A Parameter-free unsupervised framework for fMRI data analysis using batch learning growing neural gas and...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging Analysis

Background:

• Functional magnetic resonance imaging (fMRI) time-series analysis is crucial for identifying brain activity and neural patterns.
• Current clustering methods often require user-defined parameters, limiting their practical application.
• Automated and stable cluster number selection remains a challenge in fMRI data analysis.

Purpose of the Study:

• To introduce a novel parameter-free hierarchical topological structure learning clustering algorithm for fMRI data.
• To improve convergence speed and eliminate manual parameter tuning in fMRI analysis.
• To enhance the robustness of clustering through a two-stage false positive rate control mechanism.

Main Methods:

• Developed the Batch Learning Growing Neural Gas (BL-GNG) algorithm, an extension of the Growing Neural Gas (GNG) model.
• Implemented a parameter-free approach to hierarchical topological structure learning.
• Incorporated a two-stage false positive rate control using randomization inference and 3D neighborhood criteria.

Main Results:

• BL-GNG demonstrated superior performance on real fMRI data, achieving a Jaccard Coefficient of 0.99 and an Area Under the ROC Curve of 0.97.
• The algorithm showed high stability across 50 iterative runs, outperforming K-means, FCM, NG, and GNG.
• BL-GNG achieved significant computational efficiency with an average execution time of 26 seconds.

Conclusions:

• BL-GNG shows significant potential as a robust and efficient tool for fMRI data analysis.
• The algorithm can aid in diagnosing brain disorders and investigating neural subnetworks.
• BL-GNG advancements can contribute to progress in cognitive neuroscience research.