On the Search for Data-Driven and Reproducible Schizophrenia Subtypes Using Resting State fMRI Data From Multiple
Lærke Gebser Krohne1,2, Ingeborg Helbech Hansen3, Kristoffer H Madsen1,4
1Department of Applied Mathematics and Computer Science, Technical University of Denmark, 2800 Kgs. Lyngby, Denmark.
Neural Computation
|August 6, 2024
Summary
Researchers used unsupervised clustering on fMRI data to find schizophrenia subtypes. They identified one cluster showing diagnostic differences, but symptom severity did not correlate, suggesting other disease mechanisms.
Area of Science:
- Neuroscience
- Psychiatry
- Data Science
Background:
- Functional magnetic resonance imaging (fMRI) data has been explored for schizophrenia biomarkers for decades.
- High internal heterogeneity of schizophrenia hinders firm conclusions from fMRI studies.
- Identifying patient subgroups with homogeneous biological profiles is a promising approach.
Purpose of the Study:
- To apply an unsupervised multiple co-clustering (MCC) method to identify schizophrenia subtypes using functional connectivity data.
- To investigate the stability and diagnostic association of identified clusters.
- To validate findings on an external dataset.
Main Methods:
- Utilized a multisite resting-state fMRI dataset merged from two public databases.
- Split data into discovery (143 patients, 143 HC) and external test sets (63 patients, 63 HC).
- Applied MCC to identify patient subgroups and evaluated cluster solutions for diagnostic association, separability, and clinical correlation (PANSS).
Main Results:
- Clustering stability was moderate and dependent on data variations and initializations.
- One cluster solution ('view') showed a significant diagnosis association, with three subject clusters overrepresenting schizophrenia patients.
- A feature cluster exhibited a continuous trend in connectivity values correlated with schizophrenia patient proportions.
- No feature clusters in the identified view correlated with positive, negative, or generalized symptom severity.
Conclusions:
- Despite moderate clustering stability, a reproducible schizophrenia patient subgroup was identified using functional connectivity.
- The identified cluster solution reflects disease-related mechanisms beyond symptom severity.
- Further research is needed to understand the biological underpinnings of these identified subgroups.
