Extracting representations of cognition across neuroimaging studies improves brain decoding.

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Cognitive brain imaging studies generate valuable data on neural substrates of mental processes.
• Current limitations include small sample sizes, low statistical power, and the need for unified theoretical frameworks for cross-study analysis.
• Existing analytical frameworks struggle to scale for analyzing diverse cognitive tasks across multiple studies.

Purpose of the Study:

• To introduce a novel methodology for analyzing brain responses across diverse cognitive tasks without requiring a unified psychological model.
• To enhance statistical power in smaller, focused studies by integrating them with larger, less focal studies.
• To improve the scalability and analytical power of cross-study brain imaging data analysis.

Main Methods:

• Developed a new analytical framework to analyze brain responses across tasks without a joint model of psychological processes.
• Implemented a method that boosts statistical power by analyzing small, focused studies alongside large, general studies.
• Utilized a data-driven approach to identify commonalities across tasks through shared brain representations.

Main Results:

• The methodology improved decoding performance in 80% of 35 diverse functional-imaging studies.
• Identified common brain networks, predictive of mental processes, by finding data-driven commonalities across tasks.
• These identified brain networks represent interpretable and plausible neural structures tuned to psychological manipulations.

Conclusions:

• The new methodology effectively enhances statistical power and decoding performance in cognitive brain imaging.
• The approach facilitates the discovery of common brain representations across diverse cognitive tasks.
• Extracted brain networks are made available for reuse, and a multi-study decoding tool is provided for new data adaptation.