Brain maturation and aging on early-onset consumption of cocaine: Neurophysiological patterns extracted by convolutional neural networks.

DE LA FUENTE, LAURA ALETHIA

Virtual

Congreso; FLUX Society congress; 2020

FLUX

Cocaine is a dopamine (DA) reuptake inhibitor that can be smoked (Smoked Cocaine, SC) or insufflated (IC), giving a different pharmacological and subsequent behavioral pattern of consumption. SC reaches a concentration peak in blood and goes down faster than the sniffed cocaine (IC). This also motivates a different consumption behavior, in which IC is taken in doses separated by time and other activities, while SC users take as many small doses as they can to avoid the down dysphoric effects. Our work has shown that the administration route can drive interoceptive differences in consumers, involving insular adaptations. And drive different neuropsychological features, particularly related to the attentive-executive function domain related to fronto-striatal structural and functional patterns. These regions are in ongoing development during adolescence. Perion in which the DA system maturation triggers critical neoplastic events. On the other hand, accelerated aging in stimulant consumption has been previously reported. Drug exposure under adolescence can potentially accelerate/alter developmental process, resulting in developmental-aging process in stimulant users, that can depend upon the administration route. In the present work, we trained a deep convolutional neural networks (CNN) to predict age over a large database of control brain structural images. With our methodology age prediction over test images improves literature showing mean errors below 2 years. We have also extracted maps of the relevant features via "heat maps" and integrated gradients images over test and a new validation sample. We found that previously reported relevant regions to predict age are captured (both subcortical and cortical). Regional importance shows an age relation importance, with subcortical regions picking over 20s. After training, we evaluated the network in a group of 25 SC dependents, 22 CC dependent, and 25 CTR matched by age, gender, education, and SES. We found that young (below 20s) consumers (independent from the administration route) show a tendency to aging features. Future work will focus on: improving the training to reduce the error over controls, elucidating regions and texture patterns that predict age in control samples and finally describe the features that account for the tendency to aging in consumers.