Fig. 2: Study design of habitat imaging–based radiomic analysis of NSCLC patients.

a Overview of the data collected and pre-processing stages of the CT and 18F-FDG PET images. The pre-processing involved 18F-FDG PET to CT registration and fusion of segmented tumor regions from 18F-FDG PET and CT images along with their local entropy maps. b The habitat imaging analysis framework consisted of a 2-stage clustering process: Individual-level clustering, where tumor regions of each patient were over-segmented into superpixels; and population-level clustering, where clustering was performed on superpixels pooled from all patients to discover distinct tumor subregions. The multiregional spatial interaction (MSI) matrix summarizes the co-occurrence statistics among different tumor subregions. The 92 MSI features extracted from the MSI matrix identified patient subtypes. c Radiogenomic analysis along with ctDNA metrics confirm the clinical and biological relevance of the identified imaging subtypes. Created with BioRender.com.