Forensic human identification using smile photographs and digital dental casts: A semi-quantitative superimposition technique combined with deep learning

Background: Three-dimensional shape of the teeth, missing dentition, pattern of smile lines, and various morphological developmental anomalies make the whole dentition unique for each individual. In the absence of ante-mortem clinical dental data, non-clinical digital smile photographs with visible dentition sourced from social network sites or next of kin may serve as an important tool for forensic dental identification. Methods: In this paper, we present a semi-automated method of forensic identification of humans that uses deep learning for extraction of visible dentition from smile photographs. A novel semi-quantitative method was developed for comparing these photographs over digital dental casts using computerised superimposition and metric analysis of six unique dental landmarks. The proposed method was validated using a paired dataset containing fifty 2D- smile photographs captured using a digital camera (simulated antemortem data) and fifty 3D-digital dental casts (simulated postmortem data). Results: Among all the unique features, we found that the smile line in a digital photograph of an individual aligns with the visible anterior dentition in 3D-digital dental casts without statistically significant inter-arch mismatch. Conclusion: Hence we propose a method indicating the utilisation of smile photographs as a strong forensic identifier when combined with machine learning, computerised superimposition and semi-quantitative metric analysis of interarch distances.”