A Machine Learning Framework for Accurate Functional Connectome Fingerprinting and an Application of a Siamese Network

Abstract

The goal of functional connectome (FC) fingerprinting is to uniquely identify subjects based on their functional connectome. In recent years, interest in this problem has increased substantially with efforts made to understand the factors that affect the accuracy of fingerprinting and to develop more effective approaches. In this work, we developed a novel machine learning framework for FC fingerprinting. Specifically, while existing approaches match a query FC with a reference FC based on a correlation score between the two FCs, our framework employed a machine learning model to determine if two FCs are similar. This allowed us to capture more complex features from FCs and also to capture non-linear similarities that may exist among FCs. We explored multiple machine learning algorithms that include a Siamese neural network and several classification algorithms. From our experiments, we observed that the Siamese network outperformed other classification models, with an FC fingerprinting accuracy of \(99.89\%\).