B1604
Title: Advanced machine learning methods for retinal imaging genetics
Authors: Wei Chen - University of Pittsburgh (United States) [presenting]
Abstract: Age-related macular degeneration (AMD) is a multifactorial irreversible retina disease and the leading cause of blindness in the developed world. The combination of wealthy genetics, fundus image data, and well-characterized clinical phenotypes provides unprecedented opportunities to explore the new retinal imaging genetics concept. The advantage is from recent advances in deep learning methods and large-scale imaging genetics datasets for image grading, disease prediction, and disease trajectory inference. Two recent methods toward the overarching goals are discussed: (1) a new temporal-correlation-structure-guided generative adversarial network model for simultaneously grading the current fundus image and predicting the longitudinal disease severity and (2) a multi-modal genotype and phenotype mutual learning for enhancing single-modal based disease prediction. The experiments on the large-scale imaging genetics dataset with validations in independent cohorts demonstrate the superiority of the model compared to baselines for simultaneously grading and predicting future AMD severity of subjects.