---
title: "Dataset-Informed Transfer Learning (DITL) Framework for Mammography: A Unified Optimization for Breast Cancer Screening and Lesion Diagnosis"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.26043v1"
permalink: "https://feed.astralor.com/items/itm_cf4c4fc54c51248c2394644c"
published: "2026-07-28T17:52:36.000Z"
score: 74
channel: "tech"
tags: ["Dataset-Informed Transfer Learning","VinDR-Mammo","Mammography Classification","Transfer Learning"]
---

# Dataset-Informed Transfer Learning (DITL) Framework for Mammography: A Unified Optimization for Breast Cancer Screening and Lesion Diagnosis

## AI commentary

> What stands out in this paper is not just the state-of-the-art results on a specific task, but the elegant way it resolves two pain points in transfer learning: hyperparameter dependence and scalability across dataset sizes. For researchers and engineers working on clinical AI deployment, DITL offers a plug-and-play optimization strategy that is particularly suited for real-world settings with heterogeneous data quality.

## At a glance

Conventional transfer learning for mammography ignores dataset-specific characteristics, while recent neighborhood-informed methods are limited to narrow tasks. The proposed Dataset-Informed Transfer Learning (DITL) framework integrates adaptive difficulty-weighted cross-entropy and neighborhood-based triplet loss, achieving significant improvements on both large-scale VinDR-Mammo and small ROI datasets without hyperparameter tuning, bridging the screening-to-diagnosis spectrum.

## Key points

- DITL framework with Adaptive Difficulty-Weighted Cross-Entropy (A-DWCE) and Adaptive Neighborhood Representation Triplet (A-NR-Triplet).
- Hyperparameter-free, negligible computational overhead, scalable from small to large datasets.
- Achieves state-of-the-art whole-image breast density classification on VinDR-Mammo with significant improvements in accuracy, F1, and AUC (p<0.0001).
- Delivers consistent gains on small ROI datasets, covering the full screening-to-diagnosis spectrum.

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_cf4c4fc54c51248c2394644c