Skip to content

Classification Workflow

This guide describes a practical workflow for reviewing and classifying ICA components using Rica.

Overview

The goal of manual component classification is to:

  1. Verify tedana's automatic classifications
  2. Correct any misclassified components
  3. Export your decisions for use in tedana

Before You Start

Understand Your Data

Before classifying components:

  • Review the Info tab to understand processing parameters
  • Note the number of components and their distribution
  • Check the QC tab for data quality issues

Set Your Environment

  • Use a large monitor (recommended: 1920x1080 or larger)
  • Switch to dark or light theme based on your preference
  • Consider using keyboard shortcuts for efficiency

Step 1: Review the Big Picture

Start in the ICA tab and look at the scatter plot:

  1. Identify clusters: Do accepted and rejected components form distinct groups?
  2. Check elbow lines: Are tedana's thresholds reasonable?
  3. Note outliers: Which components fall in ambiguous regions?

Step 2: Quick Scan of Accepted Components

Review tedana's accepted components first:

For each accepted component:
  1. Check spatial map → Does it look like brain signal?
  2. Check time series → Is it smooth and plausible?
  3. Check FFT → Is energy in expected frequency bands?
  4. If suspicious → Consider rejecting

What to look for in accepted components

  • Spatial maps localized to gray matter
  • Time series without sudden spikes
  • Frequency content in BOLD range (0.01-0.1 Hz)

Step 3: Review Rejected Components

Check tedana's rejected components:

For each rejected component:
  1. Check if it looks like true signal
  2. If it looks like BOLD → Consider accepting
  3. Common false rejections:
     - Strong task activation
     - Unusual but real brain networks

Common noise patterns to confirm rejection

  • Ring artifacts at brain edges
  • Ventricle-localized signal
  • Stripe patterns from motion
  • Spiky time series

Step 4: Focus on Edge Cases

Components near the decision boundaries deserve extra attention:

  1. Low kappa, low rho: Often unclear - use your judgment
  2. Near elbow thresholds: Could go either way
  3. Unusual spatial patterns: May be artifact or real signal

Step 5: Save Your Classifications

When finished:

  1. Click the Save button
  2. File downloads as manual_classification.tsv
  3. Move the file to your tedana output directory

Using Classifications with Tedana

Apply your manual classifications in tedana:

tedana -d your_data.nii.gz \
       -e 14.5 29 43.5 \
       --manual-classification manual_classification.tsv

Tedana will use your classifications instead of running automatic classification.

Classification Criteria

Accept if:

Criterion Description
Spatial pattern Localized to gray matter, follows known networks
Time series Smooth, no spikes, physiologically plausible
Frequency Energy primarily in BOLD range
Kappa/Rho High kappa relative to rho

Reject if:

Criterion Description
Spatial pattern Edges, ventricles, outside brain, stripes
Time series Spiky, sudden jumps, drift
Frequency High frequency noise, respiratory/cardiac peaks
Kappa/Rho Low kappa and/or high rho

Edge Cases

Some situations require judgment:

Situation Consideration
Motion-correlated but brain-like May contain both signal and noise
Strong task activation High kappa but could look unusual
Large draining veins Real signal but may want to remove
Global signal Controversial - depends on your analysis

Tips for Efficiency

Use Keyboard Shortcuts

Key Action
→ Next component
← Previous component
A Accept
R Reject

Batch Similar Components

If you notice a pattern (e.g., all edge artifacts):

  1. Sort the table by a relevant metric
  2. Review similar components together
  3. Apply consistent criteria

Take Breaks

  • Component classification requires sustained attention
  • Take breaks every 20-30 components
  • Fresh eyes catch mistakes

Quality Control Checklist

Before finalizing your classifications:

  • [ ] Reviewed all tedana-accepted components
  • [ ] Spot-checked rejected components
  • [ ] Examined edge cases carefully
  • [ ] Pie chart shows reasonable variance distribution
  • [ ] Saved classifications to file

Documenting Your Decisions

For reproducibility, consider documenting:

  • How many components you changed
  • Criteria you used for edge cases
  • Any unusual patterns in your data

This helps when:

  • Returning to the analysis later
  • Explaining decisions to collaborators
  • Writing methods sections