Orthogonalized Kernel Debiased Machine Learning for Multimodal Data Analysis

Multimodal imaging has transformed neuroscience research. While it presents unprecedented opportunities, also imposes serious challenges. Particularly, is difficult to combine the merits of interpretability attributed a simple association model with flexibility achieved by highly adaptive nonlinear model. In this article, we propose an orthogonalized kernel debiased machine learning approach, which built upon Neyman orthogonality and form decomposition orthogonality, for multimodal data analysis. We target setting that naturally arises in almost all studies, where there primary modality interest, plus additional auxiliary modalities. establish root-$N$-consistency asymptotic normality estimated parameter, semi-parametric estimation efficiency, validity confidence band predicted effect. Our proposal enjoys, good extent, both flexibility. It considerably different from existing statistical methods integration, as well orthogonality-based high-dimensional inferences. demonstrate efficacy our method through simulations application neuroimaging study Alzheimer's disease.