Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

A deep learning framework for efficient pathology image analysis.

Nature communications·2026
Same author

MetAssist 2.0: A Generalizable Artificial Intelligence Framework for Lymph Node Metastasis Detection Across Multiple Cancer Types.

Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc·2026
Same author

Unraveling the cause of microspurs in spontaneous intracranial hypotension type 1: discogenic origin or calcified Hofmann's ligament?

Journal of neurosurgery. Spine·2025
Same author

Benchmarking foundation models as feature extractors for weakly supervised computational pathology.

Nature biomedical engineering·2025
Same author

Immunohistochemical TTF-1 and Napsin a Expression in Gastrointestinal Adenocarcinomas-Low Frequency but an Important Pitfall.

Diagnostics (Basel, Switzerland)·2025
Same author

Diagnostic accuracy of unenhanced post-mortem CT and MRI compared to the non-forensic clinical autopsy: a prospective blinded study.

Virchows Archiv : an international journal of pathology·2025

Related Experiment Video

Updated: Sep 22, 2025

Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
05:33

Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System

Published on: July 11, 2025

297

Benchmarking weakly-supervised deep learning pipelines for whole slide classification in computational pathology.

Narmin Ghaffari Laleh1, Hannah Sophie Muti1, Chiara Maria Lavinia Loeffler1

  • 1Department of Medicine III, University Hospital RWTH Aachen, Aachen, Germany.

Medical Image Analysis
|May 19, 2022
PubMed
Summary
This summary is machine-generated.

Classical weakly-supervised methods surprisingly outperformed multiple-instance learning (MIL) for cancer mutation prediction from histopathology slides. This highlights the need to compare new AI methods against simpler approaches in computational pathology.

Keywords:
Artificial intelligenceComputational pathologyConvolutional neural networksMultiple-Instance LearningVision transformersWeakly-supervised deep learning

More Related Videos

Automated Slide Scanning and Segmentation in Fluorescently-labeled Tissues Using a Widefield High-content Analysis System
09:33

Automated Slide Scanning and Segmentation in Fluorescently-labeled Tissues Using a Widefield High-content Analysis System

Published on: May 3, 2018

8.2K
Digital Analysis of Immunostaining of ZW10 Interacting Protein in Human Lung Tissues
07:40

Digital Analysis of Immunostaining of ZW10 Interacting Protein in Human Lung Tissues

Published on: May 1, 2019

5.5K

Related Experiment Videos

Last Updated: Sep 22, 2025

Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
05:33

Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System

Published on: July 11, 2025

297
Automated Slide Scanning and Segmentation in Fluorescently-labeled Tissues Using a Widefield High-content Analysis System
09:33

Automated Slide Scanning and Segmentation in Fluorescently-labeled Tissues Using a Widefield High-content Analysis System

Published on: May 3, 2018

8.2K
Digital Analysis of Immunostaining of ZW10 Interacting Protein in Human Lung Tissues
07:40

Digital Analysis of Immunostaining of ZW10 Interacting Protein in Human Lung Tissues

Published on: May 1, 2019

5.5K

Area of Science:

  • Computational pathology
  • Artificial intelligence in histopathology
  • Machine learning for biomarker discovery

Background:

  • Whole slide images in histopathology are analyzed using AI for biological insights and clinical biomarkers.
  • Weakly-supervised learning is common, where only slide-level labels are available, not tile-level ground truth.
  • Classical methods assign slide labels to all tiles, while multiple-instance learning (MIL) uses 'bags' of tiles.

Purpose of the Study:

  • To systematically compare the performance of classical weakly-supervised methods against multiple-instance learning (MIL) approaches in computational pathology.
  • To evaluate these methods on clinically relevant prediction tasks using a large dataset.
  • To provide empirical evidence for the relative strengths of different AI strategies in histopathology.

Main Methods:

  • Implemented and compared six methods: three classical weakly-supervised approaches and three MIL-based approaches (with and without attention).
  • Utilized convolutional neural networks and vision transformers (ViT) for analysis.
  • Trained on N=2980 patient data with rigorous external validation across six prediction tasks.

Main Results:

  • All methods achieved high performance (AUROC > 0.9) for renal cell carcinoma subtyping.
  • Significant performance differences were observed in mutation prediction tasks for colorectal, gastric, and bladder cancers.
  • Classical weakly-supervised methods unexpectedly outperformed MIL-based methods for mutation prediction.

Conclusions:

  • Classical weakly-supervised methods are effective and should be used as benchmarks for new computational pathology pipelines.
  • The surprising performance of simpler methods motivates research into hybrid approaches combining MIL principles with classical method strengths.
  • Open-source code is provided to facilitate the application and further development of these AI methods.