Their emotional brain analysis focuses on the difference in implicit emotional reactions between Parkinson's patients, who are generally believed to suffer from impairments in recognizing emotions, and healthy individuals. The team demonstrated they can identify patients and healthy individuals with an F1 score of 0.97 or higher, based solely on brain scan readings of emotional responses. This diagnostic performance edges very close to 100% accuracy from brainwave data alone. The F1 score is a metric that combines precision and recall, where 1 is the best possible value.
The results show that Parkinson's patients displayed specific emotional perception patterns, comprehending emotional arousal better than emotional valence, which means they are more attuned to the intensity of emotions rather than the pleasantness or unpleasantness of those emotions. The patients were also found to struggle most with recognizing fear, disgust and surprise, or to confuse emotions of opposite valences, such as mistaking sadness for happiness.
The researchers recorded electroencephalography - or EEG - data, measuring electrical brain activity in 20 Parkinson's patients and 20 healthy controls. Participants watched video clips and images designed to trigger emotional responses. After the recording of EEG data, multiple EEG descriptors were processed to extract key features and these were transformed into visual representations, which were then analyzed using machine learning frameworks such as convolutional neural networks, for automatic detection of distinct patterns in how the patients processed emotions compared to the healthy group. This processing enabled the highly accurate differentiation between patients and healthy controls.
Key EEG descriptors used include spectral power vectors and common spatial patterns. Spectral power vectors capture the power distribution across various frequency bands, which are known to correlate with emotional states. Common spatial patterns enhance interclass discriminability by maximizing variance for one class while minimizing it for another, allowing for better classification of EEG signals.
As the researchers continue refining EEG-based techniques, emotional brain monitoring has the potential to become a widespread clinical tool for Parkinson's diagnosis. The study demonstrates the promise of combining neurotechnology, AI and affective computing to provide objective neurological health assessments.
Ravikiran Parameshwara, Soujanya Narayana, Murugappan Murugappan, Ibrahim Radwan, Roland Goecke, Ramanathan Subramanian.
Exploring Electroencephalography-Based Affective Analysis and Detection of Parkinson's Disease.
Intell Comput. 2024;3:0084. doi: 10.34133/icomputing.0084