interpolate-bads

name	interpolate_bads
description	Spherical spline interpolation to reconstruct bad channel data from neighbors
layer	L3
group	channel
metadata	{"tags":["operator","channels","interpolation","repair","spline"],"modalities":["eeg","ecog"],"step_string":"interpolate_bads","analysis_goal_allowed":["classification","source_localization","feature_extraction","clinical_screening","exploratory","generic","connectivity","phase_amplitude_coupling","online_inference"],"analysis_goal_forbidden":[]}

Interpolate Bad Channels

Function

Reconstructs data for bad channels using spherical spline interpolation from neighboring good channels. Preserves total channel count while replacing corrupted data with estimated values.

Parameter Format

interpolate_bads — No parameters. Operates on channels marked as bad in metadata.

How It Works

1. Identifies channels in meta.bads or raw.info["bads"]
2. Uses spherical spline interpolation (MNE) based on electrode geometry
3. Replaces bad channel data with interpolated values
4. Channels remain in the array with repaired data

When to Use

• Need consistent channel count across subjects (e.g., for group analysis)
• Downstream analysis requires specific electrode positions (source localization, topographic maps)
• Only 1-3 channels are bad (interpolation quality degrades with many bad channels)
• Better than dropping when spatial patterns matter

When NOT to Use

• No channel location information available (interpolation needs geometry)
• Too many bad channels (> 15-20% of total) — interpolation becomes unreliable
• sEEG depth electrodes (geometry assumptions don't apply)
• Don't care about maintaining channel count → use drop_bads (simpler)

Constraints

• Requires electrode positions (montage) for accurate interpolation
• Quality degrades if neighboring channels are also bad
• Interpolated channels contain NO new information — they smooth from neighbors
• Should be applied before CAR (reconstructed channels participate in the average)

Ordering

• Apply BEFORE: car (interpolated channels should contribute to average)
• Apply BEFORE: ica (prevents wasting components on bad channel noise)
• Apply AFTER: quality check identifies which channels are bad

Reference Code

Standalone Implementation

import mne
import numpy as np

info = mne.create_info(ch_names, sfreq, ch_types='eeg')
raw = mne.io.RawArray(data, info, verbose=False)
raw.info['bads'] = bad_channels  # e.g. ['Fp1', 'T8']
raw.interpolate_bads(reset_bads=True, verbose=False)
data = raw.get_data().astype(np.float32)

Key API

• raw.info['bads'] = [...] — mark channels as bad
• raw.interpolate_bads(reset_bads=True, verbose=False) — spherical spline interpolation
• reset_bads=True clears the bads list after interpolation