Abstract
Excessive fear in response to certain stimuli may be a key indicator of anxiety disorders. Its detection makes it valuable for the diagnosis and treatment of such pathologies. Quantum computing has shown promising results in processing different types of brain signals. However, its potential for functional near-infrared spectroscopy (fNIRS) signals remains largely unexplored. The present study investigates the application of parameterized quantum circuits (PQCs) for the detection of fear in fNIRS data. To this end, two different quantum architectures and quantum kernels are presented and tested on a publicly available fNIRS dataset. The proposed models are evaluated for subject-dependent and subject-independent classification by cross-validation to measure their performance under different conditions. The cross-validation results showed good performance of the proposed architectures even when trained on a very small dataset. Both analyzed quantum kernels showed high performance as feature extractors. Surprisingly, the subject-dependent approach achieved superior results despite using a training set more than 20 times smaller than that of the subject-independent approach. These results emphasize the power of quantum models in the classification of fNIRS signals and open new avenues for the analysis of this type of brain signals beyond the limitations of classical approaches.