Introduction: The brain is one of the most vital and complex organs of the human body. For centuries, researchers have sought to unravel the intricacies of its functioning. As our understanding deepens, brain mapping has emerged as an increasingly significant area of scientific inquiry. Understanding the brain is, in many ways, synonymous with understanding human behavior, cognition, and future potential. Statistical methods have proven to be indispensable tools in advancing this research. In this paper, we introduce a novel and rapid Bayesian statistical method designed to enhance the process of brain mapping. This method offers both speed and precision, providing an efficient means of advancing our understanding of the brain's structure and function.
Methods: In this paper, we present a new spatiotemporal Bayesian method for brain mapping, building upon the fast Bayesian approach developed by Masgarov. A key innovation in our method is the replacement of the traditional estimation technique with the Integrated Nested Laplace Approximation (INLA), a powerful tool for approximate Bayesian inference. INLA offers significant advantages in terms of both computational efficiency and accuracy, especially when compared to sampling-based methods such as Markov Chain Monte Carlo (MCMC). These improvements contribute to its growing popularity in Bayesian inference. We applied this model to fMRI data obtained from an auditory task and compared its performance to that of previous models, demonstrating its enhanced capability for brain mapping.
Results: The proposed method was compared with both the conventional General Linear Model (GLM) and the fast Bayesian method. Key performance metrics, including computation time, activation areas, and the False Discovery Rate (FDR), were evaluated. While the GLM was the fastest in terms of execution time, our INLA-based fast Bayesian model demonstrated superior precision and noise reduction, with only a marginal difference in speed.
All three methods identified similar primary activation regions in the brain; however, our model detected a higher number of activated areas. Additionally, the INLA fast Bayesian model more effectively removed noise and irrelevant voxels, resulting in more accurate and detailed brain maps.
Conclusion: Brain mapping with modern statistical methods offers the potential for deeper and more accurate insights into brain function. While spatiotemporal statistical models generally require more computational time compared to conventional General Linear Models (GLM), they provide a more precise representation of the brain's complex dynamics. By accounting for both spatial and temporal data, these models are better suited to capture the true capacity of real-world data, ultimately leading to more reliable and nuanced findings in brain research.
Keywords: Brain mapping, fast Bayesian mode,INLA, Spatiotemporal