Latest Advances In Pathology And Whole Slide Imaging Research August 2025

by Sebastian Müller 74 views

Hey guys! Check out the latest and greatest in pathology and whole slide imaging research. This week's roundup is packed with exciting new papers, covering everything from foundation models to multiple instance learning. Whether you're a seasoned researcher or just getting started, there's something here for everyone. So, let's dive in and explore the cutting edge of digital pathology! Be sure to visit the Github page for an enhanced reading experience and even more papers!

Whole Slide Imaging

Title Date Comment
WeakSupCon: Weakly Supervised Contrastive Learning for Encoder Pre-training 2025-07-30
Medic...

Medical Image Computing and Computer Assisted Intervention (MICCAI) 2025 workshop on Efficient Medical AI

ModalTune: Fine-Tuning Slide-Level Foundation Models with Multi-Modal Information for Multi-task Learning in Digital Pathology 2025-07-30
Pathology Foundation Models are Scanner Sensitive: Benchmark and Mitigation with Contrastive ScanGen Loss 2025-07-29
Accep...

Accepted (Oral) in MedAGI 2025 International Workshop at MICCAI Conference

Machine learning-based multimodal prognostic models integrating pathology images and high-throughput omic data for overall survival prediction in cancer: a systematic review 2025-07-29
Main ...

Main article (50 pages, inc 3 tables, 4 figures). Supplementary material included with additional methodological information and data

SCORPION: Addressing Scanner-Induced Variability in Histopathology 2025-07-28
Accep...

Accepted in UNSURE 2025 workshop in MICCAI

PTCMIL: Multiple Instance Learning via Prompt Token Clustering for Whole Slide Image Analysis 2025-07-24
Robust sensitivity control in digital pathology via tile score distribution matching 2025-07-24
Camer...

Camera ready version. Accepted at MICCAI 2025

PreMix: Label-Efficient Multiple Instance Learning via Non-Contrastive Pre-training and Feature Mixing 2025-07-24 Under review
A Versatile Pathology Co-pilot via Reasoning Enhanced Multimodal Large Language Model 2025-07-23
Survival Modeling from Whole Slide Images via Patch-Level Graph Clustering and Mixture Density Experts 2025-07-22
A tissue and cell-level annotated H&E and PD-L1 histopathology image dataset in non-small cell lung cancer 2025-07-21
Our d...

Our dataset is available at 'https://zenodo.org/records/15674785' and our code is available at 'https://github.com/DIAGNijmegen/ignite-data-toolkit'

Leveraging Spatial Context for Positive Pair Sampling in Histopathology Image Representation Learning 2025-07-21
Probabilistic smooth attention for deep multiple instance learning in medical imaging 2025-07-20
RACR-MIL: Rank-aware contextual reasoning for weakly supervised grading of squamous cell carcinoma using whole slide images 2025-07-19
17 pa...

17 pages main text, 2 page references, 2 page appendix; under submission

WSI-Agents: A Collaborative Multi-Agent System for Multi-Modal Whole Slide Image Analysis 2025-07-19
Efficient Whole Slide Pathology VQA via Token Compression 2025-07-19
Leveraging Pathology Foundation Models for Panoptic Segmentation of Melanoma in H&E Images 2025-07-18
Accep...

Accepted by MIUA 2025

A Mixture of Experts (MoE) model to improve AI-based computational pathology prediction performance under variable levels of histopathology image blur 2025-07-18
Prototype-Based Multiple Instance Learning for Gigapixel Whole Slide Image Classification 2025-07-16
Accep...

Accepted to MICCAI 2025

Screen Them All: High-Throughput Pan-Cancer Genetic and Phenotypic Biomarker Screening from H&E Whole Slide Images 2025-07-14

Diving Deep into Whole Slide Imaging Research

Whole slide imaging (WSI) has revolutionized pathology, and this week's papers showcase some of the most cutting-edge advancements in the field. From weakly supervised contrastive learning to scanner-induced variability, these studies address critical challenges and open up exciting new possibilities. For example, WeakSupCon: Weakly Supervised Contrastive Learning for Encoder Pre-training presents a novel approach to pre-training encoders, while SCORPION: Addressing Scanner-Induced Variability in Histopathology tackles a significant hurdle in WSI analysis. These papers demonstrate the power of machine learning and artificial intelligence in transforming how we analyze and interpret pathology images. The integration of these technologies promises to improve diagnostic accuracy, accelerate research, and ultimately enhance patient care. This section is a treasure trove for anyone interested in the latest developments in WSI, offering insights into the techniques and applications that are shaping the future of pathology. Keep an eye on these trends, guys, as they're sure to make a big impact in the coming years. The work being done in WSI is truly transformative, bridging the gap between traditional microscopy and the digital age. The advancements highlighted here are not just incremental improvements; they represent a paradigm shift in how pathology is practiced and researched.

Whole Slide Images

Title Date Comment
WeakSupCon: Weakly Supervised Contrastive Learning for Encoder Pre-training 2025-07-30
Medic...

Medical Image Computing and Computer Assisted Intervention (MICCAI) 2025 workshop on Efficient Medical AI

ModalTune: Fine-Tuning Slide-Level Foundation Models with Multi-Modal Information for Multi-task Learning in Digital Pathology 2025-07-30
Pathology Foundation Models are Scanner Sensitive: Benchmark and Mitigation with Contrastive ScanGen Loss 2025-07-29
Accep...

Accepted (Oral) in MedAGI 2025 International Workshop at MICCAI Conference

Machine learning-based multimodal prognostic models integrating pathology images and high-throughput omic data for overall survival prediction in cancer: a systematic review 2025-07-29
Main ...

Main article (50 pages, inc 3 tables, 4 figures). Supplementary material included with additional methodological information and data

SCORPION: Addressing Scanner-Induced Variability in Histopathology 2025-07-28
Accep...

Accepted in UNSURE 2025 workshop in MICCAI

PTCMIL: Multiple Instance Learning via Prompt Token Clustering for Whole Slide Image Analysis 2025-07-24
Robust sensitivity control in digital pathology via tile score distribution matching 2025-07-24
Camer...

Camera ready version. Accepted at MICCAI 2025

PreMix: Label-Efficient Multiple Instance Learning via Non-Contrastive Pre-training and Feature Mixing 2025-07-24 Under review
A Versatile Pathology Co-pilot via Reasoning Enhanced Multimodal Large Language Model 2025-07-23
Survival Modeling from Whole Slide Images via Patch-Level Graph Clustering and Mixture Density Experts 2025-07-22
A tissue and cell-level annotated H&E and PD-L1 histopathology image dataset in non-small cell lung cancer 2025-07-21
Our d...

Our dataset is available at 'https://zenodo.org/records/15674785' and our code is available at 'https://github.com/DIAGNijmegen/ignite-data-toolkit'

Leveraging Spatial Context for Positive Pair Sampling in Histopathology Image Representation Learning 2025-07-21
Probabilistic smooth attention for deep multiple instance learning in medical imaging 2025-07-20
RACR-MIL: Rank-aware contextual reasoning for weakly supervised grading of squamous cell carcinoma using whole slide images 2025-07-19
17 pa...

17 pages main text, 2 page references, 2 page appendix; under submission

WSI-Agents: A Collaborative Multi-Agent System for Multi-Modal Whole Slide Image Analysis 2025-07-19
Efficient Whole Slide Pathology VQA via Token Compression 2025-07-19
Leveraging Pathology Foundation Models for Panoptic Segmentation of Melanoma in H&E Images 2025-07-18
Accep...

Accepted by MIUA 2025

A Mixture of Experts (MoE) model to improve AI-based computational pathology prediction performance under variable levels of histopathology image blur 2025-07-18
Prototype-Based Multiple Instance Learning for Gigapixel Whole Slide Image Classification 2025-07-16
Accep...

Accepted to MICCAI 2025

Screen Them All: High-Throughput Pan-Cancer Genetic and Phenotypic Biomarker Screening from H&E Whole Slide Images 2025-07-14

The Expanding Universe of Whole Slide Images

Focusing on whole slide images, this section highlights the innovative research that's pushing the boundaries of what's possible with digital pathology. The sheer volume of data contained in a whole slide image presents both a challenge and an opportunity, and these papers tackle this head-on. From ModalTune: Fine-Tuning Slide-Level Foundation Models with Multi-Modal Information for Multi-task Learning in Digital Pathology to Survival Modeling from Whole Slide Images via Patch-Level Graph Clustering and Mixture Density Experts, the studies showcase diverse approaches to extracting meaningful insights from these complex images. Leveraging Pathology Foundation Models for Panoptic Segmentation of Melanoma in H&E Images exemplifies the power of foundation models in this domain. What's particularly exciting is the emphasis on multi-modal learning, where information from different sources (e.g., images, genomic data) is integrated to provide a more comprehensive understanding of the disease. This holistic approach holds immense promise for improving diagnostic accuracy and personalized treatment strategies. Guys, the future of pathology is looking bright, with WSI at the forefront of this transformation. The integration of artificial intelligence and machine learning with WSI technology is opening up new avenues for research and clinical practice, promising a future where diagnoses are more precise, treatments are more targeted, and patient outcomes are significantly improved.

Pathology

Title Date Comment
Priority-Aware Clinical Pathology Hierarchy Training for Multiple Instance Learning 2025-07-31
10 pa...

10 pages, 4 figures, Accepted for oral presentation by The 2nd MICCAI Student Board (MSB) EMERGE Workshop

ModalTune: Fine-Tuning Slide-Level Foundation Models with Multi-Modal Information for Multi-task Learning in Digital Pathology 2025-07-30
Pathology Foundation Models are Scanner Sensitive: Benchmark and Mitigation with Contrastive ScanGen Loss 2025-07-29
Accep...

Accepted (Oral) in MedAGI 2025 International Workshop at MICCAI Conference

VLM-CPL: Consensus Pseudo Labels from Vision-Language Models for Annotation-Free Pathological Image Classification 2025-07-29 Accepted at TMI
ViCTr: Vital Consistency Transfer for Pathology Aware Image Synthesis 2025-07-25
Accep...

Accepted in ICCV 2025

Robust sensitivity control in digital pathology via tile score distribution matching 2025-07-24
Camer...

Camera ready version. Accepted at MICCAI 2025

DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis 2025-07-24
TCM-Tongue: A Standardized Tongue Image Dataset with Pathological Annotations for AI-Assisted TCM Diagnosis 2025-07-24
16 pa...

16 pages, 11 figures, 2 Tables

Deep Learning for Glioblastoma Morpho-pathological Features Identification: A BraTS-Pathology Challenge Solution 2025-07-24
Accep...

Accepted by the International Brain Tumor Segmentation (BraTS) challenge organized at MICCAI 2024 conference

A Versatile Pathology Co-pilot via Reasoning Enhanced Multimodal Large Language Model 2025-07-23
Towards Robust Foundation Models for Digital Pathology 2025-07-22
Multi-modal vision-language model for generalizable annotation-free pathology localization and clinical diagnosis 2025-07-22
Efficient Whole Slide Pathology VQA via Token Compression 2025-07-19
Leveraging Pathology Foundation Models for Panoptic Segmentation of Melanoma in H&E Images 2025-07-18
Accep...

Accepted by MIUA 2025

A Mixture of Experts (MoE) model to improve AI-based computational pathology prediction performance under variable levels of histopathology image blur 2025-07-18
Pathology-Guided Virtual Staining Metric for Evaluation and Training 2025-07-16
19 pa...

19 pages, 10 figures. Intended for submission to the Journal of Imaging Informatics in Medicine (JIIM)

Integrating Pathology Foundation Models and Spatial Transcriptomics for Cellular Decomposition from Histology Images 2025-07-09
EXAONE Path 2.0: Pathology Foundation Model with End-to-End Supervision 2025-07-09
EXAON...

EXAONE Path 2.0 technical report

Revisiting Automatic Data Curation for Vision Foundation Models in Digital Pathology 2025-07-08 MICCAI 2025
ViTaL: A Multimodality Dataset and Benchmark for Multi-pathological Ovarian Tumor Recognition 2025-07-06

The Broad Scope of Pathology Research

This section delves into the core of pathology, showcasing a diverse range of research topics that are shaping the field. From the development of Pathology Foundation Models to innovative diagnostic tools, these papers represent the cutting edge of pathology research. The study on Deep Learning for Glioblastoma Morpho-pathological Features Identification highlights the power of artificial intelligence in cancer diagnostics, while TCM-Tongue: A Standardized Tongue Image Dataset with Pathological Annotations for AI-Assisted TCM Diagnosis explores the intersection of traditional medicine and modern technology. Towards Robust Foundation Models for Digital Pathology underscores the importance of creating reliable AI systems for clinical use. The common thread running through these studies is a commitment to improving diagnostic accuracy, streamlining workflows, and ultimately enhancing patient outcomes. It's fascinating to see how researchers are leveraging machine learning, computer vision, and other advanced technologies to tackle some of the most pressing challenges in pathology. Keep an eye on the progress in foundation models, as they are poised to revolutionize the way we approach digital pathology. The integration of diverse modalities, such as imaging, genomics, and clinical data, promises a more holistic and personalized approach to patient care.

Multiple Instance Learning

Title Date Comment
Priority-Aware Clinical Pathology Hierarchy Training for Multiple Instance Learning 2025-07-31
10 pa...

10 pages, 4 figures, Accepted for oral presentation by The 2nd MICCAI Student Board (MSB) EMERGE Workshop

WeakSupCon: Weakly Supervised Contrastive Learning for Encoder Pre-training 2025-07-30
Medic...

Medical Image Computing and Computer Assisted Intervention (MICCAI) 2025 workshop on Efficient Medical AI

Pathology Foundation Models are Scanner Sensitive: Benchmark and Mitigation with Contrastive ScanGen Loss 2025-07-29
Accep...

Accepted (Oral) in MedAGI 2025 International Workshop at MICCAI Conference

Predicting Neoadjuvant Chemotherapy Response in Triple-Negative Breast Cancer Using Pre-Treatment Histopathologic Images 2025-07-26
PTCMIL: Multiple Instance Learning via Prompt Token Clustering for Whole Slide Image Analysis 2025-07-24
Robust sensitivity control in digital pathology via tile score distribution matching 2025-07-24
Camer...

Camera ready version. Accepted at MICCAI 2025

PreMix: Label-Efficient Multiple Instance Learning via Non-Contrastive Pre-training and Feature Mixing 2025-07-24 Under review
Leveraging Spatial Context for Positive Pair Sampling in Histopathology Image Representation Learning 2025-07-21
Probabilistic smooth attention for deep multiple instance learning in medical imaging 2025-07-20
RACR-MIL: Rank-aware contextual reasoning for weakly supervised grading of squamous cell carcinoma using whole slide images 2025-07-19
17 pa...

17 pages main text, 2 page references, 2 page appendix; under submission

A Transformer-Based Conditional GAN with Multiple Instance Learning for UAV Signal Detection and Classification 2025-07-19 13 pages, 7 figures
Smarter Together: Combining Large Language Models and Small Models for Physiological Signals Visual Inspection 2025-07-18
ProDisc-VAD: An Efficient System for Weakly-Supervised Anomaly Detection in Video Surveillance Applications 2025-07-17
arXiv...

arXiv admin comment: This version has been removed by arXiv administrators as the submitter did not have the rights to agree to the license at the time of submission

Prototype-Based Multiple Instance Learning for Gigapixel Whole Slide Image Classification 2025-07-16
Accep...

Accepted to MICCAI 2025

SGPMIL: Sparse Gaussian Process Multiple Instance Learning 2025-07-11
8 pag...

8 pages, 4 figures, 2 tables

Cracking Instance Jigsaw Puzzles: An Alternative to Multiple Instance Learning for Whole Slide Image Analysis 2025-07-10 Accepted by ICCV2025
GNN-ViTCap: GNN-Enhanced Multiple Instance Learning with Vision Transformers for Whole Slide Image Classification and Captioning 2025-07-09
EXAONE Path 2.0: Pathology Foundation Model with End-to-End Supervision 2025-07-09
EXAON...

EXAONE Path 2.0 technical report

Sequential Attention-based Sampling for Histopathological Analysis 2025-07-09
The Trilemma of Truth in Large Language Models 2025-07-08

Unlocking Insights with Multiple Instance Learning

Multiple instance learning (MIL) is a powerful technique for dealing with weakly labeled data, and it's particularly well-suited for analyzing whole slide images. This section highlights the latest advancements in MIL research, showcasing its versatility and potential. From Priority-Aware Clinical Pathology Hierarchy Training for Multiple Instance Learning to Prototype-Based Multiple Instance Learning for Gigapixel Whole Slide Image Classification, these papers demonstrate the diverse applications of MIL in pathology. The study on PTCMIL: Multiple Instance Learning via Prompt Token Clustering for Whole Slide Image Analysis presents a novel approach to MIL, while RACR-MIL: Rank-aware contextual reasoning for weakly supervised grading of squamous cell carcinoma using whole slide images tackles a specific clinical challenge. MIL allows researchers to train models even when they only have bag-level labels (e.g., a slide is cancerous), rather than instance-level labels (e.g., specific cells are cancerous). This is a game-changer for digital pathology, where obtaining detailed annotations can be time-consuming and expensive. This technique helps computers understand and learn from complex images, even when there isn't a lot of specific information. MIL can be really helpful for things like diagnosing diseases from medical images, where it's not always clear what parts of the image are important. Guys, MIL is a crucial tool in the digital pathology toolbox, enabling us to extract valuable information from complex images and improve diagnostic accuracy.

Pathology Reports

Title Date Comment
CLIP-IT: CLIP-based Pairing for Histology Images Classification 2025-07-29
Can human clinical rationales improve the performance and explainability of clinical text classification models? 2025-07-28
DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis 2025-07-24
Historical Report Guided Bi-modal Concurrent Learning for Pathology Report Generation 2025-06-23
On the Importance of Text Preprocessing for Multimodal Representation Learning and Pathology Report Generation 2025-06-06 11 pages, 1 figure
VLCD: Vision-Language Contrastive Distillation for Accurate and Efficient Automatic Placenta Analysis 2025-06-02
Proce...

Proceedings of the 9th International Workshop on Health Intelligence, in conjunction with the Annual AAAI Conference on Artificial Intelligence, Philadelphia, Pennsylvania, March 2025

Multimodal Survival Modeling in the Age of Foundation Models 2025-05-28
23 pa...

23 pages, 7 figures, 8 tables, updated with acknowledgements and declarations of interest

Any-to-Any Learning in Computational Pathology via Triplet Multimodal Pretraining 2025-05-20
Small or Large? Zero-Shot or Finetuned? Guiding Language Model Choice for Specialized Applications in Healthcare 2025-04-29
Global explainability of a deep abstaining classifier 2025-04-01
CancerLLM: A Large Language Model in Cancer Domain 2025-04-01
new v...

new version, add the RAG version of cancerLLM

Vision Language Models versus Machine Learning Models Performance on Polyp Detection and Classification in Colonoscopy Images 2025-03-27
Code ...

Code is available at: https://github.com/aminkhalafi/CML-vs-LLM-on-Polyp-Detection. CoI: AlSo serves on the advisory board and holds equity in Virgo Surgical Solutions. The other authors declare no conflicts of interest. Data

A Multimodal Knowledge-enhanced Whole-slide Pathology Foundation Model 2025-03-25 62 pages
ELM: Ensemble of Language Models for Predicting Tumor Group from Pathology Reports 2025-03-24
Towards Scalable and Cross-Lingual Specialist Language Models for Oncology 2025-03-11
Cancer Type, Stage and Prognosis Assessment from Pathology Reports using LLMs 2025-03-03
Pathology Report Generation and Multimodal Representation Learning for Cutaneous Melanocytic Lesions 2025-02-27
11 pa...

11 pages, 2 figures. arXiv admin note: text overlap with arXiv:2502.19285

Leveraging large language models for structured information extraction from pathology reports 2025-02-14 29 pages, 6 figures
PolyPath: Adapting a Large Multimodal Model for Multi-slide Pathology Report Generation 2025-02-14
8 mai...

8 main pages, 21 pages in total

Volumetric Reconstruction of Prostatectomy Specimens from Histology 2024-11-29

Revolutionizing Pathology Reports with AI

This section focuses on the critical area of pathology reports, highlighting research that aims to automate and improve the generation and analysis of these crucial documents. From DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis to Leveraging large language models for structured information extraction from pathology reports, these papers explore the use of artificial intelligence and natural language processing to enhance the efficiency and accuracy of pathology reporting. The study on Can human clinical rationales improve the performance and explainability of clinical text classification models? addresses the important issue of explainability in AI-driven pathology. CancerLLM: A Large Language Model in Cancer Domain showcases the potential of large language models (LLMs) in cancer diagnostics and treatment. The goal is to create systems that can not only generate reports but also extract key information, identify patterns, and ultimately provide clinicians with valuable insights to guide patient care. This is a rapidly evolving field, with LLMs playing an increasingly important role. Pathology reports are a crucial part of healthcare, but they can be long and complex. New AI technologies can help doctors write these reports more quickly and accurately. Guys, this is an area where AI can truly make a difference in the lives of both pathologists and patients.

Pathology Report Generation

Title Date Comment
DiagR1: A Vision-Language Model Trained via Reinforcement Learning for Digestive Pathology Diagnosis 2025-07-24
Historical Report Guided Bi-modal Concurrent Learning for Pathology Report Generation 2025-06-23
On the Importance of Text Preprocessing for Multimodal Representation Learning and Pathology Report Generation 2025-06-06 11 pages, 1 figure
Pathology Report Generation and Multimodal Representation Learning for Cutaneous Melanocytic Lesions 2025-02-27
11 pa...

11 pages, 2 figures. arXiv admin note: text overlap with arXiv:2502.19285

PolyPath: Adapting a Large Multimodal Model for Multi-slide Pathology Report Generation 2025-02-14
8 mai...

8 main pages, 21 pages in total

Multimodal Whole Slide Foundation Model for Pathology 2024-11-29
The c...

The code is accessible at https://github.com/mahmoodlab/TITAN

Clinical-grade Multi-Organ Pathology Report Generation for Multi-scale Whole Slide Images via a Semantically Guided Medical Text Foundation Model 2024-09-23

Automating Pathology Report Generation

This section highlights the specialized field of pathology report generation, where researchers are developing AI systems that can automatically generate comprehensive and accurate pathology reports. Historical Report Guided Bi-modal Concurrent Learning for Pathology Report Generation demonstrates a novel approach to incorporating historical data into the report generation process. On the Importance of Text Preprocessing for Multimodal Representation Learning and Pathology Report Generation emphasizes the crucial role of text preprocessing in achieving optimal performance. PolyPath: Adapting a Large Multimodal Model for Multi-slide Pathology Report Generation showcases the power of multimodal models in handling complex pathology data. The ultimate goal is to create systems that can assist pathologists in their daily work, reducing workload and improving efficiency. This field requires a deep understanding of both pathology and natural language processing, and the progress being made is truly remarkable. AI is getting better at understanding and writing medical reports, which helps doctors do their jobs more efficiently. This technology can look at images and text from pathology tests and automatically create detailed reports. Guys, the automation of pathology report generation holds the promise of streamlining workflows, reducing errors, and freeing up pathologists to focus on more complex cases.

This week's collection of papers truly showcases the dynamic and rapidly evolving landscape of pathology and whole slide imaging. From the advancements in foundation models and multiple instance learning to the exciting developments in pathology report generation, there's a wealth of knowledge and innovation to explore. Stay tuned for more updates, and let's continue to push the boundaries of what's possible in digital pathology!