交互式方法探索

在七个协议上探索解码方法如何分化

这里没有单一总分。通过去除整体水平后的二维方法轮廓图、任意双协议散点图、七协议一致性矩阵与两两配对对比器,直接观察方法在不同 EEG 范式间的行为模式与不确定性。

视图一 · 去除整体水平的二维轮廓映射

二维方法轮廓图

先将 7 个准确率协议分别标准化,再减去每个方法自身的七协议均值,从而彻底去除整体高低,只保留方法在不同任务间的相对行为形状。距离越近,表示跨范式的相对强弱模式越相似。

← 相对偏向 睡眠分期 / 心算负荷相对偏向 运动想象 / BETA 8通道 →↑ 相对偏向 BETA 4通道 / BETA 8通道↓ 相对偏向 语义 ERP / 心算负荷CBraModCodeBraineeg-fm-maskingEEGMambaEEGPTERP-FM BaseLaBraMAtom-LMolecule-LQuark-BQuark-GQuark-HQuark-LLUNA BaseLUNA LargeREVE BaseREVE LargeREVE SmallSingLEMST-EEGFormer BST-EEGFormer LZUNA 1.1Deep4NetEEGNetShallowFBCSPNetLog band-power ridgeSpectral ridge
实心点:跨协议轮廓清晰空心点:轮廓较平,位置参考意义较弱外圈弧度:二维保留的 7 维近邻比例
完整覆盖方法
27
另 9 个部分覆盖
二维保留方差
69.5%
轴 1:45.1% · 轴 2:24.4%
距离相关 r
0.92
2D 与真实 7D 轮廓距离
3-近邻保留率
53.1%
200 次区间重采样

轮廓较平的方法(空心点): 以下方法的跨协议相对波动小于其 95% 区间重采样标准差的 1.5 倍,因此在二维轮廓图上的具体落点参考意义较弱: ERP-FM Base、LaBraM at Scale Molecule-L、LaBraM at Scale Quark-L、REVE Large.

部分覆盖方法(未纳入七协议轮廓图,共 9 个): 本图不做缺失值填补。下列仅在部分协议上评测的方法保留在下方的双协议散点图与两两协议对比器中: BrainOmni Base (5/7)、CSP+LDA (5/7)、FBCCA (2/7)、MDM (2/7)、Standard CCA (2/7)、Temporal ridge (2/7)、TS+LR (2/7)、xDAWN+LDA (2/7)、XDAWNCov+TS+LR (2/7).

查看完整轮廓坐标与七协议原始分数表(27 个完整方法 + 9 个部分覆盖方法)
方法类别轴 1轴 23-近邻保留轮廓状态运动想象BETA 8通道BETA 4通道心算负荷P300 ERP语义 ERP睡眠分期
BrainOmni Base基础模型———5/758.8%57.0–60.8%21.7%18.6–25.0%13.0%10.6–15.6%61.6%58.4–64.9%——67.7%65.8–69.5%
CBraMod基础模型−0.15+0.051/3清晰62.9%59.6–66.5%33.7%29.8–37.6%27.6%24.1–31.3%62.3%58.1–66.5%55.0%53.3–56.7%55.8%54.2–57.4%71.3%68.7–73.7%
CodeBrain基础模型−0.96−0.781/3清晰58.8%54.5–62.6%11.9%10.5–13.3%11.2%9.8–12.7%63.0%59.5–66.5%50.8%49.3–52.2%54.7%53.0–56.6%70.7%68.8–72.7%
eeg-fm-masking MAE (r=9 cm, L=2)基础模型+1.35+1.702/3清晰68.1%64.3–72.3%54.8%49.4–60.0%47.1%42.1–52.1%63.6%59.1–68.1%49.5%48.1–51.0%54.8%52.7–56.9%67.9%65.4–70.4%
EEGMamba基础模型+1.02+0.882/3清晰67.3%63.6–70.6%40.2%35.7–44.7%32.7%28.9–36.7%60.3%57.0–63.4%52.5%51.4–53.5%54.4%52.5–56.3%64.1%61.6–66.4%
EEGPT基础模型−1.54−0.451/3清晰58.3%54.9–61.8%20.3%18.2–22.6%17.3%15.5–19.2%65.7%60.9–70.3%54.0%51.5–56.6%54.6%53.3–55.9%75.5%73.9–77.1%
ERP-FM Base基础模型−1.02+0.371/3较平54.8%51.8–57.8%18.4%16.3–20.5%20.9%18.4–23.5%61.1%57.3–65.1%51.2%49.3–53.2%53.3%51.7–54.8%71.0%69.5–72.6%
LaBraM基础模型−1.49−0.251/3清晰53.5%51.4–55.4%10.8%9.5–12.2%12.9%11.2–14.6%64.6%60.5–68.6%49.4%47.7–51.2%53.2%51.3–54.9%71.1%69.0–73.2%
LaBraM at Scale Atom-L基础模型−1.39+0.622/3清晰56.9%53.3–61.0%25.8%22.4–29.4%28.0%24.3–31.8%67.5%63.7–71.3%53.2%51.5–54.9%52.8%51.3–54.3%72.7%70.6–74.4%
LaBraM at Scale Molecule-L基础模型−0.52+0.321/3较平59.6%57.2–61.4%26.6%22.9–30.5%28.9%25.1–33.0%63.9%59.9–68.0%51.4%49.2–53.5%54.7%52.9–56.5%71.9%70.2–73.4%
LaBraM at Scale Quark-B基础模型−0.31−0.021/3清晰61.9%59.1–64.4%27.4%23.6–31.5%27.8%24.0–31.9%66.0%62.0–70.0%49.6%48.5–50.6%55.4%53.1–57.8%72.1%70.3–74.0%
LaBraM at Scale Quark-G基础模型−0.21+1.241/3清晰58.7%57.3–60.1%25.0%21.3–28.8%26.8%22.8–31.1%59.4%56.9–61.7%50.3%48.5–52.1%51.8%50.1–53.4%67.3%65.6–68.9%
LaBraM at Scale Quark-H基础模型−1.06+0.812/3清晰60.2%57.4–62.9%23.7%20.2–27.3%26.2%22.3–30.3%64.5%60.6–68.3%52.6%51.1–54.1%52.2%50.3–53.9%72.9%71.5–74.2%
LaBraM at Scale Quark-L基础模型−0.37+0.252/3较平59.6%56.3–62.9%27.1%23.3–31.0%28.0%24.0–32.2%63.8%59.8–67.8%51.0%49.4–52.6%54.8%53.0–56.6%70.3%68.0–72.5%
LUNA Base基础模型−0.69−1.142/3清晰52.6%50.2–54.9%8.6%7.7–9.5%9.0%8.1–10.1%60.7%56.9–64.4%48.7%46.3–51.1%56.2%54.2–58.2%67.5%65.3–69.5%
LUNA Large基础模型−0.86−0.973/3清晰49.9%46.6–53.5%10.0%8.9–11.1%11.2%9.8–12.6%57.4%54.3–60.6%48.8%46.9–50.8%57.1%55.3–58.9%72.4%70.4–74.0%
REVE Base基础模型−0.54+0.431/3清晰66.2%61.8–70.6%37.5%33.3–41.9%25.5%22.4–28.8%68.8%64.1–73.2%50.5%48.5–52.3%53.7%51.8–55.4%74.6%72.9–76.4%
REVE Large基础模型+0.55−0.051/3较平70.2%64.0–76.1%52.7%47.7–57.8%46.4%41.7–51.2%69.0%63.3–74.7%54.6%52.7–56.5%59.7%57.5–61.8%76.6%75.2–78.0%
REVE Small基础模型+0.39+0.122/3清晰61.5%58.3–64.8%39.3%35.2–43.6%35.9%32.0–40.0%65.2%60.5–70.0%54.3%52.9–55.9%56.7%54.5–59.0%64.5%62.5–66.6%
SingLEM基础模型+4.15−1.142/3清晰72.0%66.6–76.6%26.6%23.7–29.5%20.5%17.8–23.3%51.5%49.2–53.8%51.2%49.2–53.1%58.5%56.1–61.1%35.3%33.7–37.0%
ST-EEGFormer Base基础模型−1.58−0.442/3清晰58.1%53.7–63.3%17.1%15.5–18.8%17.5%15.8–19.3%64.5%59.0–70.0%53.6%52.3–54.8%54.6%52.9–56.4%77.3%75.6–78.8%
ST-EEGFormer Large基础模型−0.81−0.912/3清晰62.4%57.4–67.8%22.6%20.1–25.3%22.4%19.9–25.0%63.4%57.8–69.0%52.7%51.3–54.2%57.8%56.0–59.6%79.6%78.2–81.0%
ZUNA 1.1基础模型−0.17−0.171/3清晰50.7%49.2–52.1%11.9%10.8–13.0%8.4%7.5–9.4%53.3%49.5–57.1%51.6%50.3–53.0%53.9%52.2–55.6%60.1%57.8–62.3%
Deep4Net轻量神经网络+3.20+0.352/3清晰68.6%63.3–74.0%52.8%47.3–58.2%38.7%33.7–43.8%61.3%57.5–65.2%50.8%50.0–51.6%57.4%55.4–59.5%45.2%42.7–47.7%
EEGNet轻量神经网络+2.50−0.781/3清晰71.0%65.0–76.3%55.8%50.1–61.3%44.1%38.6–49.6%67.6%62.8–72.5%53.3%51.7–54.9%61.4%58.9–64.0%56.5%54.9–58.3%
ShallowFBCSPNet轻量神经网络+1.08−1.451/3清晰68.8%63.4–73.9%23.0%20.3–25.9%16.2%14.1–18.6%67.9%63.4–72.4%49.7%49.0–50.4%56.7%54.7–58.8%55.0%53.1–56.8%
CSP+LDA经典方法———5/747.1%43.8–50.2%——68.3%62.0–74.3%50.0%49.1–50.8%52.7%51.4–54.0%50.1%46.9–53.1%
FBCCA经典方法———2/7—79.1%75.1–82.8%72.3%67.5–76.9%————
Log band-power ridge经典方法−0.86−1.732/3清晰57.3%54.4–60.3%5.3%4.8–5.8%5.3%4.8–5.8%65.0%60.2–69.7%50.4%49.3–51.6%56.1%54.2–58.0%66.0%63.0–68.9%
MDM经典方法———2/753.8%50.8–57.9%——71.5%65.5–77.8%———
Spectral ridge经典方法+0.12+3.343/3清晰54.9%53.2–56.7%50.7%45.6–55.8%48.2%43.1–53.3%56.8%54.6–58.9%51.0%49.6–52.4%51.0%49.6–52.5%72.2%70.7–73.8%
Standard CCA经典方法———2/7—63.1%57.2–69.0%57.6%51.8–63.5%————
Temporal ridge经典方法———2/7————53.8%51.9–55.6%56.3%54.4–58.4%—
TS+LR经典方法———2/758.9%54.9–63.9%——65.5%59.6–71.8%———
xDAWN+LDA经典方法———2/7————60.4%57.7–63.2%56.8%55.1–58.6%—
XDAWNCov+TS+LR经典方法———2/7————61.9%58.8–65.0%60.3%58.2–62.6%—
查看 7 维轮廓距离最近与最远的方法对(各 10 对)
分组方法 A方法 B7 维轮廓距离2 维投影距离
最接近轮廓LaBraM at Scale Molecule-LLaBraM at Scale Quark-L0.240.17
最接近轮廓EEGPTST-EEGFormer Base0.420.04
最接近轮廓ERP-FM BaseLaBraM at Scale Molecule-L0.680.50
最接近轮廓ERP-FM BaseLaBraM at Scale Quark-L0.900.66
最接近轮廓LaBraM at Scale Atom-LLaBraM at Scale Quark-H0.900.38
最接近轮廓CodeBrainST-EEGFormer Large0.920.20
最接近轮廓REVE LargeREVE Small1.000.24
最接近轮廓LaBraM at Scale Quark-BLaBraM at Scale Quark-L1.030.28
最接近轮廓LUNA BaseLUNA Large1.080.24
最接近轮廓CodeBrainLog band-power ridge1.080.96
差异最大轮廓SingLEMSpectral ridge6.246.02
差异最大轮廓LaBraM at Scale Atom-LSingLEM6.175.81
差异最大轮廓LaBraMSingLEM6.155.71
差异最大轮廓EEGPTSingLEM5.875.74
差异最大轮廓SingLEMST-EEGFormer Base5.875.78
差异最大轮廓LaBraM at Scale Quark-HSingLEM5.755.56
差异最大轮廓REVE BaseSingLEM5.704.95
差异最大轮廓LUNA LargeSingLEM5.595.01
差异最大轮廓ShallowFBCSPNetSpectral ridge5.534.88
差异最大轮廓ERP-FM BaseSingLEM5.525.39

视图二 · 任意双协议十字误差棒

自由双协议散点图

任选两个准确率协议作为横纵坐标,每个点同时画出横轴与纵轴的 95% 区间以及蓝灰色随机水平参考线。完整覆盖与部分覆盖方法只要在两个协议上都有发布数据,均会显示。

斯皮尔曼相关 ρ = −0.15 · 共同方法 n = 29

0%0%20%20%40%40%60%60%80%80%100%100%运动想象与静息 (平衡准确率 %)头皮睡眠分期 (%)随机 50.0%随机 20.0%
查看当前散点数据表:运动想象 × 睡眠分期(共 29 个方法)
方法类别运动想象与静息头皮睡眠分期
BrainOmni Base基础模型58.8% (57.0–60.8%)67.7% (65.8–69.5%)
CBraMod基础模型62.9% (59.6–66.5%)71.3% (68.7–73.7%)
CodeBrain基础模型58.8% (54.5–62.6%)70.7% (68.8–72.7%)
eeg-fm-masking MAE (r=9 cm, L=2)基础模型68.1% (64.3–72.3%)67.9% (65.4–70.4%)
EEGMamba基础模型67.3% (63.6–70.6%)64.1% (61.6–66.4%)
EEGPT基础模型58.3% (54.9–61.8%)75.5% (73.9–77.1%)
ERP-FM Base基础模型54.8% (51.8–57.8%)71.0% (69.5–72.6%)
LaBraM基础模型53.5% (51.4–55.4%)71.1% (69.0–73.2%)
LaBraM at Scale Atom-L基础模型56.9% (53.3–61.0%)72.7% (70.6–74.4%)
LaBraM at Scale Molecule-L基础模型59.6% (57.2–61.4%)71.9% (70.2–73.4%)
LaBraM at Scale Quark-B基础模型61.9% (59.1–64.4%)72.1% (70.3–74.0%)
LaBraM at Scale Quark-G基础模型58.7% (57.3–60.1%)67.3% (65.6–68.9%)
LaBraM at Scale Quark-H基础模型60.2% (57.4–62.9%)72.9% (71.5–74.2%)
LaBraM at Scale Quark-L基础模型59.6% (56.3–62.9%)70.3% (68.0–72.5%)
LUNA Base基础模型52.6% (50.2–54.9%)67.5% (65.3–69.5%)
LUNA Large基础模型49.9% (46.6–53.5%)72.4% (70.4–74.0%)
REVE Base基础模型66.2% (61.8–70.6%)74.6% (72.9–76.4%)
REVE Large基础模型70.2% (64.0–76.1%)76.6% (75.2–78.0%)
REVE Small基础模型61.5% (58.3–64.8%)64.5% (62.5–66.6%)
SingLEM基础模型72.0% (66.6–76.6%)35.3% (33.7–37.0%)
ST-EEGFormer Base基础模型58.1% (53.7–63.3%)77.3% (75.6–78.8%)
ST-EEGFormer Large基础模型62.4% (57.4–67.8%)79.6% (78.2–81.0%)
ZUNA 1.1基础模型50.7% (49.2–52.1%)60.1% (57.8–62.3%)
Deep4Net轻量神经网络68.6% (63.3–74.0%)45.2% (42.7–47.7%)
EEGNet轻量神经网络71.0% (65.0–76.3%)56.5% (54.9–58.3%)
ShallowFBCSPNet轻量神经网络68.8% (63.4–73.9%)55.0% (53.1–56.8%)
CSP+LDA经典方法47.1% (43.8–50.2%)50.1% (46.9–53.1%)
Log band-power ridge经典方法57.3% (54.4–60.3%)66.0% (63.0–68.9%)
Spectral ridge经典方法54.9% (53.2–56.7%)72.2% (70.7–73.8%)

视图三 · 为何不设单一综合分

不同协议是否偏好相同的方法?

下表展示每对准确率协议在共同评测方法上的斯皮尔曼相关系数 ρ 与分歧度(1 − ρ)。任意两项不同协议之间的中位相关为 +0.21,范围从 语义 ERP 与 睡眠分期 的 −0.19 到 BETA 8通道 与 BETA 4通道 的 +0.95。点击任意非对角单元格即可在上方散点图中打开对应协议对。

每个单元格显示斯皮尔曼相关 ρ、分歧度 1 − ρ 及共同方法数 n。
协议运动想象BETA 8通道BETA 4通道心算负荷P300 ERP语义 ERP睡眠分期
运动想象
1.00n=31
BETA 8通道
1.00n=30
BETA 4通道
1.00n=30
心算负荷
1.00n=31
P300 ERP
1.00n=31
语义 ERP
1.00n=31
睡眠分期
1.00n=29

两两协议对比

选择任意两个方法逐协议对比

查看同一协议下两个方法的点估计与 95% 区间;仅当数据已发布配对差值区间时才显示差值检验。下拉菜单按方法族与字母顺序排列,不按分数排序。

0%25%50%75%100%随机水平 50.0%ShallowFBCSPNet68.8% [63.4–73.9]EEGNet71.0% [65.0–76.3]
已发布的配对差值ShallowFBCSPNet − EEGNet: −2.2 pp[−8.0, +3.4 pp]

95% 区间,未校正 · n = 10 名被试 · 配对区间包含零

对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差

默认展示(运动想象与静息):ShallowFBCSPNet 平衡准确率 68.8%(95% 区间 63.4–73.9%);EEGNet 71.0%(95% 区间 65.0–76.3%);随机水平 50.0%。

查看已发布的配对差值表(共 91 对)
协议对比(左减右)配对均值差主区间被试数区间口径与说明
SSVEP · 8 通道FBCCA − Standard CCA+16.0 pp[+11.6, +20.5 pp]7098.75% 区间,按四项对比做 Bonferroni 校正
SSVEP · 8 通道FBCCA − EEGNet+23.3 pp[+17.6, +29.6 pp]7098.75% 区间,按四项对比做 Bonferroni 校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
SSVEP · 4 通道FBCCA − Standard CCA+14.7 pp[+10.7, +19.1 pp]7098.75% 区间,按四项对比做 Bonferroni 校正
SSVEP · 4 通道FBCCA − EEGNet+28.2 pp[+22.8, +33.9 pp]7098.75% 区间,按四项对比做 Bonferroni 校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
运动想象与静息TS+LR − Spectral ridge+4.0 pp[−1.1, +11.5 pp]1098.75% 区间,按四项对比做 Bonferroni 校正
运动想象与静息TS+LR − EEGNet−12.1 pp[−18.1, −5.4 pp]10名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
运动想象与静息MDM − EEGNet−17.2 pp[−23.7, −10.6 pp]10名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
运动想象与静息MDM − Spectral ridge−1.1 pp[−4.7, +3.8 pp]10名义 95% 区间
运动想象与静息TS+LR − CSP+LDA+11.8 pp[+7.9, +16.8 pp]10名义 95% 区间
运动想象与静息MDM − CSP+LDA+6.8 pp[+3.3, +10.8 pp]10名义 95% 区间
运动想象与静息TS+LR − LaBraM+5.4 pp[+0.3, +11.3 pp]10名义 95% 区间
运动想象与静息TS+LR − CBraMod−4.0 pp[−8.4, +0.9 pp]10名义 95% 区间
运动想象与静息TS+LR − MDM+5.1 pp[+2.3, +7.8 pp]10名义 95% 区间
心算与静息TS+LR − Spectral ridge+8.8 pp[+1.6, +16.1 pp]3698.75% 区间,按四项对比做 Bonferroni 校正
心算与静息TS+LR − EEGNet−2.1 pp[−8.0, +3.8 pp]36名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
心算与静息MDM − EEGNet+3.8 pp[−1.3, +9.3 pp]36名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
心算与静息MDM − Spectral ridge+14.7 pp[+8.9, +20.5 pp]36名义 95% 区间
心算与静息TS+LR − LaBraM+0.9 pp[−4.6, +6.6 pp]36名义 95% 区间
心算与静息TS+LR − CBraMod+3.2 pp[−1.5, +8.0 pp]36名义 95% 区间
心算与静息TS+LR − MDM−6.0 pp[−11.6, −0.5 pp]36名义 95% 区间
P300 目标 ERPXDAWNCov+TS+LR − Spectral ridge+11.0 pp[+6.8, +15.1 pp]2198.75% 区间,按四项对比做 Bonferroni 校正
P300 目标 ERPXDAWNCov+TS+LR − EEGNet+8.6 pp[+5.2, +12.0 pp]21名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
P300 目标 ERPxDAWN+LDA − EEGNet+7.1 pp[+4.3, +10.0 pp]21名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
P300 目标 ERPxDAWN+LDA − Spectral ridge+9.5 pp[+6.5, +12.4 pp]21名义 95% 区间
P300 目标 ERPXDAWNCov+TS+LR − Temporal ridge+8.1 pp[+5.3, +11.1 pp]21名义 95% 区间
P300 目标 ERPxDAWN+LDA − Temporal ridge+6.6 pp[+3.7, +9.8 pp]21名义 95% 区间
P300 目标 ERPXDAWNCov+TS+LR − LaBraM+12.5 pp[+8.7, +16.3 pp]21名义 95% 区间
P300 目标 ERPXDAWNCov+TS+LR − CBraMod+6.9 pp[+4.4, +9.6 pp]21名义 95% 区间
P300 目标 ERPXDAWNCov+TS+LR − xDAWN+LDA+1.5 pp[−0.9, +3.7 pp]21名义 95% 区间
语义目标 ERPXDAWNCov+TS+LR − Spectral ridge+9.3 pp[+6.1, +12.8 pp]3098.75% 区间,按四项对比做 Bonferroni 校正
语义目标 ERPXDAWNCov+TS+LR − EEGNet−1.2 pp[−3.0, +0.6 pp]30名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
语义目标 ERPxDAWN+LDA − EEGNet−4.6 pp[−6.5, −2.7 pp]30名义 95% 区间对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
语义目标 ERPxDAWN+LDA − Spectral ridge+5.8 pp[+4.0, +7.8 pp]30名义 95% 区间
语义目标 ERPXDAWNCov+TS+LR − Temporal ridge+3.9 pp[+2.0, +5.9 pp]30名义 95% 区间
语义目标 ERPxDAWN+LDA − Temporal ridge+0.5 pp[−1.2, +2.1 pp]30名义 95% 区间
语义目标 ERPXDAWNCov+TS+LR − LaBraM+7.1 pp[+4.8, +9.4 pp]30名义 95% 区间
语义目标 ERPXDAWNCov+TS+LR − CBraMod+4.5 pp[+2.5, +6.4 pp]30名义 95% 区间
语义目标 ERPXDAWNCov+TS+LR − xDAWN+LDA+3.5 pp[+1.8, +5.3 pp]30名义 95% 区间
心算与静息CSP+LDA − EEGNet+0.7 pp[−4.3, +5.8 pp]3695% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
心算与静息Deep4Net − EEGNet−6.3 pp[−10.4, −2.3 pp]3695% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
心算与静息ShallowFBCSPNet − EEGNet+0.3 pp[−2.6, +3.3 pp]3695% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
心算与静息ShallowFBCSPNet − Deep4Net+6.6 pp[+2.6, +10.6 pp]3695% 区间
SSVEP · 4 通道Deep4Net − EEGNet−5.5 pp[−6.8, −4.1 pp]7095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
SSVEP · 4 通道ShallowFBCSPNet − EEGNet−27.9 pp[−31.9, −23.9 pp]7095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
SSVEP · 4 通道ShallowFBCSPNet − Deep4Net−22.4 pp[−26.1, −18.9 pp]7095% 区间
SSVEP · 8 通道Deep4Net − EEGNet−3.0 pp[−3.9, −2.1 pp]7095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
SSVEP · 8 通道ShallowFBCSPNet − EEGNet−32.7 pp[−37.0, −28.5 pp]7095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
SSVEP · 8 通道ShallowFBCSPNet − Deep4Net−29.7 pp[−33.7, −25.8 pp]7095% 区间
运动想象与静息Deep4Net − EEGNet−2.4 pp[−6.8, +2.3 pp]1095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
运动想象与静息ShallowFBCSPNet − EEGNet−2.2 pp[−8.0, +3.4 pp]1095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
运动想象与静息ShallowFBCSPNet − Deep4Net+0.3 pp[−6.8, +7.7 pp]1095% 区间
P300 目标 ERPCSP+LDA − EEGNet−3.3 pp[−5.4, −1.6 pp]2195% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
P300 目标 ERPDeep4Net − EEGNet−2.5 pp[−4.2, −0.9 pp]2195% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
P300 目标 ERPShallowFBCSPNet − EEGNet−3.6 pp[−5.4, −1.9 pp]2195% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
P300 目标 ERPShallowFBCSPNet − Deep4Net−1.1 pp[−2.1, −0.2 pp]2195% 区间
语义目标 ERPCSP+LDA − EEGNet−8.8 pp[−11.1, −6.4 pp]3095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
语义目标 ERPDeep4Net − EEGNet−4.0 pp[−5.8, −2.1 pp]3095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
语义目标 ERPShallowFBCSPNet − EEGNet−4.7 pp[−7.2, −2.2 pp]3095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
语义目标 ERPShallowFBCSPNet − Deep4Net−0.7 pp[−3.0, +1.5 pp]3095% 区间
头皮睡眠分期CSP+LDA − EEGNet−6.4 pp[−9.3, −3.3 pp]2095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
头皮睡眠分期Deep4Net − EEGNet−11.3 pp[−13.6, −9.0 pp]2095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
头皮睡眠分期ShallowFBCSPNet − EEGNet−1.6 pp[−3.4, +0.4 pp]2095% 区间,未校正对照为矩阵中单种子的 EEGNet 单元格;未包含训练方差
头皮睡眠分期ShallowFBCSPNet − Deep4Net+9.7 pp[+7.0, +12.7 pp]2095% 区间
运动想象与静息Log band-power ridge − Spectral ridge+2.4 pp[−1.2, +5.8 pp]1095% 配对被试自助法区间(未校正)
SSVEP · 8 通道Log band-power ridge − Spectral ridge−45.4 pp[−50.4, −40.6 pp]7095% 配对被试自助法区间(未校正)
SSVEP · 4 通道Log band-power ridge − Spectral ridge−42.9 pp[−47.7, −38.1 pp]7095% 配对被试自助法区间(未校正)
心算与静息Log band-power ridge − Spectral ridge+8.2 pp[+4.0, +12.6 pp]3695% 配对被试自助法区间(未校正)
P300 目标 ERPLog band-power ridge − Spectral ridge−0.5 pp[−2.4, +1.4 pp]2195% 配对被试自助法区间(未校正)
语义目标 ERPLog band-power ridge − Spectral ridge+5.1 pp[+3.1, +7.1 pp]3095% 配对被试自助法区间(未校正)
头皮睡眠分期Log band-power ridge − Spectral ridge−6.2 pp[−8.5, −3.9 pp]2095% 配对被试自助法区间(未校正)
运动想象与静息LaBraM at Scale Quark-L − LaBraM at Scale Quark-B−2.3 pp[−4.8, −0.3 pp]1095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
SSVEP · 8 通道LaBraM at Scale Quark-L − LaBraM at Scale Quark-B−0.4 pp[−1.3, +0.6 pp]7095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
SSVEP · 4 通道LaBraM at Scale Quark-L − LaBraM at Scale Quark-B+0.1 pp[−0.9, +1.2 pp]7095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
心算与静息LaBraM at Scale Quark-L − LaBraM at Scale Quark-B−2.1 pp[−5.0, +0.8 pp]3695% 配对被试自助法区间(未校正;共 21 项阶梯对比)
P300 目标 ERPLaBraM at Scale Quark-L − LaBraM at Scale Quark-B+1.4 pp[−0.3, +3.3 pp]2195% 配对被试自助法区间(未校正;共 21 项阶梯对比)
语义目标 ERPLaBraM at Scale Quark-L − LaBraM at Scale Quark-B−0.6 pp[−2.8, +1.7 pp]3095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
头皮睡眠分期LaBraM at Scale Quark-L − LaBraM at Scale Quark-B−1.8 pp[−3.3, −0.3 pp]2095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
运动想象与静息LaBraM at Scale Quark-H − LaBraM at Scale Quark-L+0.6 pp[−3.1, +3.5 pp]1095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
SSVEP · 8 通道LaBraM at Scale Quark-H − LaBraM at Scale Quark-L−3.4 pp[−4.4, −2.4 pp]7095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
SSVEP · 4 通道LaBraM at Scale Quark-H − LaBraM at Scale Quark-L−1.7 pp[−2.7, −0.8 pp]7095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
心算与静息LaBraM at Scale Quark-H − LaBraM at Scale Quark-L+0.6 pp[−1.7, +2.8 pp]3695% 配对被试自助法区间(未校正;共 21 项阶梯对比)
P300 目标 ERPLaBraM at Scale Quark-H − LaBraM at Scale Quark-L+1.6 pp[−0.4, +3.5 pp]2195% 配对被试自助法区间(未校正;共 21 项阶梯对比)
语义目标 ERPLaBraM at Scale Quark-H − LaBraM at Scale Quark-L−2.6 pp[−5.1, −0.1 pp]3095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
头皮睡眠分期LaBraM at Scale Quark-H − LaBraM at Scale Quark-L+2.5 pp[+1.0, +4.2 pp]2095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
运动想象与静息LaBraM at Scale Quark-H − LaBraM at Scale Quark-B−1.8 pp[−5.2, +1.0 pp]1095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
SSVEP · 8 通道LaBraM at Scale Quark-H − LaBraM at Scale Quark-B−3.8 pp[−4.8, −2.8 pp]7095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
SSVEP · 4 通道LaBraM at Scale Quark-H − LaBraM at Scale Quark-B−1.6 pp[−2.5, −0.7 pp]7095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
心算与静息LaBraM at Scale Quark-H − LaBraM at Scale Quark-B−1.5 pp[−4.4, +1.3 pp]3695% 配对被试自助法区间(未校正;共 21 项阶梯对比)
P300 目标 ERPLaBraM at Scale Quark-H − LaBraM at Scale Quark-B+3.0 pp[+1.3, +4.8 pp]2195% 配对被试自助法区间(未校正;共 21 项阶梯对比)
语义目标 ERPLaBraM at Scale Quark-H − LaBraM at Scale Quark-B−3.3 pp[−5.5, −0.9 pp]3095% 配对被试自助法区间(未校正;共 21 项阶梯对比)
头皮睡眠分期LaBraM at Scale Quark-H − LaBraM at Scale Quark-B+0.7 pp[−0.9, +2.5 pp]2095% 配对被试自助法区间(未校正;共 21 项阶梯对比)