Multi-dimensional persistent feature analysis identifies connectivity patterns of resting-state brain networks in Alzheimer’s disease

摘要

Objective. The characterization of functional brain network is crucial to understanding the neuralmechanisms associated with Alzheimer’s disease (AD) and mild cognitive impairment (MCI).Some studies have shown that graph theoretical analysis could reveal changes of the disease-relatedbrain networks by thresholding edge weights. But the choice of threshold depends on ambiguouscognitive conditions, which leads to the lack of interpretability. Recently, persistent homology (PH)was proposed to record the persistence of topological features of networks across every possiblethresholds, reporting a higher sensitivity than graph theoretical features in detecting network-levelbiomarkers of AD. However, most research on PH focused on zero-dimensional features(persistence of connected components) reflecting the intrinsic topology of the brain network,rather than one-dimensional features (persistence of cycles) with an interesting neurobiologicalcommunication pattern. Our aim is to explore the multi-dimensional persistent features of brainnetworks in the AD and MCI patients, and further to capture valuable brain connectivity patterns.Approach. We characterized the change rate of the connected component numbers across graphfiltration using the functional derivative curves, and examined the persistence landscapes thatvectorize the persistence of cycle structures. After that, the multi-dimensional persistent featureswere validated in disease identification using a K-nearest neighbor algorithm. Furthermore, aconnectivity pattern mining framework was designed to capture the disease-specific brainstructures. Main results. We found that the multi-dimensional persistent features can identifystatistical group differences, quantify subject-level distances, and yield disease-specific connectivitypatterns. Relatively high classification accuracies were received when compared with graphtheoretical features. Significance. This work represents a conceptual bridge linking complex brainnetwork analysis and computational topology. Our results can be beneficial for providing acomplementary objective opinion to the clinical diagnosis of neurodegenerative diseases.