Really cool new paper by our group! Much like the fingerprint on our fingertips, our brain's structural connectome may also tell individuals apart. We take it one step further and show that it can also help single out individual cognitive performance 1/n
sciencedirect.com/science/articl…
sciencedirect.com/science/articl…
We first scanned adults on 3 separate occasions: at baseline, a separate session in the same scanner 126.4 (SD = 102.8, range 12–442) days after the first scan and another session in a completely different scanner 158.4 (SD = 103.6, range 21–465) days after the first scan. 2/n
We used diffusion-weighted images from these scans to derive a personalized structural connectome for each person in each session. The novelty was that we applied connectome dynamics to transform undirected region-to-region connections to a pair of hub-directed connections 3/n 
We trained a deep learning algorithm (optimized w/ grid search approach) on two-thirds of these data and tested its performance on the remaining third of the unseen data. Our approach was able to single out individuals with 93% accuracy! 4/n
We used the backtracking technique to identify which input connectome dynamic features had the greatest contribution to classification accuracy. 5/n
Using a model based on these top 16 subnetworks, we went one step further: we attempted to predict individual performance across the lifespan. For toddlers, we looked at predicting an Early Learning Composite (ELC), a composite of fine motor, visual reception, and language 6/n
Indeed, the connectome fingerprint model predicted individual ELC scores within 7.7 points of real performance with an r = 0.7 (compared to a random model, which predicted within 22.2 points with an r = 0.14) 7/n
In adults, we predicted IQ within 4.1 points of the individual performance score (r = 0.76) compared to a randomized model, which predicted within 24.3 (r = 0.1). 8/n
This was a study heavy on computational methods but with several important *practical* implications: 1) machine-learning can accurately tell individuals apart based solely on their brain structural connectome, easily derived from diffusion-weighted MR imaging; 9/n
2) connectome fingerprinting may not just accurately identify individuals but it may also provide a prediction of their individual scores on development/cognition/skills, both in toddlers and adults; 10/n
3) identifying the subnetworks that contribute to such performance prediction could shed light on biological mechanisms of developmental milestones and intelligence but also serve as potential targets for therapy (e.g. with neuromodulation). Thanks for reading! 11/11
@threadreaderapp unroll
