Multivariate Classification of First-Episode Schizophrenia Spectrum Psychosis using EEG Microstate Dynamics

This study demonstrates that multivariate EEG microstate dynamics can distinguish first-episode schizophrenia spectrum psychosis from healthy controls and specifically correlate with negative symptom severity, despite lacking significant univariate differences.

The Gist

Imagine your brain is like a busy, high-speed train station. Every few milliseconds, a different set of trains (brain networks) takes over the tracks to send signals around the city. In a healthy brain, these trains switch on and off in a smooth, rhythmic, and predictable dance.

This paper is about a new way of looking at that train station to see if we can spot the difference between a healthy station and one where the trains are running a bit chaotically because of Schizophrenia, specifically in people who have just had their first major episode of the illness.

Here is the breakdown of what the researchers did and found, using some everyday analogies:

1. The "Snapshot" Method (EEG Microstates)

Instead of watching the whole movie of brain activity, the researchers looked at microstates. Think of these as tiny, split-second snapshots of the train station.

• In a healthy station, the snapshots show a specific pattern: "Trains A and B are running, then quickly switch to C and D."
• The researchers wanted to see if the "snapshots" of people with first-episode schizophrenia looked different from healthy people.

2. The Detective Work (The Study)

They gathered 69 people: 41 who had just been diagnosed with the condition and 28 healthy volunteers. They put them in a room, hooked them up to EEG machines (like a helmet that listens to brain waves), and asked them to sit quietly.

They didn't just look at one snapshot; they looked at 28 different clues at once. These clues included:

• How long a specific "train pattern" lasted.
• How often the trains switched from one pattern to another.
• The order in which the patterns appeared.

3. The "Group vs. Individual" Puzzle

Here is the tricky part. When the researchers looked at the clues one by one (like asking, "Did the trains run longer for Group A?"), they couldn't find a single clear difference. It was like trying to find a single broken gear in a massive clock; looking at one gear tells you nothing.

However, when they used a computer program (a "multivariate classifier") to look at all the clues together as a team, the computer could tell the difference!

• The Result: The computer was about 64-69% accurate at guessing whether a person had schizophrenia or was healthy, just by looking at the combination of how their brain snapshots behaved.
• The Analogy: It's like recognizing a friend's handwriting. You might not be able to spot a difference in just the letter "A" or just the letter "B," but when you look at the whole sentence, the unique style becomes obvious. The brain of someone with schizophrenia has a unique "handwriting" made of many small, connected patterns.

4. The Symptom Connection (The "Negative" Link)

The researchers also asked: "Do these brain patterns get worse when the person's symptoms get worse?"

They found a very specific link with Negative Symptoms.

• What are Negative Symptoms? These aren't "bad" behaviors; they are the absence of normal things, like losing motivation, feeling no emotion, or not wanting to talk to people.
• The Finding: The more severe a person's lack of motivation or emotion was, the more their brain "snapshots" changed. Specifically, the "D" pattern of brain activity became shorter, and the "A" and "B" patterns happened more often.
• The Analogy: Imagine the brain's "motivation engine" (Pattern D) is running out of gas. The more the engine sputters (shorter duration), the more the person feels unmotivated. Interestingly, the "positive symptoms" (like hearing voices or seeing things that aren't there) didn't seem to have a direct link to these specific brain patterns in this study.

The Bottom Line

This study suggests that while we can't find one single "broken switch" in the brains of people with early-stage schizophrenia, we can see a unique, complex pattern of how their brain networks dance together.

This "dance" is different enough for a computer to spot the difference between a healthy brain and a sick one. Furthermore, the way this dance changes seems to be a direct reflection of how much a person is struggling with a lack of motivation and emotion, offering a potential new tool to understand and track the illness.

Technical Summary

1. Problem Statement

Schizophrenia Spectrum Psychosis (SSP) is a severe mental disorder characterized by disruptions in brain network dynamics. While Electroencephalography (EEG) microstates—quasi-stable, topographically distinct patterns of scalp potential lasting tens of milliseconds—have shown promise in revealing large-scale brain network alterations in chronic schizophrenia, evidence regarding First-Episode Schizophrenia Spectrum Psychosis (FESSP) remains limited.

The core problem addressed is twofold:

1. Diagnostic Gap: Can rapid, large-scale brain network dynamics (microstates) serve as a biomarker to distinguish FESSP patients from healthy controls?
2. Symptom Correlation: Do these microstate dynamics correlate with specific clinical symptom severity (positive vs. negative symptoms) in the early stages of the illness?

2. Methodology

The study employed a multivariate machine learning approach combined with rigorous statistical testing on resting-state EEG data.

• Participants:
  • Total: 69 participants.
  • FESSP Group: $n=41$ (Mean age: 22.49 years).
  • Healthy Controls (HC): $n=28$ (Mean age: 21.33 years).
• Data Acquisition & Preprocessing: Resting-state EEG was recorded and analyzed to extract microstate classes A through D.
• Feature Extraction:
  • A total of 28 features were extracted, encompassing both temporal parameters (e.g., duration, occurrence, coverage) and transition probabilities between microstate classes.
• Classification Strategy:
  • Algorithm: Linear Support Vector Machine (SVM).
  • Validation: Stratified cross-validation to ensure balanced class representation.
  • Significance Testing: Permutation testing was used to determine if classification accuracy exceeded chance levels.
• Clinical Correlation Analysis:
  • Within the FESSP group, microstate features were correlated with clinical scores using:
    • Brief Psychiatric Rating Scale (BPRS)
    • Scale for the Assessment of Positive Symptoms (SAPS)
    • Scale for the Assessment of Negative Symptoms (SANS)
  • Statistical significance was adjusted for multiple comparisons using False Discovery Rate (FDR) correction.

3. Key Contributions

• Multivariate Signature Identification: The study moves beyond univariate analysis (comparing single features) to demonstrate that FESSP is characterized by a distributed multivariate pattern of correlated microstate features.
• Early-Stage Biomarker Validation: It provides evidence that EEG microstate dynamics are altered even in the first episode of psychosis, suggesting these metrics could serve as early diagnostic tools.
• Symptom-Specific Mapping: The research successfully dissociates the neural correlates of negative symptoms from positive symptoms, identifying specific microstate dynamics linked to the severity of negative symptomatology.

4. Results

• Classification Performance:
  • The multivariate model achieved above-chance discrimination between FESSP and healthy controls.
  • Balanced Accuracy: 0.644
  • Area Under the Curve (AUC): 0.688
  • Statistical Significance: $p = 0.030$ (via permutation testing).
• Univariate Analysis:
  • Crucially, no single individual feature survived multiple-comparison correction when comparing groups directly. This confirms that the diagnostic power lies in the complex interplay of features rather than isolated abnormalities.
• Clinical Correlations (FESSP Group):
  • Negative Symptoms (SANS): Strong associations were found.
    • Higher SANS scores correlated with shorter durations of Microstate D ($\rho = -0.507, p_{FDR} = 0.020$).
    • Higher SANS scores correlated with higher occurrence of Microstates A and B ($\rho = 0.434\text{--}0.443, p_{FDR} = 0.042$).
  • Positive Symptoms (SAPS) & General Severity (BPRS-18): No significant associations were found with any microstate features.

5. Significance and Conclusion

This study establishes that EEG microstate dynamics offer a viable, non-invasive window into the pathophysiology of first-episode psychosis. The findings highlight that:

1. Distributed Network Dysfunction: The pathology of FESSP is best understood as a distributed network dysfunction rather than a deficit in a single brain state, necessitating multivariate analytical approaches.
2. Negative Symptom Specificity: The specific link between Microstate D (often associated with the salience network) and negative symptoms suggests that negative symptoms in early psychosis may be driven by distinct network instability compared to positive symptoms.
3. Clinical Utility: These dynamics could potentially be developed into objective biomarkers for early diagnosis and for tracking the progression of negative symptoms, which are notoriously difficult to treat and measure subjectively.