Three-dimensional convolutional neural network for age-at-death estimation of deceased individuals through cranial computed tomography scans

Abstract

Accurate age-at-death (AAD) estimation is integral in establishing biological profiles in forensic anthropology, though standardized multivariate techniques are lacking. The current study developed and tested a three-dimensional convolutional neural network and three model variations with 1,224 de-identified cranial CT scans from the New Mexico Decedent Image Database. Each model required an input of an individual's cranial CT scan and outputted an AAD estimation. Model 3 was superior, successfully predicting AAD within 1.5 decades.