion recognition via Self-Assessment Questionnaires Emotion Recognition Accuracy    With aphasia n = 6 n = 9 Without aphasia n = 28 n = 20 Note: n = number of participants. Each participant was asked to assess an emotion by selecting a response option [81]. Following the original software [80], the participant selected one of the response options (one of the six basic emotions or neutral/unknown) by pointing at a surface (A4 size). Ten seconds of time were allowed for response to each task.
As in objective facial emotion recognition, a pre-test with ten items (initially randomised, later presented in the same order) was performed too. In addition, the examiner checked that the headphones were comfortably fitted. The volume was adjusted individually [45]. Questions asked of the patient regarding the test procedure were answered. However, no assistance was given with regard to the content of the test.
Standard values are available for accuracy (in percentages) of emotion recognition: mean = 72,67 ± 11.66; min. = 56.00; max. = 86.00 (see, also, Figure 2 and Table A4, Appendix A). However, no data were collected for time taken [45]. As proposed by Belin et al. [45] and explained above, MAVs (selected items, adapted to the circumstances of this study) were used. The MAVs, as material for auditory emotion recognition assessment, are explicitly recommended for comparisons of facial emotion recognition. They are particularly well-suited, since only the auditory modality is addressed. Furthermore, the MAVs do not contain any linguistic information, which excludes distortion or aggravated conditions for patients with aphasia [45]. Mild aphasia was not necessarily a criterion for exclusion in this study (see Table 1).

Appendix B.4 Sunnybrook Facial Grading System for Diagnosing Facial Palsy
With the Sunnybrook Facial Grading System, each face was rated in three areas by comparing the affected side of the face with the intact side. This resulted in three values: (1) Resting Symmetry Score (symmetry at rest), (2) Voluntary Movement Score (symmetry of voluntary movements) and (3) Synkinesis Score (synkinesis). With these three scores, a total score (0-100 points) was calculated. The lower the total score, the more pronounced the facial paresis respectively paralysis. The authors did not give any recommendation for a further classification according to degree of severity or the point value for a diagnosis of facial palsy actually made [52,53]. For the present study, however, an unambiguous diagnosis of the presence of facial paresis seemed indispensable to classify the participants into the appropriate target or control groups (with or without facial paresis). The severity classification of the present study was therefore based on the procedures of the House-Brackman Facial Nerve Grading System [79] and the Facial Nerve Grading System 2.0 [55]. For these measuring instruments, the total value to be achieved is divided into six groups or grades (degree I: normal function up to degree VI: total paralysis) [55,79]. This classification was also used in the present work. For this purpose, the maximum total score (100 points) to be achieved in the Sunnybrook Facial Grading System was divided by six and thus into six equally sized areas (100-84 points: normal function, no facial paresis; 83-67: light facial paresis; 66-50 moderate facial paresis; 49-33 medium facial paresis; 32-16 severe facial paresis; 15-0 complete facial paresis with respect to paralysis). Once the total score had been evaluated by the logopaedic examiner, the severity level could be determined. According to this definition, a facial paresis from grade II (≤83 points) could be presented. This, in turn, implied an admission of a natural portion of asymmetry in the face and is consistent with previous research [82].