EEG-Based Scalogram Imaging and Deep Learning for Insomnia Classification

Insomnia is one of the most widespread sleep disorders described by the difficulties with falling asleep or staying asleep. Ensuring fast, accurate, and efficient diagnosis and treatment of the disorder is often quite challenging. Classical methods of diagnostics are quite expensive and are often limited to subjective observation and polysomnography, which is also expensive, time-consuming, and ineffective for various complicated sleep disorders. To address these issues, the following work offers a novel approach based on the pictures of the quarried scalograms, which are obtained after training the deep learning model called DenseNet-121 and learning the electroencephalogram signals.Our work aims to overcome the limitations of common clinical approaches through novel non-invasive techniques enabled by recent progress in machine learning and neuroimaging technologies. Scalogram images derived from the analysis of EEG signals implicated in sleep, provides an essential combination of the temporal and spectral information to depict sleeping patterns. Finally, while using the cutting-edge deep learning architecture DenseNet-121, which has proven a high classification performance in images to distill the critical features of scalogram images created through processing previously obtained EEG signals to successfully detect and classify the presence of insomnia.With a dataset consisting of EEG signals captured at 512 Hz and converted into 1024 scalogram pictures, we were able to identify insomnia with an astonishing 99.51% classification accuracy through careful experimentation and analysis. With this high degree of accuracy, our approach, called ScaloNet-121, exhibits promise as a reliable and workable replacement for diagnosing sleep issues, providing a promising path for data-driven, non-invasive sleep problem diagnosis.