DCSM Sleep Staging Dataset

--| Overview This archive stores the Danish Center for Sleep Medicine (DCSM) sleep staging dataset. The DCSM dataset consists of 255 randomly selected and fully anonymized overnight lab-based PSG recordings from patients visiting the DCSM for the diagnosis of non-specific sleep related disorders. The DCSM dataset represents a diverse cohort of Danish patients with respect to demographic characteristics, diagnostic background and sleep/non-sleep related medication usage. The PSGs were collected between 2015-2018. EEG and EOG signals were recorded at 256 Hz and bandpass filtered to the interval 0.3 Hz - 70 Hz (3dB limits). Hypnograms were scored according to the AASM criteria. The intended purpose of the DCSM dataset is development and benchmarking of clinically applicable sleep staging algorithms. --| Publication This dataset was published in: U-Sleep: Resilient High-Frequency Sleep Staging Mathias Perslev, Sune Darkner, Lykke Kempfner, Miki Nikolic, Poul Jørgen Jennum and Christian Igel npj Digital Medicine, 2021 --| Dataset Structure Each subject in the dataset is represented by an anonymously named folder storing three data files, 'hypnogram.ids', 'psg.edf' and 'psg.h5', for example: /dcsm_dataset/ |- tp8cbd54f0_4751_4def_b11e_6e6353c69562 |----| hypnogram.ids |----| psg.edf |----| psg.h5 ---| Downloading the Data The dataset can be downloaded as a single zip file from this archive. Alternatively, the U-Time software package allows the entire or a subset of the dataset to be downloaded. Please see https://github.com/perslev/U-Time ---| Loading Data Data may be loaded manually (please refer to details below) or using the U-Time software package. Please see to https://github.com/perslev/U-Time. ---| H5 & EDF PSG File Details Each PSG file (psg.h5) is a HDF5 formatted file storing data from multiple modalities such as EEG, EOG and EMG as separate channel-wise entries. The archive stores each channel under a group named 'channels'. For example, to access data channel 'C3-M2', the directory '/channels/C3-M2' should be considered. Each archive also has the attributes 'date' and 'sample_rate': 'date' is always None (anonymized) and 'sample_rate' is always 256 (Hz). Each PSG recording is also supplied in EDF format. ---| Hypnogram File Details Each hypnogram file (hypnogram.ids) is a text file storing data in "init-duration-stage" format (ids). It contains three comma-separated columns of values: 1) The first column ('init') stores a time point at which a scored segment of sleep begins in seconds offset from the start of the PSG recording. 2) The second column ('duration') stores the duration of time in seconds that the scored segment of sleep lasts. 3) The third column ('stage') stores the clinician's sleep stage annotation for the segment, which is always one of ['Wake', 'N1', 'N2', 'N3', 'REM']. For example, if a subject was scored to be in sleep stage "Wake" for 120 seconds starting at second 60 in the file, the hypnogram file would contain the following line: "60,120,Wake".

--| 数据集概览
本归档文件存储丹麦睡眠医学中心（Danish Center for Sleep Medicine, DCSM）的睡眠分期数据集。该数据集包含255份经随机选取且完全匿名化的夜间实验室多导睡眠图（Polysomnography, PSG）记录，均来自因非特异性睡眠相关疾病就诊于丹麦睡眠医学中心的患者。该数据集涵盖了在人口统计学特征、诊断背景以及睡眠/非睡眠相关药物使用情况方面具有多样性的丹麦患者群体。上述多导睡眠图数据采集于2015年至2018年间。脑电图（Electroencephalogram, EEG）与眼电图（Electrooculogram, EOG）信号以256 Hz的采样率采集，并经过带通滤波，滤波频段为0.3 Hz至70 Hz（3dB截止点）。睡眠分期图谱按照美国睡眠医学会（American Academy of Sleep Medicine, AASM）标准进行标注。本数据集的设计用途为开发与基准测试临床可用的睡眠分期算法。
--| 相关出版物
本数据集已发表于：《U-Sleep：高鲁棒性高频睡眠分期算法》，作者为Mathias Perslev、Sune Darkner、Lykke Kempfner、Miki Nikolic、Poul Jørgen Jennum与Christian Igel，刊载于《npj Digital Medicine》，2021年。
--| 数据集结构
数据集中的每一位受试者对应一个匿名命名的文件夹，内含三份数据文件：`hypnogram.ids`、`psg.edf`与`psg.h5`。示例如下：
/dcsm_dataset/
└─ tp8cbd54f0_4751_4def_b11e_6e6353c69562
├─ hypnogram.ids
├─ psg.edf
└─ psg.h5
--| 数据下载
本数据集可作为单个zip压缩包从本归档文件下载。此外，通过U-Time软件包可下载完整数据集或其子集。详情请访问：https://github.com/perslev/U-Time
--| 数据加载
数据可手动加载（详见下文说明），或通过U-Time软件包加载。详情请访问：https://github.com/perslev/U-Time
--| H5与EDF格式PSG文件说明
每份PSG文件（`psg.h5`）均为HDF5格式文件，以独立的逐通道条目存储多模态数据，涵盖脑电图（Electroencephalogram, EEG）、眼电图（Electrooculogram, EOG）与肌电图（Electromyogram, EMG）等。归档文件将所有通道存储于名为`channels`的组下。例如，若要访问通道`C3-M2`的数据，可通过路径`/channels/C3-M2`进行读取。每份归档文件还包含`date`与`sample_rate`两个属性：`date`始终为`None`（已匿名化），`sample_rate`固定为256 Hz。所有PSG记录同时提供EDF格式版本。
--| 睡眠分期图谱文件说明
每份睡眠分期图谱文件（`hypnogram.ids`）为文本文件，采用「起始时间-持续时长-睡眠分期」（init-duration-stage, IDS）格式存储数据。该文件包含三列以逗号分隔的数值：
1. 第一列（`init`）：存储睡眠分期片段的起始时间点，以距PSG记录开始的秒数为单位；
2. 第二列（`duration`）：存储该睡眠分期片段的持续时长，单位为秒；
3. 第三列（`stage`）：存储临床医师对该片段标注的睡眠分期，其取值仅包含以下五种：「Wake（清醒）」、「N1期」、「N2期」、「N3期」、「REM（快速眼动睡眠）」。
例如，若某受试者在记录开始后第60秒至第180秒期间被标注为清醒状态，则睡眠分期图谱文件中将包含如下一行：`60,120,Wake`。

搜集汇总

背景与挑战

背景概述

DCSM睡眠分期数据集是一个包含255份匿名过夜PSG记录的临床数据集，用于睡眠分期算法的开发和基准测试。数据采集于2015-2018年，来自丹麦睡眠医学中心的患者，涵盖多样化的患者群体，信号采样率为256 Hz并遵循AASM评分标准。数据集以HDF5和EDF格式存储PSG信号，并以文本文件存储睡眠阶段注释，便于研究和算法验证。