مقالات پذیرفته شده در هشتمین کنگره بین المللی زیست پزشکی
Application of a new Wavelet based method for working memory fMRI data
Application of a new Wavelet based method for working memory fMRI data
Fateme Goodarzi,1Anoshirvan Kazemnejad,2Azam Saffar,3,*
1. Department of Biostatistics Faculty of Medical Sciences Tarbiat Modares University (TMU) Tehran, Iran 2. Department of Biostatistics Faculty of Medical Sciences Tarbiat Modares University (TMU) Tehran, Iran
Introduction: Functional Magnetic Resonance Imaging (fMRI) is a powerful tool for studying brain activity, but traditional models often fail to capture the dynamic, time-varying nature of neural processes. Most conventional models, such as the General Linear Model (GLM), rely on static assumptions, often averaging brain activity over time, and therefore may not account for transient or rapidly evolving patterns in brain function. This can lead to less accurate representations of how brain regions respond to cognitive tasks, particularly for tasks that engage multiple neural processes at different time scales. To address this limitation, we applied a generalized wavelet analysis model, which incorporates the dynamic properties of brain activity. This method allows for more precise brain mapping, especially in identifying both short-term and long-term neural activations. In this study, we sought to compare the performance of the generalized wavelet model with the more commonly used GLM approach, specifically in the context of a memory task. Our goal was to assess the accuracy, efficiency, and noise reduction capabilities of the wavelet model in fMRI analysis.
Methods: We began by formulating the generalized wavelet model for analyzing time-varying brain activity. Wavelet analysis is particularly well-suited for this type of data because it allows for multi-resolution decomposition, enabling us to examine both slow and fast changes in brain activity simultaneously. The model was adapted to handle fMRI data, which is characterized by high dimensionality and complex temporal structures. Following the model formulation, the data used for this study were acquired from an fMRI memory task, where participants engaged in recalling specific information over a given period.
Before applying the models, the fMRI data underwent extensive preprocessing, including motion correction, spatial smoothing, and normalization to account for head movement, variability in brain anatomy, and other common sources of noise in fMRI studies. Once the data were preprocessed, both the generalized wavelet model and the traditional GLM were fitted to the dataset.
The General Linear Model was chosen as the baseline for comparison, as it remains the most widely used method for fMRI data analysis. The GLM treats brain activity as a linear response to external stimuli, assuming a stationary relationship between the stimulus and brain response over time. However, because this model averages activity across the entire scanning period, it may not detect transient brain responses or changes in brain activation patterns.
On the other hand, the generalized wavelet model allows for time-frequency decomposition of the fMRI signal, meaning that it can capture brain activity at multiple time scales. This is particularly advantageous for memory tasks, where different brain regions may be engaged at different points in time. After fitting both models to the data, brain activity maps were generated based on each method's analysis.
Results: The brain activity maps produced by both the generalized wavelet model and the GLM were compared in terms of accuracy and the regions identified as active. In both models, the primary regions of brain activity related to the memory task were consistent, demonstrating the validity of the wavelet approach in detecting key brain areas associated with memory retrieval. However, the wavelet-based model provided several key advantages over the GLM:
Increased sensitivity in primary regions: The wavelet model showed higher levels of activation in the primary brain regions involved in the memory task. These areas, such as the prefrontal cortex and hippocampus, were more prominently activated in the wavelet maps compared to the GLM maps.
Reduced noise in peripheral regions: Unlike the GLM, the wavelet model reduced false-positive activations or noisy signals in peripheral areas of the brain, particularly in regions not typically associated with memory tasks. This highlights the wavelet model’s ability to reduce noise and improve the clarity of the brain maps.
Slightly increased computation time: While the wavelet model provided more detailed and noise-resistant results, it was computationally more intensive than the GLM. The generalized wavelet analysis took slightly longer to compute, though the difference was relatively small and outweighed by the increase in accuracy and noise reduction.
Conclusion: In this study, we introduced a novel approach using generalized wavelet analysis for fMRI data, demonstrating its effectiveness in capturing the dynamic nature of brain activity. As expected, this method was able to identify brain regions with greater precision compared to the commonly used General Linear Model. The wavelet-based model performed particularly well in identifying transient brain activations and reducing noise, which are critical in understanding complex cognitive processes like memory. Although the method required slightly more computational resources due to its complexity, the results indicate that it is a superior approach for analyzing non-stationary brain activity. Moreover, the main regions identified by both the wavelet and GLM models corresponded well to areas known to be involved in memory tasks, further validating the accuracy of the wavelet method. Future studies could explore further optimizations of this approach and apply it to other cognitive functions to assess its broader applicability.