EGFDA: Experience-guided Fine-grained Domain Adaptation for cross-domain pneumonia diagnosis

Haoran Zhao, Tao Ren, Wei Li, Danke Wu, Zhe Xu

Abstract: Highlights•A novel experience-guided fine-grained domain adaptation framework is proposed to address lesion-preference bias and class-preference bias in cross-domain pneumonia diagnosis.•A gradient-aware lesion area matching mechanism is proposed to preserve lesion-related information for global domain matching.•A reweighing smooth certainty-aware matching mechanism is proposed to ensure the smoothness and reliability of class matching.