QC Group Report

QC Group Report#

QC Group reports provide a dataset-level quality control summary centered on the Global Quality Index (GQI). They visualize how individual recordings compare across multiple QC components and help identify outliers that may need attention.

For execution instructions, see the Tutorial.

What QC Group Reports Show#

The QC Group report answers the question: “How do recordings in this dataset rank by quality, and what drives the differences?”

Unlike QA reports (which profile raw signal characteristics), QC reports summarize quality decisions based on configurable thresholds. The GQI score (0-100%) provides a single quality estimate per recording, while component breakdowns explain which factors contribute to lower scores.

Input and Attempt Selection#

QC Group reads GQI results from attempt-indexed TSV files:

summary_reports/group_metrics/Global_Quality_Index_attempt_<n>.tsv

Attempt resolution order:

  1. Explicit --input_tsv path (if provided)

  2. Explicit --attempt <n> number

  3. Latest available attempt (default)

Each attempt has a matching configuration snapshot:

summary_reports/config/global_quality_index_<n>.ini

This enables comparing how different GQI threshold settings affect quality rankings without re-running the full calculation pipeline.

Understanding the GQI Tab#

The GQI tab is the primary triage interface. It shows:

  1. Global score distribution - Histogram/density of GQI scores across all recordings

  2. Penalty decomposition - How much each penalty family contributes to score reduction

  3. Ranking table - Recordings sorted by GQI score with penalty breakdowns

Interpreting GQI Scores#

Score Range

Interpretation

90-100%

Excellent quality, minimal artifacts

70-89%

Good quality, some artifacts present

50-69%

Moderate quality, notable artifact burden

Below 50%

Poor quality, significant issues

Penalty Families#

The GQI score is reduced by four penalty families:

  • ch (Channel variability): Penalizes recordings with high percentages of noisy or flat channels

  • corr (Correlation): Penalizes recordings with high ECG/EOG contamination

  • mus (Muscle): Penalizes recordings with frequent muscle artifacts

  • psd (PSD noise): Penalizes recordings with high spectral noise burden

Understanding Component Tabs#

Each component tab (STD, PtP, PSD, ECG, EOG, Muscle) provides detailed views of that specific quality metric:

Distribution Plots#

  • Violin/box plots: Show the spread of values across recordings

  • Density plots: Reveal clustering and outlier patterns

  • Subject markers: Individual recordings shown as points for identification

Ranking Tables#

  • Recordings sorted by component value (worst first)

  • Hover information shows full recording identifiers

  • Helps quickly identify which recordings drive group-level patterns

How to Interpret QC Group Reports#

Recommended workflow:

  1. Start with the GQI tab - Get an overview of quality distribution and identify low-scoring recordings

  2. Check penalty breakdown - Determine which penalty families drive low scores

  3. Open component tabs - Investigate specific metrics (e.g., if ch penalty is high, check STD/PtP tabs)

  4. Compare channel types - Switch between MAG/GRAD to see if issues are sensor-type specific

  5. Track attempts - If you’ve run multiple GQI attempts with different thresholds, compare how rankings change

Notes on Combined View#

The Combined (mag+grad) tab provides a unified overview but requires careful interpretation:

  • Unit mixing: Amplitude-based metrics combine pT (MAG) and pT/m (GRAD) values

  • Best for: Quick triage and identifying recordings with issues in either channel type

  • Validate in: MAG and GRAD tabs for unit-specific interpretation

Practical Example#

Suppose you see a recording with GQI = 62% and high ch penalty:

  1. Open the STD tab → Check if noisy channel % is elevated

  2. Open the PtP tab → Check if there are transient amplitude issues

  3. Switch to MAG vs GRAD → Determine if the problem is sensor-type specific

  4. Return to QA Subject report for that recording → Inspect the actual channel×epoch heatmaps

This drill-down approach helps you understand not just that a recording has issues, but why and where.