Assessing the effectiveness of online emotion recognition training in healthy volunteers

Background Difficulties in facial emotion recognition are associated with a range of mental health and neurodevelopmental conditions and can negatively impact longer term social functioning. Interventions that target facial emotion recognition may therefore have important clinical potential, for example for autistic individuals. We investigated the effect of an emotion recognition training (ERT) task on emotion recognition ability and, importantly, whether training generalises to novel (non-trained) faces.

Methods We conducted three online experimental studies with healthy volunteers completing a single ERT session to test: 1) the efficacy of our four-emotion ERT (training to improve recognition of angry, happy, sad and scared emotional expressions) (N=101), 2) the efficacy of our six-emotion ERT (adding disgusted and surprised) (N=109), and 3) the generalisability of ERT to novel (non-trained) facial stimuli (N=120). In all three studies, our primary outcome was total correct hits across all emotions. In Studies 1 and 2, this was compared across active training and control (sham) training groups (randomised). In Study 3, this was compared across groups who were trained on stimuli that were either the same identity (stimulus-congruent), or a different identity (stimulus-incongruent) to those they were tested on (randomised). Linear mixed effects models were used to test for effects of training.

Results The effect estimate from Study 1 was in the direction of improvement in the active training group, however, confidence intervals were wide (b=0.02, 95% CI=-0.02 to 0.07, p=0.27) and our effect may have been reduced due to ceiling effects. Study 2, with the additional emotions, indicated total hits were greater following active (vs. sham) training, which remained following inclusion of baseline covariates (b=0.07, 95% CI=0.03 to 0.12, p=0.002). Study 3 demonstrated that improvement post-training was similar across stimulus-congruent and incongruent groups (b=-0.01, 95% CI=-0.05 to 0.02, p = 0.52).

Conclusion Our results indicate that ERT improves emotion recognition and that this improvement generalises to novel stimuli. Our data suggest six emotions should be used rather than four to avoid ceiling effects in training. Future studies should explore generalisability of facial stimuli of different ages and ethnicities as well as examining longer-term effects of ERT. The application of ERT as an intervention may be particularly beneficial to populations with known emotion recognition difficulties, such as autistic individuals.