# Margin-Consistent Deep Subtyping of Invasive Lung Adenocarcinoma via Perturbation Fidelity in Whole-Slide Image Analysis

arXiv: https://arxiv.org/abs/2603.06650 · HTML: https://mlsanity.com/paper/2603.06650/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 19 of its 33 references that are in the map.
- First posted on arXiv 27 Feb 2026.

## Builds on (most cited first, 19 of 19)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [A Simple Framework for Contrastive Learning of Visual Representations](https://mlsanity.com/paper/2002.05709/) (2020)
- [Momentum Contrast for Unsupervised Visual Representation Learning](https://mlsanity.com/paper/1911.05722/) (2019)
- [Towards Evaluating the Robustness of Neural Networks](https://mlsanity.com/paper/1608.04644/) (2016)
- [DeepFool: a simple and accurate method to fool deep neural networks](https://mlsanity.com/paper/1511.04599/) (2015)
- [Prevalence of Neural Collapse during the terminal phase of deep learning training](https://mlsanity.com/paper/2008.08186/) (2020)
- [Overfitting in adversarially robust deep learning](https://mlsanity.com/paper/2002.11569/) (2020)
- [Geometry-aware Instance-reweighted Adversarial Training](https://mlsanity.com/paper/2010.01736/) (2020)
- [Large Margin Deep Networks for Classification](https://mlsanity.com/paper/1803.05598/) (2018)
- [Probabilistic Margins for Instance Reweighting in Adversarial Training](https://mlsanity.com/paper/2106.07904/) (2021)
- [Pathologist-level classification of histologic patterns on resected lung adenocarcinoma slides with deep neural networks](https://mlsanity.com/paper/1901.11489/) (2019)
- [Multi-Layer Pseudo-Supervision for Histopathology Tissue Semantic Segmentation using Patch-level Classification Labels](https://mlsanity.com/paper/2110.08048/) (2021)
- [WSSS4LUAD: Grand Challenge on Weakly-supervised Tissue Semantic Segmentation for Lung Adenocarcinoma](https://mlsanity.com/paper/2204.06455/) (2022)
- [Exploring and Exploiting Decision Boundary Dynamics for Adversarial Robustness](https://mlsanity.com/paper/2302.03015/) (2023)
- [Understanding the Decision Boundary of Deep Neural Networks: An Empirical Study](https://mlsanity.com/paper/2002.01810/) (2020)
- [Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers](https://mlsanity.com/paper/2406.18451/) (2024)
- [Double InfoGAN for Contrastive Analysis](https://mlsanity.com/paper/2401.17776/) (2024)
- [Dynamic Label Adversarial Training for Deep Learning Robustness Against Adversarial Attacks](https://mlsanity.com/paper/2408.13102/) (2024)
- [Semi-Supervised Contrastive VAE for Disentanglement of Digital Pathology Images](https://mlsanity.com/paper/2410.02012/) (2024)

## Abstract (arXiv)

Whole-slide image classification for invasive lung adenocarcinoma subtyping remains vulnerable to real-world imaging perturbations that undermine model reliability at the decision boundary. We propose a margin consistency framework evaluated on 203,226 patches from 143 whole-slide images spanning five adenocarcinoma subtypes in the BMIRDS-LUAD dataset. By combining attention-weighted patch aggregation with margin-aware training, our approach achieves robust feature-logit space alignment measured by Kendall correlations of 0.88 during training and 0.64 during validation. Contrastive regularization, while effective at improving class separation, tends to over-cluster features and suppress fine-grained morphological variation; to counteract this, we introduce Perturbation Fidelity (PF) scoring, which imposes structured perturbations through Bayesian-optimized parameters. Vision Transformer-Large achieves 95.20 +/- 4.65% accuracy, representing a 40% error reduction from the 92.00 +/- 5.36% baseline, while ResNet101 with an attention mechanism reaches 95.89 +/- 5.37% from 91.73 +/- 9.23%, a 50% error reduction. All five subtypes exceed an area under the receiver operating characteristic curve (AUC) of 0.99. On the WSSS4LUAD external benchmark, ResNet50 with an attention mechanism attains 80.1% accuracy, demonstrating cross-institutional generalizability despite approximately 15-20% domain-shift-related degradation and identifying opportunities for future adaptation research.