Constructing Compact Signatures for Individual Fingerprinting of Brain Connectomes

Vikram Ravindra, Petros Drineas, Ananth GramaFrontiers in Neuroscience 2021 · DOI

Abstract

Recent neuroimaging studies have shown that functional connectomes are unique to individuals, acting as a neural "fingerprint." This study presents a novel method to identify the specific, minimal components of resting-state and task-specific connectomes responsible for these unique signatures. By employing advanced matrix sampling techniques, we demonstrate that a remarkably small subset of connections—fewer than 100—can achieve high accuracy in individual identification. These features are statistically significant, robust, and localized to specific structural regions, primarily the prefrontal and parietal cortices.

Introduction

In the field of neuroimaging, the traditional focus has been on identifying "average" brain patterns to characterize healthy or pathological states. While useful, this approach often overlooks the significant individual variations that define personal identity. Recent research has established that the functional connectome—the map of neural communications—can serve as a reliable "fingerprint." However, previous identification methods typically required the entire connectome or relied on pre-defined anatomical networks.

This study addresses a fundamental question: Can we automatically identify the minimal set of connections that constitute an individual's unique neural signature without prior anatomical knowledge? Furthermore, can this signature be derived from a single imaging session?

Methodology

The researchers developed an unsupervised approach based on Randomized Numerical Linear Algebra, specifically utilizing Leverage Score Sampling. The process involves several key steps:

  1. Data Vectorization: Region-to-region correlation matrices (connectomes) from fMRI scans are transformed into a large matrix where rows represent subjects and columns represent individual connections (edges).
  2. Leverage Score Calculation: Using Singular Value Decomposition (SVD), the method calculates "statistical leverage scores" for every connection. High leverage scores indicate connections that are highly influential in distinguishing between individuals.
  3. Principal Features Subspace Method: Instead of analyzing the tens of thousands of possible connections, the method selects only the top $t$ connections with the highest leverage scores. This creates a "compact signature" that filters out common background noise shared across the population.

The study utilized the Human Connectome Project (HCP) dataset, involving 100 unrelated healthy adults, covering both resting-state and seven distinct task-based fMRI conditions.

Results: The Power of Compact Signatures

The most striking finding is the extreme compactness of the individual signature. Identification accuracy was found to plateau with as few as 61 to 95 connections—a tiny fraction of the 64,000+ possible edges in the full model.

Accuracy vs. Number of Features

As shown in the figure above, accuracy rises sharply and stabilizes early. The method achieved a 96% training accuracy and 93.11% test accuracy in matching individuals across different resting-state sessions. This performance significantly outperformed random feature selection (approx. 54% accuracy) and even slightly exceeded the accuracy of using the entire connectome (approx. 89%), likely due to the removal of non-discriminating noise.

Anatomical Localization and Task Consistency

The discriminating connections identified by the leverage score method are not randomly distributed across the brain. Instead, they are highly localized in specific "hotspots," primarily within the prefrontal and parietal cortices.

Anatomical hotspots of discriminating connections

These regions are associated with high-level executive functions, complex thought, and attention, suggesting that individual identity is rooted in the brain's most sophisticated processing centers. Furthermore, when applied to task-based fMRI, the "fingerprint" regions shifted predictably to match the functional areas relevant to the specific task (e.g., auditory and speech regions for language tasks).

Conclusion

This research demonstrates that individual neural signatures are both highly localized and extremely efficient. By identifying a compact set of fewer than 100 connections, researchers can identify individuals with high precision. While the current study is limited to the high-quality HCP dataset, it paves the way for more efficient neuroimaging analysis and a deeper understanding of the biological basis of individual identity. Future work will focus on testing these methods across more diverse datasets and investigating the minimum scan duration required to extract a stable signature.

Summary updated

Abstract

Recent neuroimaging studies have shown that functional connectomes are unique to individuals, i.e., two distinct fMRIs taken over different sessions of the same subject are more similar in terms of their connectomes than those from two different subjects. In this study, we present new results that identify specific parts of resting state and task-specific connectomes that are responsible for the unique signatures. We show that a very small part of the connectome can be used to derive features for discriminating between individuals. A network of these features is shown to achieve excellent training and test accuracy in matching imaging datasets. We show that these features are statistically significant, robust to perturbations, invariant across populations, and are localized to a small number of structural regions of the brain. Furthermore, we show that for task-specific connectomes, the regions identified by our method are consistent with their known functional characterization. We present a new matrix sampling technique to derive computationally efficient and accurate methods for identifying the discriminating sub-connectome and support all of our claims using state-of-the-art statistical tests and computational techniques.
BibTeX
@inproceedings{ravindra2021constructing,
  author = {Vikram Ravindra and Petros Drineas and Ananth Grama},
  title = {{Constructing Compact Signatures for Individual Fingerprinting of Brain Connectomes}},
  booktitle = {Frontiers in Neuroscience},
  year = {2021},
  doi = {10.3389/fnins.2021.549322},
}
Figure