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

2026-03-10 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Digital Pathology · Depth: Advanced, extended

Summary

A new margin consistency framework significantly improves the robustness and accuracy of deep learning models for invasive lung adenocarcinoma subtyping from whole-slide images (WSIs). Evaluated on 203,226 patches from 143 WSIs across five subtypes in the BMIRDS-LUAD dataset, the approach combines attention-weighted patch aggregation with margin-aware training. It achieves robust feature-logit space alignment, with Kendall correlations of 0.88 during training and 0.64 during validation. The framework introduces Perturbation Fidelity (PF) scoring, which uses Bayesian-optimized structured perturbations to counteract the over-clustering effect of contrastive regularization, preserving fine-grained morphological distinctions. Vision Transformer-Large achieved 95.20% accuracy, a 40% error reduction, while ResNet101 with attention reached 95.89% accuracy, a 50% error reduction, both outperforming a 92.00% baseline. All five subtypes exceeded ROC-AUCs of 0.99. Cross-institutional validation on the WSSS4LUAD benchmark showed 80.1% accuracy with ResNet50, demonstrating generalizability despite a 15-20% domain shift.

Key takeaway

For research scientists developing robust AI for digital pathology, this framework offers a significant advancement in lung adenocarcinoma subtyping. Your models can achieve higher accuracy and substantially reduced variance by integrating attention-enhanced margin consistency with Perturbation Fidelity. Consider adopting this multi-loss approach and Bayesian optimization for hyperparameter tuning to improve prediction reliability and generalizability across diverse clinical imaging conditions, crucial for real-world deployment.

Key insights

Margin consistency with Perturbation Fidelity enhances deep learning robustness and accuracy for lung adenocarcinoma subtyping.

Principles

• Attention mechanisms increase decision margins by down-weighting noise.
• Perturbation Fidelity preserves fine-grained morphological distinctions.
• Margin-aware training improves robustness against decision boundary brittleness.

Method

The method combines cross-entropy, supervised-contrastive loss, and Perturbation Fidelity loss, weighted by margin-aware sampling. Bayesian optimization tunes perturbation parameters, and attention-weighted aggregation forms slide-level features.

In practice

• Use attention-based pooling for WSI feature aggregation.
• Apply Perturbation Fidelity to prevent feature over-clustering.
• Employ margin-aware weighting to focus training on brittle cases.

Topics

• Digital Pathology
• Lung Adenocarcinoma Subtyping
• Margin Consistency
• Perturbation Fidelity
• Whole-Slide Imaging

Best for: Computer Vision Engineer, Research Scientist, AI Researcher, AI Scientist, Deep Learning Engineer

Related on AIssential

• Implementing trust in non-small cell lung cancer diagnosis with a conformalized uncertainty-aware AI framework — TRUECAM enhances pathology AI trustworthiness by integrating uncertainty quantification, data filtering, and error rate control.
• Generative Deep Learning for Computational Destaining and Restaining of Unregistered Digital Pathology Images — Preprocessing-based domain adaptation enables cGAN models to generalize for computational pathology across institutions without retraining.
• When is 3D Worth It? A Resource-Performance Frontier for CNNs and Transformers in Lung CT — Input dimensionality significantly impacts model performance and stability in volumetric medical imaging.
• Trusting Right Predictions for Wrong Reasons: A LIME Based Analysis of Deep Learning Interpretability in Lung Cancer Diagnosis — Deep learning models can achieve high predictive accuracy in medical imaging for different reasons, requiring independent interpretability validation.
• Align then Refine: Text-Guided 3D Prostate Lesion Segmentation — Text-guided, phase-scheduled training improves 3D prostate lesion segmentation by integrating semantic alignment and confidence-aware refinement.
• BreastGPT: A Multimodal Large Language Model for the Full Spectrum of Breast Cancer Clinical Routine — Workflow-aligned data and cross-scale visual modeling are critical for clinically grounded medical MLLMs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.