中国科学院自动化研究所机构知识库

青少年是人类发育过程中从儿童到成人、从依赖到独立的一个重要过渡阶段，伴随着身体、神经、心理、认知和社会角色等的变化。青少年时期也是多种精神疾病的高发期，而这些精神疾病的发生与异常的脑发育有关。对于青少年大脑、行为与认知发育的研究有助于个性化教育方案的制定和异常发育的早期干预。近年来，非侵入式磁共振成像技术的快速发展推动了青少年脑发育及其与行为和认知发育之间关系的研究。但是当前关于青少年发育的研究大多利用横断面数据在组平均水平上进行分析，忽略了青少年发育的个体差异。针对这一局限，本文利用大样本的青少年纵向数据探究了青少年个体脑发育及其与行为和认知发育的关系。本论文的主要工作和创新点归纳如下：

1. 青少年脑结构和脑功能的个体差异及发育研究

利用一批包含14岁和19岁磁共振成像的纵向数据，系统地研究了青少年脑结构和脑功能的个体差异及发育，其中脑结构指标包括灰质体积、皮层表面积、平均曲率和皮层厚度，脑功能指标为功能连接。结果发现，14岁和19岁功能连接的个体差异存在相似的层级结构，即在高级皮层个体差异较大，而在初级皮层个体差异较小。该模式与灰质体积的个体差异模式呈显著正相关。从14岁到19岁，功能连接的个体差异整体增加，而各脑结构指标的个体差异整体降低。此外，功能连接从14岁到19岁变化的模式与皮层厚度变化的模式显著相关。综上，该研究揭示了青少年脑结构和功能个体差异及发育的规律，并且探究了脑结构和脑功能的关系，为研究青少年个体行为和认知的发育提供了基础。

2. 基于脑灰质体积的青少年身高体重比及其发育的个体化预测研究

基于8个采集中心的青少年纵向数据以及多变量的机器学习方法，利用脑灰质体积作为特征，构建了青少年身高体重比及其发育的预测模型。在14岁和19岁，基于体素的全脑灰质体积可以显著预测个体的身高体重比，且模型能够在8个独立中心数据集上进行泛化。14岁和19岁在预测中起作用的脑区有一定的差别。从14 岁到19岁，基于体素的小脑灰质体积的变化可以预测个体身高体重比的变化，且模型可以在7个独立中心数据集上进行泛化。横断面的预测中，基于全脑特征的预测模型性能优于基于局部大脑特征的模型，而在纵向预测中则相反。综上，该研究找到了能够预测青少年个体身高体重比及其发育的脑区，可能为青少年肥胖的早期干预提供帮助。此外，该研究发现了纵向研究中局部大脑信息相对于全脑信息的优势，为未来的纵向研究提供了重要参考。

3. 基于脑功能连接的青少年抑制控制能力发育的个体化预测研究

基于大样本的青少年纵向数据，找到了预测青少年抑制控制能力个体化发育的早期神经预测因子。从14岁到19岁，抑制控制能力的发育存在个体差异。利用数据驱动的方法，发现14岁腹侧注意力网络与皮下核团之间的功能连接可以在个体水平预测5年后抑制控制能力的变化。该预测模型在不同的脑区划分以及不同的功能连接计算方式上均可重复，且在独立样本上具有良好的泛化性能。此外，药物滥用是一种抑制控制能力缺陷有关的行为障碍。结果发现，预测模型中所识别的功能连接特征还与未来5年内的药物滥用情况有关。综上，该研究揭示了青少年中期到晚期抑制控制能力发育的个体差异，并且找到了该发育的早期神经预测因子，为青少年抑制控制能力缺陷及其相关的行为障碍的早期干预提供了神经依据。

Adolescence, the transitional stage of life from childhood dependence on a caregiver to adult independence, accompanied with profound body, neural, psychological, cognitive and social change. Adolescence is also a time of increasing incidence of several psychiatric illnesses, which are thought to be linked to abnormal brain development. Studies on the development of adolescent brain and behavior can help to tailor more reasonable and individualized educational schemes, and provide opportunities for early recognition and intervention for abnormal development. In recent years, the fast development of non-invasive magnetic resonance imaging (MRI) technology have advanced the progress on the studies of brain and behavior development during adolescence. However, most previous studies on adolescence development mainly focused on cross-sectional data and group-average methods, which obscured the meaningful individual differences in development. In view of this situation, large-scale longitudinal datasets of adolescents are used to study the individual differences of brain development and its relationship with behavioral development during adolescence. The main contributions are summarized as follows:

1. Individual differences and development of adolescent brain structure and function across the cortex

Using a longitudinal adolescent dataset, which acquired MRI data at 14 and 19 years old, this study systematically examined the individual variations of brain structure and function, as well as their development during adolescence. The brain structure information includes gray matter volume, surface area, surface mean curvature and surface thickness, while the functional information used here was functional connectivity. We found a higher individual variation in heteromodal association cortex, and a lower inter-subject variability within the unimodal sensory and motor cortices at both 14 and 19 years old. This pattern was positively correlated with the individual differences of gray matter volume. From 14 to 19 years old, the overall inter-subject variability of brain function increased, while the overall individual differences of brain structure decreased. Moreover, from age 14 to age 19, the development of brain function was significantly correlated with the development of surface thickness. In conclusion, this study revealed the individual differences and development of adolescent brain structure and function, and also studied the relationship between brain structure and function. These provided foundation for studying the individualized development of behaviors and cognitive abilities in adolescents.

2. Gray matter volume predicts individual body mass index and its development during adolescence

Utilizing the longitudinal adolescent dataset from 8 acquisition sites and the multivariate machine learning method, we constructed predictive models for predicting individual body mass index and its development using gray matter volume as features. We found that the whole brain voxel-wise gray matter volume could predict the body mass index at the individual level at both 14 and 19 years old, and these predictive models could be generalized well to each of the 8 sites. The contributions of each brain region had considerable differences between the baseline and follow-up predictions. Moreover, the voxel-wise relative development of cerebellum could predict the relative development of body mass index individually, and this predictive model generalized well in 7 of the 8 sites. In the cross-sectional predictions, the prediction based on whole-brain gray matter volume performed better than that based on regional gray matter volume, whereas in the longitudinal prediction the cerebellum gray matter volume gave better prediction accuracy than that based on whole-brain gray matter volume. This study found brain regions playing an important role in predicting individual body mass index and its development during adolescence, which may help guide early preventions and interventions for obesity in adolescents. Moreover, our results implied the advantages of local brain information in longitudinal studies, which could provide insights for future longitudinal studies.

3. Functional connectivity predicts individual development of inhibitory control during adolescence

Based on a large-scale longitudinal adolescent dataset, we found the early neural predictors for the individualized development of inhibitory control during adolescence. From 14 to 19 years old, the participants had distinct between-subject trajectories in their inhibitory control ability. Utilizing the data-driven method, we found that the 14-year-old functional connectivity pattern between the ventral attention network and the subcortical network could reliably predict the development of inhibitory control at the individual level. This prediction model was reproducible with different methods of brain division and functional connectivity construct, and could be generalized to previously unseen individuals. Moreover, substance abuse is a maladaptive behavior related to deficit of inhibitory control, we found that the functional connections that could predict the development of inhibitory control were also related to substance abuse problems within 5 years. This study revealed the individual differences in inhibitory control development from mid- to late-adolescence, and found the neural predictor for it. This may provide the neural evidence for the early intervention for the deficit of inhibitory control and its related maladaptive behaviors in adolescents.