AI-powered interpretable imaging phenotypes noninvasively characterize tumor microenvironment associated with diverse molecular signatures and survival in breast cancer

Tianxu Lv, Xiaoyan Hong, Yuan Liu, Kai Miao, Heng Sun, Lihua Li, Chuxia Deng, Chunjuan Jiang, Xiang Pan

Abstract: Highlights•An artificial intelligence (AI)-driven computational pipeline is proposed to characterize tumor microenvironment (TME).•We propose signature-driven spatial-kinetic voxel subcluster graph (SigN-TG) to construct intratumoral cluster graphs.•We devise interpretable imaging phenotypes (IMPs) to measure morphology, interaction, and proximity between the subclusters.•Experiments demonstrate that IMPs can effectively predict distinct molecular signatures.•Our method provides an interpretable perspective to characterize TME in a noninvasive and clinically relevant manner.