Dynamic brain fluctuations outperform connectivity measures and mirror pathophysiological profiles across dementia...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Atypical brain network fluctuations are characteristic of neurodegeneration.
• Traditional static and average functional magnetic resonance imaging (fMRI) connectivity methods have yielded inconsistent results in neurodegenerative diseases due to overlooking dynamic changes.
• Resting-state networks (RSNs) are crucial for understanding brain function and dysfunction in conditions like Alzheimer's disease (AD) and behavioral variant frontotemporal dementia (bvFTD).

Purpose of the Study:

• To introduce and validate a data-driven machine learning pipeline using dynamic connectivity fluctuation analysis (DCFA) for classifying neurodegenerative diseases.
• To investigate the non-linear oscillatory patterns across key RSNs (Salience Network, Default Mode Network, Executive Network, Motor Network, Visual Network) in bvFTD and AD patients compared to healthy controls.
• To compare the efficacy of DCFA with traditional static and dynamic connectivity approaches in differentiating between AD, bvFTD, and healthy controls.

Main Methods:

• A multicenter study involving 300 participants (bvFTD patients, AD patients, and healthy controls) using resting-state fMRI (RS-fMRI) data.
• Development of a machine learning pipeline employing dynamic connectivity fluctuation analysis (DCFA) and a Gradient Boosting Machines (GBM) algorithm with Bayesian hyperparameter tuning.
• Analysis of non-linear oscillatory patterns across individual and combined RSNs, including the SN, DMN, EN, MN, and VN, as features for classification across four independent datasets.

Main Results:

• The DCFA pipeline achieved high classification accuracy for bvFTD (SN + EN: 86.43%), AD (DMN + EN: 86.63%), and bvFTD vs. AD (DMN + SN: 82.67%).
• The analysis revealed a systematic and unique architecture of RSN disruption specific to each neurodegenerative condition.
• DCFA classification significantly outperformed traditional static and linear dynamic connectivity methods in differentiating between the studied groups.

Conclusions:

• Non-linear dynamical fluctuations in RSNs provide a more accurate pathophysiological characterization of neurodegenerative conditions like AD and bvFTD than traditional connectivity approaches.
• The developed DCFA machine learning pipeline offers a robust and sensitive tool for classifying neurodegenerative diseases using multicenter RS-fMRI data.
• This study highlights the importance of analyzing dynamic brain network fluctuations for a deeper understanding of neurodegeneration.