GitHub

Treatment MI

This document describes all the changes applied to downloaded Motor Imagery databases from MOABB or base repository.

AlexMI

Class Labels:

  • Original: 2-right_hand, 3-feet, 4-rest
  • Modified: No changes required

Sampling Rate:

  • Original: 512Hz
  • Modified: Downsampled to 256Hz (decimation factor = 2)

Specific Treatments:

  • Epoch labeling correction: In the original data, entire epochs were tagged with the corresponding event class, which can cause problems during epoch slicing. This was corrected by maintaining the non-zero tag only at the first sample of each epoch, setting the tag of all other samples in the epoch to 0.

Technical Note: Typically, the stimulation channel has a non-zero value when an event is triggered (e.g., 1 for left-hand) at one specific sample. When data is sliced into epochs, an epoch starts at the trigger sample and ends at the trigger sample + window length samples (window length = trial length × sampling rate, e.g., for this database: trial length = 3s, sampling rate = 256Hz, window length = 3×256 = 768 samples).

BNCI2014001

Class Labels:

  • Original: 1-lefthand, 2-righthand, 3-feet, 4-tongue
  • Modified: 4-tongue → 6-tongue (standardization)

Sampling Rate:

  • Original: 250Hz
  • Modified: No changes required

Specific Treatments:

  • Offset addition: 2-second offset added as per the original paper's instructions (first 2 seconds after trigger correspond to visual cue indicating incoming MI task)
  • Channel removal: EOG channels removed
  • Session concatenation: Runs concatenated into corresponding sessions (identical experimental conditions)

BNCI2014002

Class Labels:

  • Original: 1-right_hand, 2-feet
  • Modified: 1-righthand → 2-righthand, 2-feet → 3-feet (standardization)

Sampling Rate:

  • Original: 512Hz
  • Modified: Downsampled to 256Hz (decimation factor = 2)

Specific Treatments:

  • Data cleaning: Samples with NaN and/or zero values removed (first and last lines)

  • Offset addition: 3-second offset added as per the original paper's instructions (first 3 seconds after trigger correspond to visual cue indicating incoming MI task)

  • Database separation:

    • BNCI2014002-Train: Runs 1-5 (no feedback, used for classifier training)
    • BNCI2014002-Test: Runs 6-8 (with feedback, used for classifier testing)

BNCI2014004

Class Labels:

  • Original: 1-lefthand, 2-righthand
  • Modified: No changes required

Sampling Rate:

  • Original: 250Hz
  • Modified: No changes required

Specific Treatments:

  • Data cleaning: Samples with NaN and/or zero values removed
  • Channel removal: EOG channels removed
  • Offset addition: 3-second offset added as per the original paper's instructions (first 3 seconds after trigger correspond to visual cue indicating incoming MI task)
  • Database separation:
    • BNCI2014004-Train: Sessions 1-2 (no feedback, used for classifier training)
    • BNCI2014004-Test: Sessions 3-5 (with feedback, used for classifier testing)

Warning: Session 3 is labeled as training in MOABB but as testing in the original paper. We followed the original paper classification.

BNCI2015001

Class Labels:

  • Original: 1-right_hand, 2-feet
  • Modified: 1-righthand → 2-righthand, 2-feet → 3-feet (standardization)

Sampling Rate:

  • Original: 512Hz
  • Modified: Downsampled to 256Hz (decimation factor = 2)

Specific Treatments:

  • Session handling: 2-3 sessions per subject with identical experimental conditions (no separation required)

Cho2017

Class Labels:

  • Original: 1-lefthand, 2-righthand
  • Modified: No changes required

Sampling Rate:

  • Original: 512Hz
  • Modified: Downsampled to 256Hz (decimation factor = 2)

Specific Treatments:

  • Channel removal: EMG channels removed

GrossWentrup2009

Class Labels:

  • Original: 1-lefthand, 2-righthand
  • Modified: No changes required

Sampling Rate:

  • Original: 500Hz
  • Modified: Downsampled to 250Hz (decimation factor = 2)

Specific Treatments:

  • No additional changes required

Lee2019MI

Class Labels:

  • Original: 1-righthand, 2-lefthand
  • Modified: 1-righthand → 2-righthand, 2-lefthand → 1-lefthand (standardization)

Sampling Rate:

  • Original: 1000Hz
  • Modified: Downsampled to 200Hz (decimation factor = 5)

Specific Treatments:

  • Channel removal: EMG channels removed
  • Session handling: 2 sessions with identical experimental conditions (no separation required)

PhysionetMI

Class Labels:

  • Original: 1-rest, 2-lefthand, 3-righthand, 4-both_hands, 5-feet
  • Modified: 1-rest → 4-rest, 2-lefthand → 1-lefthand, 3-righthand → 2-righthand, 4-bothhands → 5-bothhands, 5-feet → 3-feet (standardization)

Sampling Rate:

  • Original: 160Hz
  • Modified: No changes required

Specific Treatments:

  • Data cleaning: Samples with NaN and/or zero values removed
  • Task selection: Only imagined movement tasks included (Tasks 1 and 3 with executed movements excluded)
  • Database separation:
    • PhysionetMI-T2: Task 2 (lefthand and righthand classes) - Runs 4, 8, 12 concatenated
    • PhysionetMI-T4: Task 4 (both_hands and feet classes) - Runs 6, 10, 14 concatenated

Schirrmeister2017

Class Labels:

  • Original: 1-feet, 2-lefthand, 3-rest, 4-righthand
  • Modified: 1-feet → 3-feet, 2-lefthand → 1-lefthand, 3-rest → 4-rest, 4-righthand → 2-righthand (standardization)

Sampling Rate:

  • Original: 500Hz
  • Modified: Downsampled to 250Hz (decimation factor = 2)

Specific Treatments:

  • Session concatenation: Train and test runs concatenated into single session (identical experimental conditions)

Warning: When using paradigm.get_data() from MOABB for .yml metadata acquisition, class labels differ from the original paper. Original labels were used for CSV to NY conversion.

Shin2017A

Class Labels:

  • Original: 1-lefthand, 2-righthand
  • Modified: No changes required

Sampling Rate:

  • Original: 200Hz
  • Modified: No changes required

Specific Treatments:

  • Channel removal: EOG channels removed
  • Session handling: 3 sessions with identical experimental conditions (no separation required)

SPSM2025

Class Labels:

  • Original: 1-lefthand, 2-righthand, 3-rest
  • Modified: 3-rest → 4-rest (standardization)

Sampling Rate:

  • Original: 250Hz
  • Modified: No changes required

Specific Treatments:

  • Session concatenation: 1 session of 2 runs each with identical conditions, runs concatenated into corresponding session

Weibo2014

Class Labels:

  • Original: 1-lefthand, 2-righthand, 3-both_hands, 4-feet, 7-rest
  • Modified: 3-bothhands → 5-bothhands, 4-feet → 3-feet, 7-rest → 4-rest (standardization)

Sampling Rate:

  • Original: 200Hz
  • Modified: No changes required

Specific Treatments:

  • Offset addition: 3-second offset added as per the original paper's instructions (first 3 seconds after trigger correspond to visual cue indicating incoming MI task)
  • Channel removal: EOG and CB channels removed (CB electrodes purpose unclear)
  • Data cleaning: Samples with NaN and/or zero values removed

Zhou2016

Class Labels:

  • Original: 1-lefthand, 2-righthand, 3-feet
  • Modified: No changes required

Sampling Rate:

  • Original: 250Hz
  • Modified: No changes required

Specific Treatments:

  • Session concatenation: 3 sessions of 2 runs each with identical conditions, runs concatenated into corresponding sessions
  • Channel removal: EOG channels removed