FaceTTD: Time-to-Death from Imageomic Facial Time Series for Rapid Mortality Risk Profiling and Longevity Interventions

Mortality risk assessment remains a fundamental challenge in healthcare, with current methods relying on chronological age that fails to capture individual variation in biological aging and proximity to death. Direct prediction of time-to-death from accessible, non-invasive phenotypic signals could enable more precise aging risk stratification and targeted longevity interventions. We present *FaceTTD*, a framework for predicting time-to-death (TTD) from facial images as a measure of biological aging. Treating portraits as time series inputs, we train XGBoost and Random Forest regressors on curated IMDB and Wikipedia datasets. In-distribution performance reaches $R^2 = 0.67$, but falls to $R^2 = 0.12$--$0.25$ out-of-distribution depending on TTD subset. Longitudinal facial trajectories improve predictive accuracy, indicating value in temporal coverage. Our findings highlight the promise and limitations of mortality modeling from phenotypic time series, positioning mortality horizon estimation as an imageomics problem where facial trajectories serve as accessible phenotypes of biological aging, and motivating multimodal extensions (voice, video, wearables, EHRs) for robust health applications. A live demo is available at https://huggingface.co/spaces/doubleblindanonymous/facettd.