Prediction of activation patterns preceding hallucinations in patients with schizophrenia using machine learning with...

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Detecting functional MRI (fMRI) signal changes during hallucinations is challenging.
• Existing machine learning methods for fMRI data analysis can be difficult to interpret.

Purpose of the Study:

• To develop an automated machine learning algorithm for identifying pre-hallucination fMRI states.
• To improve the interpretability of fMRI-based classification models for hallucinations.
• To characterize the variability in brain activity patterns preceding hallucinations.

Main Methods:

• Applied structured sparsity with total variation penalty to resting-state fMRI data for classification.
• Utilized an extension of principal component analysis with spatial constraints to analyze pattern variability.
• Investigated whole-brain fMRI data from individuals experiencing hallucinations.

Main Results:

• Achieved reliable classification performance with interpretable predictive patterns, identifying clusters in speech-related brain regions.
• Characterized the variability within pre-hallucination patterns, revealing intrinsic data structures.
• Demonstrated the potential for identifying distinct functional patterns preceding hallucinations.

Conclusions:

• The developed machine learning approach offers interpretable insights into pre-hallucination brain activity.
• Understanding pattern variability is crucial for effective hallucination detection and treatment.
• Results support the development of fMRI-guided therapies, such as neurofeedback, for drug-resistant hallucinations.