Abstract
We present a per-patient probabilistic digital twin for adverse-event risk in aesthetic medicine. By registering facial landmarks via thin-plate splines, a generic 3D anatomical template is warped to individual patient anatomy, carrying per-zone Bayesian posteriors over complication rates. Given a proposed injection plan, the twin returns a posterior-predictive distribution over complication counts, operating as an uncertainty-aware world model for outpatient procedures. Constructing this twin is constrained by severe data scarcity: critical filler complications (vascular occlusion, vision loss, infection, nodules, bruising) are rare at single practices, while facial photographs are protected biometric identifiers under HIPAA and GDPR that cannot be centralized. We propose BFed-RDP, a federated Bayesian protocol where distributed clinics jointly refine per-zone Gamma-Poisson posteriors by exchanging Rényi-DP natural parameters. On a manufacturer partition of the FDA MAUDE database ($K{=}8$ sites, $Z{=}5$ complication classes, 552 records), BFed-RDP at $(\varepsilon{\approx}2.09, \delta{=}10^{-5})$ achieves a Poisson score of $-11.61$ per test event versus $-18.61$ for FedAvg$+$DP, with Fed-Heart-Disease ($K{=}4$ hospitals) as an out-of-domain benchmark. Because MAUDE lacks procedure denominators, outputs represent relative rate orderings rather than absolute risks. While posteriors remain well calibrated on synthetic data and non-private baselines, calibration degrades under strict privacy budgets on real data, a trade-off we report explicitly.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MWM_121.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MWM_121_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=W8PGYgs2OU
BibTex
@InProceedings{RijDam_AFederated_MICCAISAT2026,
author = { Rijhwani, Damini},
title = { { A Federated Probabilistic Digital Twin for Adverse-Event Risk in Aesthetic Medicine } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17262},
month = {pending},
page = {pending}
}