Overview
QADER, or Quick Age at Death Estimation using Radiographs, is an innovative AI-driven model for estimating age-at-death using single bone radiographs from archaeological contexts. Developed by Moustafa Abdalla and Katherine D. Van Schaik, this technology utilizes a deep learning convolutional neural network (CNN) to achieve accurate predictions of age-at-death, especially in the context of limited or damaged skeletal remains, offering a compelling and innovative way to enhance archaeologists’ demographic and paleopathological assessments.
Advantages
Versatility: QADER can predict age-at-death within 10 years from a single radiograph of long bones (e.g., humerus, pelvis, femur, tibia) with even when the epiphyses and metaphyses of those bones (the areas that are typically used for traditional age assessments) are damaged or absent.
Convolutional Neural Network (CNN): The technology leverages an 11-layer deep CNN, demonstrating its ability to capture spatial dependencies in bone radiographs and extract high-level features crucial for age prediction.
Accuracy: The current version of the model achieved a remarkable 94% accuracy in predicting the decade-of-death when utilizing all available bones for an individual. QADER maintains an 82% accuracy in predicting age-at-death from a single femoral radiograph (even if the femur is incomplete), showcasing its reliability and practicability in the common scenario in which only limited skeletal remains are present.
Applications
Archaeological Studies: QADER serves as a valuable support tool for bioarchaeologists, enabling precise age estimation even from incomplete skeletal remains.
Non-Destructive Age Estimation: The technology offers a non-destructive alternative to molecular age estimation methods, facilitating preservation efforts in archaeological contexts.
Future Directions
Generalizability: Ongoing research aims to assess the model's generalizability to diverse archaeological sites and contexts.
Confidence Calibration: Further developments include calibrating confidence intervals for age predictions and extending the model to provide numerical age-at-death predictions.