Application of artificial neural network to fMRI regression analysis

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Functional magnetic resonance imaging (fMRI) is a key tool for understanding brain activity.
• Analyzing the relationship between experimental events and fMRI signals is crucial for neuroscience research.
• Traditional methods may have limitations in detecting complex or continuous signal modulations.

Purpose of the Study:

• To introduce and evaluate an artificial neural network (ANN) approach for analyzing event-related fMRI data.
• To demonstrate the capability of ANNs in detecting correlations between event sequences and fMRI signals.
• To highlight the advantages of ANNs for exploratory fMRI analyses, particularly for parametrically modulated responses.

Main Methods:

• A layered feed-forward artificial neural network (ANN) was employed.
• The ANN performed non-linear regression analysis, using event sequences as input and fMRI signals as the supervised output.
• Cross-validation and an early stopping procedure were utilized to mitigate issues with autocorrelation noise inherent in fMRI data.

Main Results:

• The ANN successfully detected various fMRI responses with different temporal dynamics.
• Simulations confirmed the ANN's ability to detect parametrically modulated responses without requiring a priori assumptions.
• The ANN approach proved effective in identifying continuous changes in brain responses correlated with input variations.

Conclusions:

• ANN-based regression analysis is a powerful and flexible tool for exploratory fMRI studies.
• This method can uncover complex relationships between experimental events and brain activity, including continuous modulations.
• The ANN approach offers an advantage over traditional methods by not requiring pre-defined assumptions about response shapes.