Re-thinking Mammography Transfer Learning: The Dataset-Informed Transfer Learning (DITL) Framework for Breast Cancer Screening and Lesion Diagnosis

2026-07-28

AI Summary

This work addresses the limitations of existing transfer learning approaches in mammography classification, which often neglect dataset-specific characteristics and suffer from narrow task scope and poor scalability, thereby struggling to balance small lesion diagnosis with large-scale screening. To overcome these challenges, we propose a Dataset-aware Transfer Learning (DITL) framework that integrates dataset difficulty signals with neighborhood-aware triplet supervision. DITL introduces two adaptive loss functions—Adaptive Difficulty-Weighted Cross-Entropy (A-DWCE) and Adaptive Neighborhood Representation Triplet Loss (A-NR-Triplet)—which require no hyperparameter tuning and enable end-to-end optimization through learnable margins and sample weighting. Difficulty modeling is guided by k-nearest neighbor label purity estimated in a self-supervised feature space. Experiments demonstrate that DITL significantly improves breast density classification performance on VinDR-Mammo (p<0.0001) and consistently enhances accuracy across multiple small-scale lesion ROI datasets.

Abstract

Enhancing classification performance in mammography remains a persistent challenge across both small curated datasets and large-scale clinical cohorts. Conventional transfer learning approaches often neglect dataset-specific characteristics, while recent neighborhood-informed methods have been restricted to narrow tasks with rigid formulations, limiting their scalability to population-level datasets. To address these challenges, we propose the Dataset-Informed Transfer Learning (DITL) framework, which integrates dataset-derived difficulty signals with neighborhood-based triplet supervision in a unified objective. DITL introduces two adaptive components: (i) Adaptive Difficulty-Weighted Cross-Entropy (A-DWCE), which assigns per-sample weights based on k-nearest neighbor label purity in a self-supervised feature space, and (ii) Adaptive Neighborhood Representation Triplet (A-NR-Triplet), which enforces intra-class compactness and inter-class separation using a learnable margin. Unlike focal loss, DITL requires no hyperparameter tuning, removes heuristic weighting and fixed margins, and incurs negligible computational overhead, yielding a robust and scalable optimization strategy. On the large-scale VinDR-Mammo dataset, DITL achieves state-of-the-art performance for whole-image breast density classification, with significant improvements across accuracy, F1-score, and AUC (p < 0.0001). Beyond large cohorts, DITL also delivers consistent, statistically significant gains on small ROI datasets (p < 0.0001). By bridging small-scale lesion analysis with large-scale density estimation, DITL establishes a clinically relevant, scalable, and generalizable framework for mammography classification, spanning the full breast cancer screening-to-diagnosis spectrum.

Problem

• mammography
• transfer learning
• breast cancer screening
• lesion diagnosis
• dataset-specific characteristics

Innovation

• Dataset-Informed Transfer Learning
• Adaptive Difficulty Weighting
• Neighborhood Representation Triplet
• Mammography Classification
• Self-Supervised Feature Space