Article

Contrast-Induced Class Overlap as a Fairness Bottleneck in Dermatological AI: Evidence from HAM10000

2026-06-25

Abstract excerpt

<title>Abstract</title> <p>AI-assisted skin-cancer triage systems, when calibrated for sensitivity, systematically over-predict on darker skin: at a fixed operating point we measure a specificity deficit of -11.0 pp on darker-skin patients (0.715 vs. 0.826 on a high-confidence skin-tone subset, permutation p < 0.0002, Cohen's h = 0.27), translating to an estimated ~89 excess unnecessary referrals per 1,000 darker...

Topics

Open a Topic to create a Post that cites this publication.

Identifiers and source

Literature Corpus work: 248306c4-dc1f-5186-9947-903b3571b18e
DOI: 10.21203/rs.3.rs-10132969/v1

Open publication

Related research

Semantic proximity does not establish scientific evidence.

Click a neighbor to travelStep 1 · 12 closest

Contrast-Induced Class Overlap as a Fairness Bottleneck in Dermatological AI: Evidence from HAM10000DOI 10.21203/rs.3.rs-10132969/v1Read

A Multimodal Large Reasoning Model For Fair and Interpretable Dermatological Diagnosis Across Skin TonesSemantic proximity Read

Evaluating Conversational Image Segmentation for Medicine: Performance, Failure Modes, and a Fairness Audit Across Seven ModalitiesSemantic proximity Read

A Framework for Evaluating the Efficacy of Foundation Embedding Models in HealthcareSemantic proximity Read

Multi-label transformer for dermoscopic artifact detection with an imbalance-aware and confidence-penalizing lossSemantic proximity Read

Bias-constrained multimodal intelligence for equitable and reliable clinical AISemantic proximity Read

Integrating Explainable AI for Skin Lesion Classifications: A Systematic Literature ReviewSemantic proximity Read

SkinLiTE: Lightweight Supervised Contrastive Learning Model for Enhanced Skin Lesion Detection and Disease Typification in Dermoscopic ImagesSemantic proximity Read

Case-level artificial intelligence for multi-photo teledermatology submissions: development and internal validation using patient-submitted dermatology imagesSemantic proximity Read

Multimodal Image Dataset for AI-based Skin Cancer (MIDAS) BenchmarkingSemantic proximity Read

QUANTIFYING EXPLAINABLE AI METHODS IN MEDICAL DIAGNOSIS: A STUDY IN SKIN CANCERSemantic proximity Read

DREAM: A framework for discovering mechanisms underlying AI prediction of protected attributesSemantic proximity Read

A Global Atlas of Digital Dermatology to Map Innovation and DisparitiesSemantic proximity Read