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

Meghdad Sabouri Rad; Junze; Huang; Mohammad Mehdi Hosseini; Rakesh Choudhary; Saverio J. Carello; Ola El-Zammar; Michel R. Nasr

cs.CVFeb 27, 2026

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

Authors: Meghdad Sabouri Rad, Junze, Huang, Mohammad Mehdi Hosseini, Rakesh Choudhary, Saverio J. Carello, Ola El-Zammar, Michel R. Nasr,

Organizations: Vincent · Department of Pathology, SUNY Upstate Medical University, Syracuse, NY, 13210, USA. · Columbia University, New York, NY, 10027, USA.

Abstract

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.

Figures & tables

Figure 1: Proposed attention-based robust histopathology subtyping framework integrating cross-entropy and contrastive learning with margin consistency. Whole-slide images (WSIs) are divided into fixed-size patches and normalized before passing through a patch encoder and an attention-based feature extractor $h_{\psi_{a}}(x)$ . The resulting feature representation $z$ is used for both classification via a linear classifier $f_{\theta_{a}}(x)=w^{\top}z+b$ and representation learning via supervised contrastive loss. Cross-entropy loss $\mathcal{L}_{CE}$ , contrastive loss $\mathcal{L}_{CON}$ , and Perturbation Fidelity loss $\mathcal{L}_{PF}$ are combined as total loss $\mathcal{L}_{T}$ . Margin consistency is computed using logit differences and feature margins to guide robust decision boundaries. This multitask learning scheme enhances feature discriminability, maintains high-margin consistency, and improves robustness against subtyping brittleness.

Model	$\mathcal{L}_{CE}$	$\mathcal{L}_{CE}+\mathcal{L}_{CON}$	$\mathcal{L}_{CE}+\mathcal{L}_{CON}+\mathcal{L}_{PF}$
ResNet50+Attn	90.63 $\pm$ 9.36	92.54 $\pm$ 6.90**	94.80 $\pm$ 6.12 ***
	[87.3, 93.9]	[89.8, 95.3]	[91.8, 97.8]
ViT-L	92.00 $\pm$ 5.36	93.78 $\pm$ 4.98*	95.20 $\pm$ 4.65 ***
	[89.1, 94.9]	[91.2, 96.4]	[92.6, 97.8]
ResNet101+Attn	91.73 $\pm$ 9.23	93.18 $\pm$ 4.51**	95.89 $\pm$ 5.37 ***
	[88.4, 95.0]	[90.5, 95.9]	[93.3, 98.5]

Table 1: Ablation analysis on BMIRDS-LUAD dataset with statistical significance. Values represent accuracy (%)

\pm

standard deviation. Statistical significance assessed via McNemar’s test comparing each configuration against CE-only baseline: *

p<0.05

, **

p<0.01

, ***

p<0.001

. 95% bootstrap confidence intervals shown in brackets.

† Re-implemented and evaluated on our dataset using identical protocols.
Method	Accuracy (%)	95% CI	$p$ -value
DeepSlide Wei et al. (2019) †	90.4	[87.1, 93.7]	$<$ 0.001***
Independent ResNet101 †	79.5	[75.2, 83.8]	$<$ 0.001***
Unsupervised Sheikh et al. (2022) †	94.6	[91.8, 97.4]	0.042*
Proposed (Ours)	95.89	[93.3, 98.5]	—

Table 2: Comparison of classification accuracy on the BMIRDS-LUAD dataset. All baseline methods were re-implemented and evaluated on our dataset using identical preprocessing, train/validation splits, and evaluation protocols to ensure fair comparison. Statistical significance assessed via McNemar’s test comparing each method against our proposed approach.

Architecture	Accuracy (%)	95% CI	$\Delta$ vs Internal	Domain Shift
ResNet50+Attn	80.1 $\pm$ 11.5	[74.2, 86.0]	$-$ 14.7%	Moderate
ResNet101+Attn	78.7 $\pm$ 8.3	[73.9, 83.5]	$-$ 17.2%	Moderate
ViT-L	73.1 $\pm$ 10.1	[67.4, 78.8]	$-$ 22.1%	Substantial

Table 3: Cross-dataset validation results on WSSS4LUAD benchmark demonstrating generalizability. Models trained on BMIRDS-LUAD and evaluated on WSSS4LUAD without fine-tuning. Performance degradation quantified relative to internal validation.

Figure 2: Domain shift analysis for external validation on WSSS4LUAD. (a) Per-subtype accuracy comparison. (b) Domain shift sources. (c) External confusion matrix. (d) Error sources by subtype.

Architecture	Baseline (%)	Our Method (%)	Improv. (%)
ResNet50+Attn	90.63±9.36	94.80±6.12	+4.17
ViT-L	92.00±5.36	95.20±4.65	+3.20
ResNet101+Attn	91.73±9.23	95.89±5.37	+4.16

Table 4: Statistical significance analysis of accuracy improvements with experimental results.

CE vs. CE+CON+PF (Full Method)
Comparison	McNemar’s $\chi^{2}$	$p$ -value	Cohen’s $d$	Effect Size
ResNet50+Attn	14.23	$<$ 0.001***	0.53	Medium
ViT-L	11.87	$<$ 0.001***	0.64	Medium
ResNet101+Attn	16.42	$<$ 0.001***	0.55	Medium
CE+CON vs. CE+CON+PF (PF Contribution)
ResNet50+Attn	5.12	0.024*	0.31	Small

Table 5: Statistical significance of accuracy improvements using McNemar’s test for paired classifier comparison. McNemar’s test is the appropriate statistical test for comparing classifiers evaluated on the same dataset, as it accounts for the paired nature of predictions and focuses on discordant cases Dietterich (1998) . Effect sizes quantified using Cohen’s

d

Note: Micropapillary merged with Papillary due to limited samples ( $n<10$ ).
Subtype	Sensitivity	Specificity	PPV	NPV	AUC	Fisher’s $p$
Solid	0.988	0.992	0.981	0.995	0.9996	$<$ 0.001***
Lepidic	0.964	0.997	0.986	0.992	0.9983	$<$ 0.001***
Acinar	0.989	0.988	0.982	0.993	0.9988	$<$ 0.001***
Papillary	0.979	0.996	0.989	0.993	0.9998	$<$ 0.001***

Table 6: Per-subtype discrimination performance for ResNet101 with attention mechanism with full loss (

\mathcal{L}_{CE}+\mathcal{L}_{CON}+\mathcal{L}_{PF}

). Statistical significance assessed using Fisher’s exact test comparing classification performance between baseline (CE-only) and proposed method.

Architecture	Baseline CI	Our Method CI	Var. Red.	Levene’s $p$
ResNet50+Attn	[87.3, 93.9]	[91.8, 97.8]	57.2%	0.003**
ViT-L	[89.1, 94.9]	[92.6, 97.8]	24.7%	0.041*
ResNet101+Attn	[88.4, 95.0]	[93.3, 98.5]	66.2%	$<$ 0.001***

Table 7: Bootstrap confidence intervals (95%, 1,000 iterations) and variance reduction analysis. Variance reduction calculated as

(\sigma^{2}_{\text{baseline}}-\sigma^{2}_{\text{ours}})/\sigma^{2}_{\text{baseline}}\times 100\%

. Statistical significance of variance reduction assessed via Levene’s test.

Figure 3: Bootstrap confidence intervals (95%) demonstrating accuracy improvements and enhanced model stability across architectures. Our method (green) shows consistently higher accuracy with reduced variance compared to baseline (red).

Architecture	Mean Accuracy (%)	Min-Max Range	CV (%)
ResNet50+Attention	94.80	[91.7, 97.9]	6.46
ViT-L	95.20	[92.6, 97.8]	4.88
ResNet101+Attention	95.89	[93.0, 98.8]	5.60

Table 8: Cross-validation stability analysis simulating multi-center deployment.

Architecture Type	Baseline	Our Method	$\Delta$ (%)
CNN (ResNet)	90.63-91.73	94.80-95.89	4.16-4.17
Transformer (ViT)	92.00	95.20	3.20
Overall Range	90.63-92.00	94.80-95.89	3.20-4.17

Table 9: Cross-architectural performance consistency analysis.

Histopathology Image Classification

Histopathology image classification uses computational models to identify tissue types, tumors, grades, and biomarkers from stained pathology images.

See all 139 papers

Attention-based multiple instance learning for predominant growth pattern prediction in lung adenocarcinoma wsi using foundation models
Apr 23, 2026Laura Valeria Perez-Herrera, M. J. Garcia-Gonzalez, Karen Lopez-LinaresComputational Pathology Pathology Foundation Models
* Vicomtech Foundation, Basque Research and Technology Alliance (BRTA) · ± eHealth Group, Bioengineering Area, Biogipuzkoa Health Research Institute
Multimodal Knowledge Distillation for Gastric Adenocarcinoma Classification from Whole-Slide Images
Oct 6, 2026Shrihari Dumbre, Bikash SantraCross-Modal Knowledge Distillation Multimodal Classification
Indian Institute of Technology, Jodhpur
Bag-of-Visual-Words for Spatial Mapping of Lung Adenocarcinoma Growth Patterns
Aug 5, 2026Darya Ardan, Valentin Oreiller, Henning MüllerComputational Pathology Oncology
Department of Computer Science, University of Geneva, Geneva, Switzerland · Informatics Institute, University of Applied Sciences Western Switzerland (HES-SO Valais), Sierre, Switzerland · Radiology Service, Medical Faculty, University of Geneva, Geneva, Switzerland

Apr 23, 2026J/K move · Enter open · S save

Attention-based multiple instance learning for predominant growth pattern prediction in lung adenocarcinoma wsi using foundation models

Laura Valeria Perez-Herrera, M. J. Garcia-Gonzalez, Karen Lopez-Linares

* Vicomtech Foundation, Basque Research and Technology Alliance (BRTA) · ± eHealth Group, Bioengineering Area, Biogipuzkoa Health Research Institute

Computational Pathology Pathology Foundation Models Oncology Multiple Instance Learning Whole-Slide Image Analysis

Open paper AskPDF arXiv

Lung adenocarcinoma (LUAD) grading depends on accurately identifying growth patterns, which are indicators of prognosis and can influence treatment decisions. Common deep learning approaches to determine the predominant pattern rely on patch-level classification or segmentation, requiring extensive annotations. This study proposes an attention-based multiple instance learning (ABMIL) framework to predict the predominant LUAD growth pattern at the whole slide level to reduce annotation burden. Our approach integrates pretrained pathology foundation models as patch encoders, used either frozen or fine-tuned on annotated patches, to extract discriminative features that are aggregated through attention mechanisms. Experiments show that fine-tuned encoders improve performance, with Prov-GigaPath achieving the highest agreement (\k{appa} = 0.699) under ABMIL. Compared to simple patch-aggregation baselines, ABMIL yields more robust predictions by leveraging slide-level supervision and spatial attention. Future work will extend this framework to estimate the full distribution of growth patterns and validate performance on external cohorts.

Abstract

Histopathology Image Classification

Explore similar work

Attention-based multiple instance learning for predominant growth pattern prediction in lung adenocarcinoma wsi using foundation models

Multimodal Knowledge Distillation for Gastric Adenocarcinoma Classification from Whole-Slide Images

Bag-of-Visual-Words for Spatial Mapping of Lung Adenocarcinoma Growth Patterns