Comprehensive Polysomnography (CPS) dataset

Comprehensive Polysomnography (CPS) dataset是由IT-Designers Gruppe Esslingen am Neckar等机构创建的，旨在用于机器学习在临床实践中进行唤醒检测的综合评估。该数据集包含多模态的睡眠多导图数据，涵盖脑电波、血氧水平、呼吸、眼动和腿部运动等多种生理参数。数据集的创建过程考虑了临床注释的约束，特别是唤醒事件的开始和结束注释。该数据集的应用领域主要集中在睡眠障碍诊断，特别是阻塞性睡眠呼吸暂停（OSA）的检测，旨在通过多模态数据提高唤醒检测的准确性，从而更好地支持临床决策。Comprehensive Polysomnography (CPS) dataset was created by institutions including IT-Designers Gruppe Esslingen am Neckar and others, aiming to enable comprehensive evaluations of machine learning-based arousal detection approaches in clinical practice. This dataset contains multi-modal polysomnography data covering various physiological parameters such as electroencephalogram (EEG), blood oxygen levels, respiration, eye movement, and leg movement. The development of the dataset takes into account the constraints of clinical annotations, particularly the annotations for the onset and termination of arousal events. Its main application fields focus on sleep disorder diagnosis, especially obstructive sleep apnea (OSA) detection, with the goal of improving the accuracy of arousal detection via multi-modal data to better support clinical decision-making.

搜集汇总

数据集介绍

构建方式

The Comprehensive Polysomnography (CPS) dataset was meticulously constructed to reflect clinical annotation constraints, particularly focusing on the onset of arousals as per clinical practice. This dataset includes modalities not present in existing polysomnographic datasets, such as pulse transit time and beat-by-beat blood pressure estimations. The data was collected during clinical practice from 2021-2022 in a state-of-the-art sleep laboratory at Klinikum Esslingen, Germany, and comprises 113 diagnostic polysomnographic recordings. The dataset was curated to align with the need for detecting arousal onsets, which is crucial for diagnosing sleep disorders, and was released alongside the paper to demonstrate the benefits of leveraging multimodal data for arousal onset detection.

特点

The CPS dataset is distinguished by its extensive channels and novel beat-by-beat blood pressure annotations, which are not commonly found in other polysomnographic datasets. It includes annotations indicating whether arousals were first detected in the EEG or as a consequence of other physiological changes, along with detailed medical outcomes such as sleep diagnoses, Baveno classification, and T90 value. This dataset is unique for its focus on aligning with clinical practices, particularly in annotating only the onset of arousals, which is essential for bridging the gap between technological advancements and clinical needs.

使用方法

The CPS dataset is designed to be used for training and evaluating machine learning models for arousal detection in clinical practice. Researchers can utilize this dataset to develop models that focus on detecting arousal onsets, which aligns with clinical needs. The dataset can be accessed on the PhysioNet platform under the PhysioNet Credentialed Health Data License 1.5.0. Users are encouraged to focus on detecting arousal onsets and to explore the impact of this shift on current training and evaluation schemes. The dataset is accompanied by detailed documentation and instructions on how to load and use the data, ensuring that researchers can effectively leverage its multimodal features for advancing arousal detection models.

背景与挑战

背景概述

The Comprehensive Polysomnography (CPS) dataset was created to address the critical misalignment between clinical protocols and machine learning (ML) methods in the detection of arousals during sleep, a key aspect in diagnosing sleep disorders. Developed by a collaborative team from IT-Designers Gruppe, Aalen University of Applied Sciences, Klinikum Esslingen, and Technical University of Munich, the dataset reflects clinical annotation constraints and includes modalities not present in existing polysomnographic datasets. Released in 2024, the CPS dataset aims to bridge the gap between technological advancements and clinical needs by providing a robust resource for ML-based arousal detection models, thereby enhancing the integration of ML in clinical settings.

当前挑战

The creation of the CPS dataset faced several challenges, primarily stemming from the divergence between clinical annotation practices and ML requirements. Clinicians typically annotate only the onset of arousals, while ML methods necessitate annotations for both the beginning and end of events. Additionally, the lack of standardized evaluation methodologies tailored to clinical needs posed significant hurdles. The dataset also had to address the diversity of equipment, software, and protocols across different laboratories, which complicates the development of universally applicable ML models. Furthermore, the absence of large-scale time series datasets and clear evaluation metrics, along with limited consensus on theoretical and practical understanding of time series, further impeded progress. The CPS dataset addresses these challenges by introducing a novel post-processing and evaluation framework, ALPEC, which emphasizes approximate localization and precise event count of arousals, aligning with clinical practice.

常用场景

经典使用场景

在临床实践中，综合多导睡眠图（CPS）数据集被广泛用于机器学习驱动的觉醒检测。该数据集通过整合多模态数据，如脑电图、血氧水平、呼吸和眼动等，为研究人员提供了一个全面的睡眠生理参数记录。经典的使用场景包括开发和评估基于机器学习的觉醒检测模型，这些模型旨在识别睡眠中的觉醒事件，这对于诊断睡眠障碍如阻塞性睡眠呼吸暂停（OSA）至关重要。

解决学术问题

CPS数据集解决了当前机器学习方法在临床实践中应用时面临的关键问题，特别是临床协议与机器学习方法之间的不匹配。传统上，临床医生仅标注觉醒的开始，而机器学习方法依赖于觉醒的开始和结束的标注。此外，缺乏针对临床需求的觉醒检测模型的标准化评估方法。CPS数据集通过引入新的后处理和评估框架（ALPEC），强调近似定位和精确事件计数，为这些问题提供了有效的解决方案，从而促进了机器学习技术在临床环境中的整合。

衍生相关工作

CPS数据集的发布和ALPEC框架的提出，激发了大量相关研究工作。例如，基于该数据集的研究已经开发出多种先进的觉醒检测模型，这些模型利用了数据集中的多模态数据。此外，ALPEC框架本身也成为了评估时间序列数据中异常检测模型性能的标准方法之一。其他相关工作还包括对睡眠阶段分类、睡眠质量评估以及个性化睡眠治疗方案的研究。