Polysomnography Dataset for Sleep Analysis in Indian Stroke Patients

1. Overview Sleep health is vital for cognitive function and overall well-being, especially in populations such as ischemic stroke patients, where sleep-disordered breathing is highly prevalent. To address the paucity of stroke-specific sleep data, particularly within the Indian context, we present the Polysomnography Dataset for Sleep Analysis in Indian Ischemic Stroke Patients (iSLEEPS). This dataset comprises 100 polysomnography (PSG) recordings from ischemic stroke patients, with detailed annotations of sleep parameters, as well as patient demographics, lesion location, and stroke characteristics. Adhering to ethical guidelines, the iSLEEPS is publicly accessible and designed for a variety of applications, ranging from training machine learning models to in-depth exploration of sleep disorders in stroke patients. We evaluated the dataset using state-of-the-art deep learning models, including Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and Transformer based architectures. The CNN-based model achieved an accuracy of 61.65%, the LSTM-based model 74.70%, and the Transformer-based model 67.44% in automated sleep stage classification. These results highlight the dataset’s utility in training machine learning models and its potential to enhance the sleep architecture in stroke patients. By offering this resource, we aim to bridge the gap in Indian-specific sleep disorder data and foster innovation in the field of sleep medicine. 2. Data Structure The dataset is organized into a main folder named Dataset, which contains five batches of subject data: 40 subjects [Avialable on Zenodo & iHUB-Data] All 100 subjects [Avialable on iHUB-Data] Each batch contains files corresponding to 20 subjects in the following formats: 2.1 Polysomnography Data (EDF Files) Each subject's PSG recording is stored in European Data Format (EDF), commonly used for physiological signals. File naming convention: SN<subject_number>.edf (e.g., SN1.edf, SN2.edf, …, SN100.edf). These files contain multi-channel electrophysiological signals recorded during sleep. 2.2 Sleep Annotations (XLSX Files) Each subject has an annotation file in Excel (.xlsx) format, containing detailed sleep-related information across multiple sheets. File naming convention: SN<subject_number>.xlsx (e.g., SN1.xlsx, SN2.xlsx, …, SN100.xlsx). Each annotation file consists of 17 sheets, detailing various sleep characteristics: Sleep Profile Automatic Arousals Sleep Staging Light Position … (12 more sheets with additional annotations related to sleep study parameters) 2.3 Subject Metadata (XLSX File) A separate file named subject_description.xlsx contains 63 columns detailing subject-specific demographic and clinical information for all 100 subjects. Key attributes include: Demographic information: Age, Sex, Occupation, BMI, etc. Clinical parameters: AHI index, Cardiovascular disease, Blood Sugar, etc. Other characteristics: risk factors, coronary artery disease, etc.