ROFI: A Deep Learning-Based Ophthalmic Sign-Preserving and Reversible Patient Face Anonymizer

2025-10-13

AI Summary

This study addresses the challenge of simultaneously preserving patient facial privacy and retaining diagnostically critical ophthalmic features in clinical imaging. We propose a deep learning framework integrating weakly supervised learning with invertible neural identity transformation. The method anonymizes facial identity while faithfully preserving key ophthalmic signs—including conjunctival hyperemia, scleral icterus, and eyelid edema—enabling secure image reconstruction for medical auditing and longitudinal follow-up. Crucially, the framework is plug-and-play compatible with existing AI diagnostic systems without requiring downstream model modification. Evaluated across three independent cohorts and 11 ocular diseases, it achieves >95% successful anonymization, 100% diagnostic sensitivity, excellent diagnostic agreement with original images (Cohen’s κ > 0.80), and >98% perceptual similarity in reconstructed images. The core innovation lies in the joint optimization of reversible identity erasure and lossless preservation of clinically relevant anatomical and pathological features.

Technology Category

• Computer Vision: Bias, Fairness & Privacy
• Machine Learning: Privacy
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Security and Privacy: Data transparency and provenance
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Patient face images provide a convenient mean for evaluating eye diseases, while also raising privacy concerns. Here, we introduce ROFI, a deep learning-based privacy protection framework for ophthalmology. Using weakly supervised learning and neural identity translation, ROFI anonymizes facial features while retaining disease features (over 98% accuracy, $κ> 0.90$). It achieves 100% diagnostic sensitivity and high agreement ($κ> 0.90$) across eleven eye diseases in three cohorts, anonymizing over 95% of images. ROFI works with AI systems, maintaining original diagnoses ($κ> 0.80$), and supports secure image reversal (over 98% similarity), enabling audits and long-term care. These results show ROFI's effectiveness of protecting patient privacy in the digital medicine era.

Problem

• Anonymizes facial features while preserving eye disease characteristics
• Maintains diagnostic accuracy across multiple ophthalmic conditions
• Enables secure image reversal for medical audits and care

Innovation

• Uses weakly supervised learning for anonymization
• Employs neural identity translation to preserve features
• Supports secure image reversal for audit purposes