Indexing metadata

BACKGROUND: The analysis of sleep patterns in patients with chronic disorders of consciousness is attracting increasing attention in terms of diagnosis, prognosis, and treatment of severe brain damage. The study describes a software package based on an artificial intelligence (AI) expert system designed to classify the phases and stages of sleep, taking into account the characteristics of impaired cortical rhythm in such patients.

AIM: To develop a specialized AI-based software package focused on patients with chronic impairment of consciousness for automatic classification of sleep phases and stages, with an emphasis on identifying sleep spindles and non-rapid eye movement (REM) and REM sleep phases.

MATERIALS AND METHODS: To ensure the correct operation of the software package, receiver operating characteristic (ROC) curves were analyzed considering the binary classification of slow sleep, REM sleep, and wakefulness.

RESULTS: The average sensitivity and specificity of the algorithm were 87.9 and 70.1, respectively. The average area under the ROC curve was 0.790. The algorithm for determining the REM phase demonstrates low specificity with high sensitivity, and its graph was similar to that of wakefulness, as well as the irregularity of the presence of REMs in the REM sleep phase in patients with CNS and the frequent presence of nystagmus in the waking state. Information about the presence of nystagmus, entered at the start of the program, allowed us to slightly increase the efficiency of the algorithm; however, this aspect probably needs further improvement.

CONCLUSION: A software package that takes into account the features of electroencephalography of patients with chronic disorders of consciousness and analyzes sleep and wakefulness automatically is not only useful as a diagnostic tool for neurologists and somnologists but also contributes to a wider dissemination of this technique in clinical practice.