Artificial intelligence application in lymphoma diagnosis with Vision Transformer using weakly supervised training

Alex Banerjee; Amer Wahed; Andy Quesada; Hanadi El Achi; Jie Xu; Jocelyn Ursua; L. Jeffrey Medeiros; Nghia (Andy) Nguyen; Sahib Kalra; Yasir Ali

arxiv: 2604.13795 · v1 · submitted 2026-04-15 · 💻 cs.CV · cs.LG

Artificial intelligence application in lymphoma diagnosis with Vision Transformer using weakly supervised training

Nghia (Andy) Nguyen , Amer Wahed , Andy Quesada , Yasir Ali , Hanadi El Achi , Y. Helen Zhang , Jocelyn Ursua , Alex Banerjee

show 3 more authors

Sahib Kalra L. Jeffrey Medeiros Jie Xu

This is my paper

Pith reviewed 2026-05-10 13:14 UTC · model grok-4.3

classification 💻 cs.CV cs.LG

keywords vision transformerweakly supervised learninglymphoma diagnosisanaplastic large cell lymphomaclassic hodgkin lymphomapathology image classificationdeep learning in medicine

0 comments

The pith

A Vision Transformer trained with weak supervision on 100,000 patches classifies anaplastic large cell lymphoma versus classic Hodgkin lymphoma at 91.85 percent accuracy.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper shows that Vision Transformers can distinguish anaplastic large cell lymphoma from classic Hodgkin lymphoma when trained only on slide-level labels rather than expert-labeled patches. The approach extracts 100,000 image patches from whole-slide images and assigns each patch the label of its parent slide, then trains the model on this larger but noisier dataset. The resulting model reaches 91.85 percent accuracy, an F1 score of 0.92, and an AUC of 0.98 on an independent test set. These results indicate the method could become a practical module inside clinical deep-learning systems that rely on automated patch extraction. Readers would care because the technique removes the main barrier of needing pathologists to annotate every small image region.

Core claim

The authors demonstrate that their Vision Transformer model, trained via weak supervision on a dataset of 100,000 image patches, achieves 91.85 percent accuracy, 0.92 F1 score, and 0.98 AUC when classifying anaplastic large cell lymphoma against classic Hodgkin lymphoma on held-out test data, and they conclude this performance qualifies the model for use in clinical deep-learning pipelines that employ automated image patch extraction.

What carries the argument

Vision Transformer architecture trained under weak supervision, where every patch automatically receives the diagnostic label of the whole-slide image from which it was extracted.

If this is right

Training no longer requires pathologists to label every individual patch.
The model can slot into existing workflows that automatically extract patches from scanned slides.
Vision Transformers become viable for this morphology task even when only slide-level labels are available.
Performance remains high enough to support further development of automated diagnostic support tools.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same weak-supervision recipe could be tried on other lymphoma subtypes or additional cancer types where slide-level reports already exist.
Hospitals could run such models on incoming slides to pre-screen cases and flag those needing urgent pathologist review.
Combining the model output with other clinical data might further improve diagnostic consistency across different labs.

Load-bearing premise

Slide-level labels supply accurate enough supervision for reliable patch-level predictions, and the independent test set matches the variability of real clinical cases.

What would settle it

A new test collection in which pathologists directly label individual patches and the model accuracy drops well below 80 percent would show the approach does not meet the claimed suitability for clinical use.

Figures

Figures reproduced from arXiv: 2604.13795 by Alex Banerjee, Amer Wahed, Andy Quesada, Hanadi El Achi, Jie Xu, Jocelyn Ursua, L. Jeffrey Medeiros, Nghia (Andy) Nguyen, Sahib Kalra, Yasir Ali, Y. Helen Zhang.

**Figure 3.** Figure 3: Lymphoma image processing workflow: 60 lymphoma cases stained with Hematoxylin and Eosin (H&E). Epredia P1000 scans slides at 40x magnification. Classical Hodgkin Lymphoma and Anaplastic Large T-Cell Lymphoma: two lymphoma types analyzed, shown with marked regions (blue and red squares). Image patches: 100x100 pixel patches were extracted from each lymphoma cohort for detailed analysis [PITH_FULL_IMAGE:fi… view at source ↗

**Figure 4.** Figure 4: The workflow for two automated image extraction methods used in the study: FAST Python library (upper) and QuPath (lower) [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗

read the original abstract

Vision transformers (ViT) have been shown to allow for more flexible feature detection and can outperform convolutional neural network (CNN) when pre-trained on sufficient data. Due to their promising feature detection capabilities, we deployed ViTs for morphological classification of anaplastic large cell lymphoma (ALCL) versus classic Hodgkin lymphoma (cHL). We had previously designed a ViT model which was trained on a small dataset of 1,200 image patches in fully supervised training. That model achieved a diagnostic accuracy of 100% and an F1 score of 1.0 on the independent test set. Since fully supervised training is not a practical method due to lack of expertise resources in both the training and testing phases, we conducted a recent study on a modified approach to training data (weakly supervised training) and show that labeling training image patch automatically at the slide level of each whole-slide-image is a more practical solution for clinical use of Vision Transformer. Our ViT model, trained on a larger dataset of 100,000 image patches, yields evaluation metrics with significant accuracy, F1 score, and area under the curve (AUC) at 91.85%, 0.92, and 0.98, respectively. These are respectable values that qualify this ViT model, with weakly supervised training, as a suitable tool for a deep learning module in clinical model development using automated image patch extraction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Weakly supervised ViT scales up lymphoma classification but missing details and label noise raise questions about the reported performance.

read the letter

The paper extends their prior fully supervised Vision Transformer work on anaplastic large cell lymphoma versus classic Hodgkin lymphoma by switching to weak supervision. They label 100,000 patches automatically from slide-level annotations instead of hand-labeling 1,200 patches, and report 91.85% accuracy, 0.92 F1, and 0.98 AUC on an independent test set. This shows a more scalable way to train the model without needing expert patch-level labels. The strength here is the focus on practicality. Fully supervised training is resource-heavy, so demonstrating that slide-level labels can still produce usable performance is a reasonable step forward for medical imaging applications. The main issues are in the experimental setup. The abstract and available details do not describe the data sources, the exact train-test split, or how patches were extracted and assigned labels. In lymphoma slides, many patches will not contain the key diagnostic cells, so labeling all patches from an ALCL slide as ALCL introduces noise. This likely contributes to the performance drop from the earlier 100% accuracy. There are no comparisons to other models like CNNs, no statistical tests, and no discussion of potential distribution shifts in the test set. The claim that these metrics qualify the model for clinical development feels premature without that support. This work is aimed at pathologists and AI researchers working on lymphoma diagnosis or similar weakly supervised tasks in digital pathology. A reader already familiar with ViTs in medical imaging might find the scaling experiment useful as a case study, though they would have to replicate the details themselves. It is solid enough in its core idea to warrant peer review, where the authors can supply the missing information on methods and validation. I would recommend sending this to referees rather than rejecting it outright, as the practical contribution could be strengthened with clearer reporting.

Referee Report

2 major / 0 minor

Summary. The manuscript presents a Vision Transformer (ViT) model for distinguishing anaplastic large cell lymphoma (ALCL) from classic Hodgkin lymphoma (cHL) via weakly supervised training. It contrasts a prior fully supervised ViT achieving 100% accuracy and F1=1.0 on 1,200 patches with a new model trained on 100,000 automatically slide-level-labeled patches that reports 91.85% accuracy, 0.92 F1 score, and 0.98 AUC on an independent test set, claiming this qualifies the approach as suitable for clinical deep-learning modules using automated patch extraction.

Significance. If the weak-supervision results prove robust after proper validation, the work could demonstrate a scalable route to ViT-based lymphoma classification that reduces expert annotation costs compared with fully supervised patch labeling. The reported AUC of 0.98 on a larger dataset would support further development of automated pathology tools, provided label noise and distribution-shift issues are addressed.

major comments (2)

[Abstract] Abstract: The reported metrics (91.85% accuracy, 0.92 F1, 0.98 AUC) are presented without any description of data sources, number of WSIs, train-test split details, patch extraction procedure, how slide-level labels were propagated to patches, baseline comparisons, or statistical validation (e.g., confidence intervals). This information is load-bearing for the central claim that the model is suitable for clinical use.
[Abstract] Abstract: The performance drop from 100% accuracy (fully supervised, 1,200 patches) to 91.85% (weakly supervised, 100k patches) is not analyzed in light of the known risk that slide-level labels assign the same class to all patches within a WSI, many of which contain stroma, necrosis, normal lymphoid tissue, or artifacts rather than diagnostic lymphoma morphology. This label noise is a plausible explanation for the observed drop and directly challenges the suitability claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and insightful comments on our work. We address each major comment below and have prepared revisions to the manuscript to incorporate the suggested improvements.

read point-by-point responses

Referee: [Abstract] The reported metrics (91.85% accuracy, 0.92 F1, 0.98 AUC) are presented without any description of data sources, number of WSIs, train-test split details, patch extraction procedure, how slide-level labels were propagated to patches, baseline comparisons, or statistical validation (e.g., confidence intervals). This information is load-bearing for the central claim that the model is suitable for clinical use.

Authors: We concur that the abstract would benefit from additional context to support the reported metrics and the claim of suitability for clinical use. The manuscript's Methods section details the data sources (pathology slides from our institution), the number of whole-slide images, the automated patch extraction process, slide-level label assignment to patches, and comparisons to the prior fully supervised model. We will revise the abstract to briefly mention the scale (100,000 patches from multiple WSIs) and refer readers to the full methods for specifics. Additionally, we will include statistical validation such as confidence intervals in the results. This change will be implemented in the revised version. revision: yes
Referee: [Abstract] The performance drop from 100% accuracy (fully supervised, 1,200 patches) to 91.85% (weakly supervised, 100k patches) is not analyzed in light of the known risk that slide-level labels assign the same class to all patches within a WSI, many of which contain stroma, necrosis, normal lymphoid tissue, or artifacts rather than diagnostic lymphoma morphology. This label noise is a plausible explanation for the observed drop and directly challenges the suitability claim.

Authors: The observation regarding label noise is valid and represents a known limitation of weak supervision. The performance drop is expected when moving from a small, expertly curated patch dataset to a large set with automatically propagated slide-level labels that include non-diagnostic regions. Our approach prioritizes scalability and reduced expert annotation requirements, which is crucial for clinical deployment. The maintained high AUC of 0.98 demonstrates that the model learns relevant features despite the noise. We will add a dedicated analysis and discussion of label noise effects in the revised manuscript, including potential mitigation strategies, and will moderate the suitability claim to reflect this as an initial demonstration requiring further clinical validation. We believe this addresses the concern while preserving the contribution. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical ML evaluation

full rationale

The paper reports an empirical study of training a Vision Transformer on 100k image patches for ALCL vs cHL classification under weakly supervised slide-level labeling, with standard accuracy/F1/AUC metrics on an independent test set. No mathematical derivations, equations, fitted parameters, or ansatzes appear; the central claims are direct experimental outcomes rather than reductions of inputs by construction. Prior self-citation to a small fully-supervised run is present but is not load-bearing for any derivation and does not create a self-referential chain. The work is self-contained against external benchmarks and contains none of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central performance claim rests on standard deep-learning training assumptions plus the domain-specific premise that slide-level labels suffice for patch classification.

free parameters (1)

ViT model weights and training hyperparameters
Fitted during optimization on the 100,000 patches to minimize the classification objective.

axioms (1)

domain assumption Slide-level labels are reliable proxies for the presence of diagnostic features in individual patches
Invoked by the choice of weakly supervised training described in the abstract.

pith-pipeline@v0.9.0 · 5595 in / 1324 out tokens · 50185 ms · 2026-05-10T13:14:33.723060+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

[1]

Deep Learning for Medical Image Processing: Overview, Challenges and the Future

Razzak MI, Naz S, Zaib A. Deep Learning for Medical Image Processing: Overview, Challenges and the Future. In: Dey N., Ashour A., Borra S. (eds) Classification in BioApps. First ed. Springer International Publishing; 2018:323-350

work page 2018
[2]

MIT Technol. Rev. 2013. Available at (last accessed on 10/30/18): https://www.technologyreview.com/s/513696/ deep-learning

work page 2013
[3]

Deep learning

LeCun, Y , Bengio, Y, Hinton, G. Deep learning. Nature. 2015;521:436–444

work page 2015
[4]

Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases

Janowczyk, A, Madabhushi, A. Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases. J Pathol Inform. 2016;7:29

work page 2016
[5]

WHO Classification of Tumours Editorial Board

WHO Classification of Tumours: Haematolymphoid Tumours, 5th Edition, V olume 11, 2024. WHO Classification of Tumours Editorial Board. 69008 Lyon, France: International Agency for Research on Cancer (IARC)

work page 2024
[6]

Feature Extraction for CBIR and Biometrics Applications

Choras RS. Feature Extraction for CBIR and Biometrics Applications. 7th WSEAS International Conference on Applied Computer Science. V ol. 7. 2007

work page 2007
[7]

Machine learning: an algorithmic perspective

Marsland S. Machine learning: an algorithmic perspective. Chapman and Hall/CRC, 2011

work page 2011
[8]

Machine learning

Mitchell TM, Mitchell TM. Machine learning. V ol. 1. No. 9. New York: McGraw-Hill, 1997

work page 1997
[9]

Patch-based system for classification of breast histology images using deep learning

Roy K, Banik D, Bhattacharjee D, Nasipuri M. Patch-based system for classification of breast histology images using deep learning. Computerized Medical Imaging and Graphics 71 (2019): 90-103

work page 2019
[10]

Revolutionizing Digital Pathology With the Power of Generative Artificial Intelligence and Foundation Models

Waqas, A, Bui, MM, Glassy, EF, et al. Revolutionizing Digital Pathology With the Power of Generative Artificial Intelligence and Foundation Models. Laboratory Investigation. V olume 103, Issue 11,100255, November 2023 17

work page 2023
[11]

Attention Is All You Need

Vaswani A, Shazeer, N, Parmar, N, et al. Attention Is All You Need. 31st Conference on Neural Information Processing Systems (NIPS 2017)

work page 2017
[12]

A foundation model for clinical-grade computational pathology and rare cancers detection

V orontsov, E, Bozkurt, A, Casson, A, Shaikovski, G, Zelechowski, M, Severson, K, et al. A foundation model for clinical-grade computational pathology and rare cancers detection. Nature Medicine | V olume 30 | October 2024 | 2924–2935

work page 2024
[13]

Vision transformer introduces a new vitality to the classification of renal pathology

Zhang, J, Lu, JD, Chen, B, et al. Vision transformer introduces a new vitality to the classification of renal pathology. BMC Nephrology (2024) 25:337

work page 2024
[14]

The classification of the bladder cancer based on Vision Transformers (ViT)

Khedr, OS, Wahed, ME, Al‑Attar, AR, et al. The classification of the bladder cancer based on Vision Transformers (ViT). Nature Scientific Reports | (2023) 13:20639

work page 2023
[15]

Better plain ViT baselines for ImageNet-1k

Beyer, L, et al. Better plain ViT baselines for ImageNet-1k. Google Brain Research, Zurich https://github.com/google-research/big_vision

work page
[16]

An image is worth 16x16 words: transformers for image recognition at scale

Dosovitskiy, A, Beyer, L, Kolesnikov, A, et al. An image is worth 16x16 words: transformers for image recognition at scale. Proceedings of ICLR 2021

work page 2021
[17]

Rivera D, Banerjee A, Zhang R, El Achi H, Wahed A, Ho L, et al(2025) Vision Transformers for Diagnostic Classification of Lymphomas: A Matched Comparison with a Convolutional Neural Network, 21st Century Pathol, V olume 5 (1): 160

work page 2025
[18]

Rivera, D, Ali, K, Zhang, R, Mai, B, El Achi, H, Armstrong, J, Wahed, A, Nguyen, A. Deep Learning-Based Morphological Classification between Classical Hodgkin Lymphoma and Anaplastic Large Cell Lymphoma: A Proof of Concept and Literature Review, 21st Century Pathology, V olume 4 (1): 159

work page
[19]

Foundations of Machine Learning

Mohri, M, Rostamizadeh, A,Talwalkar, A. Foundations of Machine Learning. MIT Press, Second Edition, 2018. 18

work page 2018
[20]

Vision transformer-based weakly supervised histopathological image analysis of primary brain tumors

Li, Z, Cong, Y, Chen, X, et al. Vision transformer-based weakly supervised histopathological image analysis of primary brain tumors. iScience 26, 105872, January 20, 2023

work page 2023
[21]

High performance neural network inference, streaming, and visualization of medical images using FAST

Smistad, E, Østvik, A, Pedersen, A. High performance neural network inference, streaming, and visualization of medical images using FAST. IEEE Access, volume 7, 2019

work page 2019
[22]

https://doi.org/10.1038/s41598-017-17204-5, https://www.nature.com/articles/s41598-017-17204-5

Bankhead, P., et al. QuPath: Open source software for digital pathology image analysis. Scientific Reports (2017). https://doi.org/10.1038/s41598-017-17204-5

work page doi:10.1038/s41598-017-17204-5 2017
[23]

https://www.geeksforgeeks.org/machine- learning/metrics-for-machine-learning-model/

Evaluation metrics in machine learning. https://www.geeksforgeeks.org/machine- learning/metrics-for-machine-learning-model/

work page
[24]

ALCL" or

Chaurasia, A, Toohey, PW, Harris, H, Hewitt, AW. Multi-resolution vision transformer model for histopathological skin cancer subtype classification using whole slide images. Computers in Biology and Medicine. V olume 196, Part A, September 2025, 110724. LEGENDS Figure 1. Representative histology of anaplastic large cell lymphoma (L) and classical Hodgkin ...

work page 2025

[1] [1]

Deep Learning for Medical Image Processing: Overview, Challenges and the Future

Razzak MI, Naz S, Zaib A. Deep Learning for Medical Image Processing: Overview, Challenges and the Future. In: Dey N., Ashour A., Borra S. (eds) Classification in BioApps. First ed. Springer International Publishing; 2018:323-350

work page 2018

[2] [2]

MIT Technol. Rev. 2013. Available at (last accessed on 10/30/18): https://www.technologyreview.com/s/513696/ deep-learning

work page 2013

[3] [3]

Deep learning

LeCun, Y , Bengio, Y, Hinton, G. Deep learning. Nature. 2015;521:436–444

work page 2015

[4] [4]

Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases

Janowczyk, A, Madabhushi, A. Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases. J Pathol Inform. 2016;7:29

work page 2016

[5] [5]

WHO Classification of Tumours Editorial Board

WHO Classification of Tumours: Haematolymphoid Tumours, 5th Edition, V olume 11, 2024. WHO Classification of Tumours Editorial Board. 69008 Lyon, France: International Agency for Research on Cancer (IARC)

work page 2024

[6] [6]

Feature Extraction for CBIR and Biometrics Applications

Choras RS. Feature Extraction for CBIR and Biometrics Applications. 7th WSEAS International Conference on Applied Computer Science. V ol. 7. 2007

work page 2007

[7] [7]

Machine learning: an algorithmic perspective

Marsland S. Machine learning: an algorithmic perspective. Chapman and Hall/CRC, 2011

work page 2011

[8] [8]

Machine learning

Mitchell TM, Mitchell TM. Machine learning. V ol. 1. No. 9. New York: McGraw-Hill, 1997

work page 1997

[9] [9]

Patch-based system for classification of breast histology images using deep learning

Roy K, Banik D, Bhattacharjee D, Nasipuri M. Patch-based system for classification of breast histology images using deep learning. Computerized Medical Imaging and Graphics 71 (2019): 90-103

work page 2019

[10] [10]

Revolutionizing Digital Pathology With the Power of Generative Artificial Intelligence and Foundation Models

Waqas, A, Bui, MM, Glassy, EF, et al. Revolutionizing Digital Pathology With the Power of Generative Artificial Intelligence and Foundation Models. Laboratory Investigation. V olume 103, Issue 11,100255, November 2023 17

work page 2023

[11] [11]

Attention Is All You Need

Vaswani A, Shazeer, N, Parmar, N, et al. Attention Is All You Need. 31st Conference on Neural Information Processing Systems (NIPS 2017)

work page 2017

[12] [12]

A foundation model for clinical-grade computational pathology and rare cancers detection

V orontsov, E, Bozkurt, A, Casson, A, Shaikovski, G, Zelechowski, M, Severson, K, et al. A foundation model for clinical-grade computational pathology and rare cancers detection. Nature Medicine | V olume 30 | October 2024 | 2924–2935

work page 2024

[13] [13]

Vision transformer introduces a new vitality to the classification of renal pathology

Zhang, J, Lu, JD, Chen, B, et al. Vision transformer introduces a new vitality to the classification of renal pathology. BMC Nephrology (2024) 25:337

work page 2024

[14] [14]

The classification of the bladder cancer based on Vision Transformers (ViT)

Khedr, OS, Wahed, ME, Al‑Attar, AR, et al. The classification of the bladder cancer based on Vision Transformers (ViT). Nature Scientific Reports | (2023) 13:20639

work page 2023

[15] [15]

Better plain ViT baselines for ImageNet-1k

Beyer, L, et al. Better plain ViT baselines for ImageNet-1k. Google Brain Research, Zurich https://github.com/google-research/big_vision

work page

[16] [16]

An image is worth 16x16 words: transformers for image recognition at scale

Dosovitskiy, A, Beyer, L, Kolesnikov, A, et al. An image is worth 16x16 words: transformers for image recognition at scale. Proceedings of ICLR 2021

work page 2021

[17] [17]

Rivera D, Banerjee A, Zhang R, El Achi H, Wahed A, Ho L, et al(2025) Vision Transformers for Diagnostic Classification of Lymphomas: A Matched Comparison with a Convolutional Neural Network, 21st Century Pathol, V olume 5 (1): 160

work page 2025

[18] [18]

Rivera, D, Ali, K, Zhang, R, Mai, B, El Achi, H, Armstrong, J, Wahed, A, Nguyen, A. Deep Learning-Based Morphological Classification between Classical Hodgkin Lymphoma and Anaplastic Large Cell Lymphoma: A Proof of Concept and Literature Review, 21st Century Pathology, V olume 4 (1): 159

work page

[19] [19]

Foundations of Machine Learning

Mohri, M, Rostamizadeh, A,Talwalkar, A. Foundations of Machine Learning. MIT Press, Second Edition, 2018. 18

work page 2018

[20] [20]

Vision transformer-based weakly supervised histopathological image analysis of primary brain tumors

Li, Z, Cong, Y, Chen, X, et al. Vision transformer-based weakly supervised histopathological image analysis of primary brain tumors. iScience 26, 105872, January 20, 2023

work page 2023

[21] [21]

High performance neural network inference, streaming, and visualization of medical images using FAST

Smistad, E, Østvik, A, Pedersen, A. High performance neural network inference, streaming, and visualization of medical images using FAST. IEEE Access, volume 7, 2019

work page 2019

[22] [22]

https://doi.org/10.1038/s41598-017-17204-5, https://www.nature.com/articles/s41598-017-17204-5

Bankhead, P., et al. QuPath: Open source software for digital pathology image analysis. Scientific Reports (2017). https://doi.org/10.1038/s41598-017-17204-5

work page doi:10.1038/s41598-017-17204-5 2017

[23] [23]

https://www.geeksforgeeks.org/machine- learning/metrics-for-machine-learning-model/

Evaluation metrics in machine learning. https://www.geeksforgeeks.org/machine- learning/metrics-for-machine-learning-model/

work page

[24] [24]

ALCL" or

Chaurasia, A, Toohey, PW, Harris, H, Hewitt, AW. Multi-resolution vision transformer model for histopathological skin cancer subtype classification using whole slide images. Computers in Biology and Medicine. V olume 196, Part A, September 2025, 110724. LEGENDS Figure 1. Representative histology of anaplastic large cell lymphoma (L) and classical Hodgkin ...

work page 2025