CoFiDA-M: Concept-Aware Feature Modulation for Cross-Domain Adaptation with Image-Only Inference

2026-05-29

AI Summary

This work addresses the significant performance degradation of AI-based skin cancer screening models when transferring from expert dermoscopic images to consumer-grade clinical photographs, a challenge exacerbated by the inaccessibility of clinical semantic metadata during testing. To overcome this, the authors propose CoFiDA-M, a novel framework that leverages probabilistic clinical concepts predicted by the foundation model MONET as privileged information during training. A teacher network utilizes these concepts to guide a FiLM modulator in constructing a semantically enriched feature space, which is then transferred via feature-level knowledge distillation to a student model that relies solely on image inputs at test time. Evaluated across multiple datasets, CoFiDA-M consistently outperforms existing domain adaptation methods—particularly achieving superior melanoma recall—while demonstrating strong generalization and practical deployment potential without requiring test-time metadata.

Abstract

Models for AI-based skin cancer screening suffer a severe performance drop when shifting from expert dermoscopic (source) images to consumer-grade clinical (target) images, hindering real-world deployment. Existing domain adaptation methods often ignore crucial semantic invariants, such as clinical concepts. While new foundation models like MONET can provide this semantic information as dense, probabilistic scores, this metadata is unavailable at test time, creating a deployment paradox for practical image-only screening tools. We address this gap by proposing CoFiDA-M, a privileged information framework that learns from concepts at training time but deploys as an image-only model. Our method trains a teacher network that uses MONET concept probabilities to guide a FiLM modulator, transforming visual features into a semantically ``edited" feature space. A lightweight, image-only student is then trained to reproduce this edited representation, not just the teacher's final predictions. This distillation ``bakes" the clinical reasoning into the student's weights. On a challenging multi-dataset benchmark, our image-only student significantly outperforms state-of-the-art approaches, especially in melanoma recall. Our work provides a practical and generalizable framework for leveraging noisy, probabilistic metadata as privileged information, demonstrating strong cross-dataset robustness and potential for real-world deployment beyond dermatology. Implementation code is available at: https://github.com/mmu-dermatology-research/CoFiDA.git

Problem

• domain adaptation
• skin cancer screening
• privileged information
• clinical concepts
• image-only inference

Innovation

• Concept-Aware Feature Modulation
• Cross-Domain Adaptation
• Privileged Information Distillation
• Image-Only Inference
• Clinical Concept Guidance