Comprehensive Summary
This paper by Rezaei and colleagues propose a machine learning algorithm that differentiates the brain waves between people who have claustrophobic tendencies and healthy brain activity. Traditional records of claustrophobia such as self-reporting or interviewing may not fully encompass the brain’s interpretation of the condition, but capturing EEG data through activity related to fear and anxiety proves to be a better option. This study involves the use of traditional machine learning classifiers as well as newer deep learning classifiers to explore the different brain regions and frequency bands that may influence claustrophobic tendency. Nine individuals with self-identifying claustrophobia and thirteen healthy individuals participated in the study. Three rounds of EEG recording, which lasted 300 seconds each, were performed for each subject: R0, T1, and T2. In the R0 condition, the individual was placed in a large, well-lit laboratory. In the T1 and T2 conditions, the participant was placed in moderately-lit wooden chambers, with the T2 condition having smaller dimensions and no doors. The result was that the MLP and CNN-BiLSTM deep learning models performed the best in accuracy and sensitivity in classifying frequency bands as well as determining the correct brain region, although all models in the latter experiment lacked sensitivity. This study used leave-one-out cross-validation (LOOCV) for subject-independent classification, where one participant’s data was used to test the algorithm while the other twenty-one subjects’ data was used to train the model. This method was repeated for all twenty-two individuals. In this test, the CNN-BiLSTM surprisingly underperformed with an accuracy of 55% while the other classifiers all achieved roughly 73% accuracy, with slightly less accuracy regarding the experimental group. This is likely due to fewer participants with claustrophobia, highlighting the need for a larger sample size. EEG analysis showed that the most significant differences between individuals with and without claustrophobia lies in the frontal, temporal, and parietal regions of the brain as well as beta and theta frequency bands.
Outcomes and Implications
Using EEG provides an objective and noninvasive way to diagnose patients with claustrophobia. Deep learning algorithms may be used to help physicians analyze EEG data and detect patterns that may conform to either a healthy brain or waves in certain areas that may show signs of claustrophobic tendencies. However, it is important to note the sample size of this study was relatively small. Additionally, rather than relying on self-identification, factors such as heart rate can be incorporated to provide a more standardized depiction of claustrophobia. With this study serving as a foundation for future research, deep learning algorithms may be on track to diagnosing claustrophobia or other related anxiety disorders.