Search papers, labs, and topics across Lattice.
This paper introduces a multi-stage AI pipeline for legal age estimation from clavicle CT scans, combining automated clavicle detection, Integrated Gradients-guided slice selection, and a multi-slice convolutional neural network. The pipeline incorporates conformal prediction intervals for uncertainty quantification, aligning with forensic standards. Evaluated on 1,158 post-mortem CT scans, the model achieves a state-of-the-art mean absolute error of 1.55 years, surpassing human experts and previous methods.
An AI model can estimate legal age from clavicle CT scans with higher accuracy than human experts, potentially revolutionizing forensic age assessment.
Legal age estimation plays a critical role in forensic and medico-legal contexts, where decisions must be supported by accurate, robust, and reproducible methods with explicit uncertainty quantification. While prior artificial intelligence (AI)-based approaches have primarily focused on hand radiographs or dental imaging, clavicle computed tomography (CT) scans remain underexplored despite their documented effectiveness for legal age estimation. In this work, we present an interpretable, multi-stage pipeline for legal age estimation from clavicle CT scans. The proposed framework combines (i) a feature-based connected-component method for automatic clavicle detection that requires minimal manual annotation, (ii) an Integrated Gradients-guided slice selection strategy used to construct the input data for a multi-slice convolutional neural network that estimates legal age, and (iii) conformal prediction intervals to support uncertainty-aware decisions in accordance with established international protocols. The pipeline is evaluated on 1,158 full-body post-mortem CT scans from a public forensic dataset (the New Mexico Decedent Image Database). The final model achieves state-of-the-art performance with a mean absolute error (MAE) of 1.55 $\pm$ 0.16 years on a held-out test set, outperforming both human experts (MAE of approximately 1.90 years) and previous methods (MAEs above 1.75 years in our same dataset). Furthermore, conformal prediction enables configurable coverage levels aligned with forensic requirements. Attribution maps indicate that the model focuses on anatomically relevant regions of the medial clavicular epiphysis. The proposed method, which is currently being added as part of the Skeleton-ID software (https://skeleton-id.com/skeleton-id/), is intended as a decision-support component within multi-factorial forensic workflows.
