A comparison of fMRI and behavioral models for predicting inter-temporal choices

Area of Science:

• Neuroscience
• Cognitive Science
• Decision Science

Background:

• Inter-temporal choice (IteCh) involves choosing between immediate smaller rewards and delayed larger rewards.
• Understanding the neural basis of IteCh is crucial for explaining decision-making processes.
• Established behavioral models effectively predict IteCh, but neural correlates remain less understood.

Purpose of the Study:

• To develop a machine learning classifier using trial-by-trial fMRI data to predict IteCh.
• To compare the predictive performance of the fMRI-based classifier against established behavioral models.
• To investigate the potential of single-trial brain activity for identifying state-like factors influencing IteCh.

Main Methods:

• Utilized fMRI data from 363 recording sessions during an IteCh task.
• Employed a support vector machine (SVM) classifier with GLM coefficients as features.
• Implemented a searchlight approach for feature selection and cross-validation, comparing against four behavioral models.

Main Results:

• Behavioral models achieved high accuracies (≥90%).
• The fMRI-based classifier achieved a maximum accuracy of 54.84% in specific brain regions, notably the value-tracking network.
• Simulations confirmed that low accuracies are expected when classifying based on signals with realistic correlations to subjective value.

Conclusions:

• Single-trial fMRI data has limitations in predicting human choices compared to behavioral models.
• The low accuracy highlights challenges in decoding complex decisions from noisy fMRI signals.
• Future research may benefit from paradigms enhancing signal-to-noise ratio, such as miniblocks, for improved predictive power.