mindUP

Please view the video for a shortened explanation of the project.

Anxiety, a ubiquitous emotional state, accounts for a third of all youth mental health issues and has been identified as a major hurdle in education for students. BCI-based Neurofeedback Training (NFT) has emerged as an effective method to address developmental disorders such as ADHD and autism as well as psychiatric disorders such as anxiety and depression. Concomitantly, breath control has been shown to be an effective non-pharmacological intervention to alleviate symptoms of anxiety, particularly in students with anxiety. In addition, electroencephalography (EEG) studies have shown controlled breathing to have an impact on autonomic nervous system activity and oscillatory brain activity with links to improved cognitive performance during attentional and executive functions. With the promise of effectively downregulating anxiety and ameliorating academic performance in students with learning differences, BCI-based breathing entrainment protocols can be developed that adapt to the individual’s neurophysiological correlates in real-time. To that end, we recorded physiological and EEG data on LD students performing an unguided and a guided breathing exercise following an arithmetic mental task under time constraint to investigate the effects of each breathing method on subjective ratings of anxiety and identify corresponding neurophysiological correlates. Based on our findings, we built an LSTM recurrent neural net classifier that can identify predictive features of anxiety in EEG for prospective use in an individualized neuroadaptive system able to adapt to and modulate individual neurophysiological correlates of anxious states and provide optimized breathing intervention.

The recurrent neural net classifier with 2 LSTM layers was able to assess the level of anxiety of each participant, yielding an average accuracy of 86.7% on the test data. To ensure the validity of the model, we analyzed various metrics of its prediction characteristics including false positives and false negatives across all four classes. Our results are the first to show that subjective anxiety states can be reliably classified based on ongoing EEG, opening new avenues of decoding covert mental states for BCI-based neuroadaptive applications. As a next step, we plan on implementing a closed-loop BCI able to feedback the current anxiety level to a user and input this information into a feedback controller, which in response would adapt the breathing entrainment pace, ultimately finding the individualized optimum to effectively alleviate anxiety.

Built With

• keras
• mne
• python
• tensorflow