Alzheimer's Disease Diagnosis Based on Derivative Dynamic Time Warping Functional Connectivity Networks.

Due to the dynamic changes and complex time-delay characteristics of signal transmission between brain regions, this information can help identify early signs in individuals with Alzheimer's Disease (AD). Conventional Functional Connectivity Networks (FCN) may overlook interaction delays, leading to inaccurate representations of activity between brain regions. Recognizing the varying delays between patients and healthy individuals, we constructed a dynamic FCN using a Sliding Window based on Derivative Regularity Correlation (SWDRC) and a Functional Delay Network (FDN). These methods aim to improve the detection and analysis of brain networks through the Correlation-based Derivative Regularity (CDR) algorithm. The key advancement of this study is the CDR algorithm, which enables nonlinear time series alignment, unlike traditional correlation methods. This improvement allows for the analysis of asynchronous and nonlinear features in brain activity. Using CDR, SWDRC identifies local and asynchronous characteristics via a sliding window, while FDN quantifies measurable delays between brain regions in both healthy subjects and patients. Our methods show strong potential for revealing the mechanisms underlying neurodegenerative conditions. In classification experiments, combining complementary features from SWDRC and FDN achieved higher accuracy, effectively extracting disease-related patterns. FDN analysis of ADNI data indicates an increasing network delay from Healthy Control to Mild Cognitive Impairment to AD. Our code and models are available at https://github.com/hxpotato/SWDRC.