Abstract
Sub-Saharan Africa (SSA) has fewer than one dermatologist per million people. Community health workers photograph lesions and send them via WhatsApp, which degrades image quality before any AI sees it. Existing classifiers are forced-choice, conflating two distinct failures: a blurry photo needs a retake; an ambiguous case needs specialist referral. A third danger, secondary impetiginization, means steroids on an infected lesion cause active harm. We present TRIDENT, a 4 M-parameter multi-task EfficientNet-B0 that routes each image to {retake, treat, refer} via three lightweight heads for quality, uncertainty, and impetiginization risk. Quality is supervised for free via on-the-fly JPEG degradation. A safety gate forces referral whenever secondary infection is suspected. The FP16 ONNX model (8.1 MB) runs at a median of 73 ms (p95 76 ms) on a Samsung Galaxy A34; retake and treat require no server connectivity. On PASSION (Fitzpatrick III–VI, four SSA countries), TRIDENT matches the ResNet-50 baseline (subject BA 0.703 vs. 0.70) while adding calibrated, source-aware triage; impetiginization detection reaches F1 0.647 (AUROC 0.884). At the default threshold, 57% of encounters resolve offline at a cost: 25.8% of infections reach local treatment (τ_imp = 0.1 recovers 98.5% sensitivity). The retake action and safety gate cut clinical cost 12% over confidence-only deferral on test.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_025.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_025_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=BVdCkc6xCe
BibTex
@InProceedings{OlaOlu_TRIDENT_MICCAISAT2026,
author = { Oladepo, Oluwaferanmi AND Shiaki, Nunsi},
title = { { TRIDENT: Safety-Gated, On-Device Triage for Paediatric Teledermatology on Pigmented Skin } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17265},
month = {pending},
page = {pending}
}