A Novel PCA-Based Wishart Filtering Method for Reduction of Unstructured Noise in fMRI and Connectomes to Improve Diagnosis of Neurodegenerative Diseases

Overview

With increasing interest in using fMRI and functional connectivity networks to understand and diagnose neurodegeneration, advanced method for image processing are necessary to produce highly structured images. fMRI, as a noninvasive in vivo method for measuring brain function, is a convenient and effective analysis tool to understand the presence and progression of neurodegenerative diseases such as Parkinson’s and Alzheimer’s Disease. Unstructured noise reduction is an essential step in fMRI processing for removing nonBOLD noise to reveal underlying neural function; however, no intuitive methods exist for this step in processing. Conventional processes such as spatiotemporal smoothing cannot effectively differentiate BOLD signal and noise, causing reduced image quality and shifted neural signal. This makes using fMRI for clinical diagnosis challenging. Therefore, this study developed a novel process for differentiating and retaining neural signal more effectively called the Wishart Filtering method, which utilizes a noise eigenspectrum subtraction to remove unstructured eigenvalues that pollute the PCA eigenspectrum of the fMRI and connectome. This study analyzed the Wishart Filter’s effects on random noise, connectivity, and gradience in fMRI and connectomes. Repeated iterations of Wishart Filtering were compared to a temporal low pass filter and spatial smoothing. This study showed that Wishart Filtering significantly reduces noise and improves connectivity and gradience in fMRI and connectomes more effectively than spatiotemporal smoothing without causing additional detrimental effects. This novel tool in fMRI processing and analysis has the potential to improve fMRI as a more effective clinical diagnostic technique for neurodegeneration.