科研 Skill

睡眠分期自动评估

neuro-sleep-staging-eval

Builds and evaluates automatic sleep staging from EEG, EOG and EMG against AASM rules and clinical summary metrics. Use when scoring sleep stages or reporting staging accuracy.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Sleep staging evaluation

Staging is scored in 30 s epochs under AASM rules (W, N1, N2, N3, REM). Epochs from one night are strongly dependent and stage proportions are very unequal, so evaluation choices change the headline number a lot.

1. Prepare labels and data

  • Map legacy R&K labels to AASM (S3 and S4 to N3; movement and unscored epochs excluded, and say how many).
  • Trim long wake before lights-off and after lights-on (or use the lights markers). Untrimmed wake inflates accuracy because wake is easy.
  • Keep channel choice consistent with the claim (single-channel EEG, EEG+EOG, full PSG, wearable).

2. Split by subject

Subject-wise (or night-wise with all nights of a person on one side). For public datasets use the published folds when comparing to papers.

3. Report the right metrics

  • Overall accuracy and Cohen's kappa (the field's standard), plus macro-F1.
  • Per-stage F1 and the confusion matrix: N1 is always worst; say so.
  • Per-subject distribution, not only pooled: show median and range.
  • Human context: inter-scorer agreement on PSG is typically around kappa 0.75 to 0.8, and lower for N1; a model at that level is near the ceiling the labels allow.

4. Check clinically used summary metrics

Compare predicted vs scored per night: total sleep time, sleep efficiency, sleep onset latency, WASO, % N3, % REM, REM latency. Report bias and limits of agreement (Bland-Altman). A model can have good kappa and still bias REM latency.

5. Generalisation

Test on another dataset or device before claiming robustness; cohort age and disorders change stage distributions (older adults have less N3).

Report

Dataset and cohort, channels, label mapping, trimming rule, split, kappa, accuracy, macro-F1, per-stage F1, confusion matrix, per-subject spread, summary-metric agreement, and external validation if any.

示例提示词

帮我用Sleep-EDF数据训练睡眠分期模型并和人工评分比较一致性。

所属插件

神经科学面向神经科学研究:EEG/MEG、fMRI、BIDS、spike sorting、连接组、脑机接口解码评估、神经刺激安全与公开神经数据检索。

安装该插件会一并获得同领域的 45 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

睡眠分期自动评估能做什么?

Builds and evaluates automatic sleep staging from EEG, EOG and EMG against AASM rules and clinical summary metrics. Use when scoring sleep stages or reporting staging accuracy。

如何安装睡眠分期自动评估?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「睡眠分期自动评估」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

睡眠分期自动评估需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

睡眠分期自动评估经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。