Abstract The structural network damages in amyotrophic lateral sclerosis patients are evident but contradictory due to the high heterogeneity of disease. We hypothesized that patterns of structural network impairments would be different in amyotrophic lateral sclerosis subtypes by a data-driven method using 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET)/magnetic resonance (MR) hybrid imaging. The data of PET, structural MRI and diffusion tensor imaging in fifty patients with amyotrophic lateral sclerosis and twenty-three healthy controls were collected by a 18F-FDG PET/MR hybrid. Two amyotrophic lateral sclerosis subtypes were identified as the optimal cluster based on gray matter volume and standardized uptake value ratio. Network metrics at the global, local and connection levels were compared to explore the impaired patterns of structural network in the identified subtypes. Compared with healthy controls, the two amyotrophic lateral sclerosis subtypes displayed a pattern of a locally impaired structural network centralized in the sensorimotor network and a pattern of an extensively impaired structural network in the whole brain. When comparing the two amyotrophic lateral sclerosis subgroups by a support vector machine classifier based on the decreases in nodal efficiency of structural network, the individualized network scores were obtained in every amyotrophic lateral sclerosis patient and demonstrated a positive correlation with disease severity. We clustered two amyotrophic lateral sclerosis subtypes by a data-driven method, which encompassed different patterns of structural network impairments. Our results imply that amyotrophic lateral sclerosis may possess the intrinsic damaged pattern of white matter network and thus provide a latent direction for stratification in clinical research.