Predicting behavior from the multimodal profile of brain regions

根据大脑区域的多模态特征预测行为

• 批准号: 519107208
• 负责人: Dr. Sarah Genon, Ph.D.
• 金额: --
• 依托单位: Institut für Systemische Neurowissenschaften
• 依托单位国家: 德国
• 项目类别: Research Grants
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/519107208?language=en
• 关键词: Predicting; behavior; multimodal; profile; brain

Individual behavior prediction based on neuroimaging data and machine learning models have become increasingly popular. However, several challenges are present including the overall lower prediction accuracies in large samples, as well as the difficulty in interpreting the prediction models and outcomes. In this project, we address these challenges using our recently developed region-wise psychometric prediction framework with multimodal neuroimaging data. First, we aim to perform a systematic evaluation of multimodal psychometric prediction making use of multiple large datasets encompassing functional, structural and diffusion magnetic resonance imaging data. Then, we aim to improve model interpretability by specifically quantifying the contribution of neuroimaging features from each brain region and of each confounding variable. Overall, we propose to develop a multimodal psychometric prediction framework for improving predictive power beyond unimodal approaches, and for characterizing brain regions’ or confounding variables’ contribution to psychometric predictions.

基于神经成像数据和机器学习模型的个体行为预测越来越受到人们的欢迎。然而，也存在一些挑战，包括在大样本中总体预测精度较低，以及难以解释预测模型和结果。在这个项目中，我们使用我们最近开发的区域心理测量预测框架和多模式神经成像数据来解决这些挑战。首先，我们的目标是利用包含功能、结构和扩散磁共振成像数据的多个大型数据集，对多模式心理测量预测进行系统评估。然后，我们的目标是通过具体量化每个大脑区域和每个混杂变量的神经成像特征的贡献来提高模型的可解释性。总体而言，我们建议开发一个多模式心理测量预测框架，以提高预测能力，超越单峰方法，并表征大脑区域或混杂变量对心理测量预测的贡献。