Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models

Abstract High-quality image data is essential for training deep learning (DL) classifiers, yet data sharing is often limited by privacy concerns. We hypothesized that generative adversarial networks (GANs) could synthesize bone marrow smear (BMS) images suitable for classifier training. BMS from 125...

Full description

Saved in:
Bibliographic Details
Main Authors: Jan-Niklas Eckardt, Ishan Srivastava, Zizhe Wang, Susann Winter, Tim Schmittmann, Sebastian Riechert, Miriam Eva Helena Gediga, Anas Shekh Sulaiman, Martin M. K. Schneider, Freya Schulze, Christian Thiede, Katja Sockel, Frank Kroschinsky, Christoph Röllig, Martin Bornhäuser, Karsten Wendt, Jan Moritz Middeke
Format: Article
Language:English
Published: Nature Portfolio 2025-03-01
Series:npj Digital Medicine
Online Access:https://doi.org/10.1038/s41746-025-01563-9
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850054274931752960
author Jan-Niklas Eckardt
Ishan Srivastava
Zizhe Wang
Susann Winter
Tim Schmittmann
Sebastian Riechert
Miriam Eva Helena Gediga
Anas Shekh Sulaiman
Martin M. K. Schneider
Freya Schulze
Christian Thiede
Katja Sockel
Frank Kroschinsky
Christoph Röllig
Martin Bornhäuser
Karsten Wendt
Jan Moritz Middeke
author_facet Jan-Niklas Eckardt
Ishan Srivastava
Zizhe Wang
Susann Winter
Tim Schmittmann
Sebastian Riechert
Miriam Eva Helena Gediga
Anas Shekh Sulaiman
Martin M. K. Schneider
Freya Schulze
Christian Thiede
Katja Sockel
Frank Kroschinsky
Christoph Röllig
Martin Bornhäuser
Karsten Wendt
Jan Moritz Middeke
author_sort Jan-Niklas Eckardt
collection DOAJ
description Abstract High-quality image data is essential for training deep learning (DL) classifiers, yet data sharing is often limited by privacy concerns. We hypothesized that generative adversarial networks (GANs) could synthesize bone marrow smear (BMS) images suitable for classifier training. BMS from 1251 patients with acute myeloid leukemia (AML), 51 patients with acute promyelocytic leukemia (APL), and 236 stem cell donors were digitized, and synthetic images were generated using StyleGAN2-Ada. In a blinded visual Turing test, eight hematologists achieved 63% accuracy in identifying synthetic images, confirming high image quality. DL classifiers trained on real data achieved AUROCs of 0.99 across AML, APL, and donor classifications, with performance remaining above 0.95 even when incrementally substituting real data for synthetic samples. Adding synthetic data to real training data offered performance gains for an exceptionally rare disease (APL). Our study demonstrates the usability of synthetic BMS data for training highly accurate image classifiers in microscopy.
format Article
id doaj-art-1dc7ce8daaba4825b86ff26797334cf1
institution DOAJ
issn 2398-6352
language English
publishDate 2025-03-01
publisher Nature Portfolio
record_format Article
series npj Digital Medicine
spelling doaj-art-1dc7ce8daaba4825b86ff26797334cf12025-08-20T02:52:19ZengNature Portfolionpj Digital Medicine2398-63522025-03-018111010.1038/s41746-025-01563-9Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification modelsJan-Niklas Eckardt0Ishan Srivastava1Zizhe Wang2Susann Winter3Tim Schmittmann4Sebastian Riechert5Miriam Eva Helena Gediga6Anas Shekh Sulaiman7Martin M. K. Schneider8Freya Schulze9Christian Thiede10Katja Sockel11Frank Kroschinsky12Christoph Röllig13Martin Bornhäuser14Karsten Wendt15Jan Moritz Middeke16Department of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyChair of Software Technology, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyChair of Software Technology, TUD Dresden University of TechnologyElse Kröner Fresenius Center for Digital Health, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyChair of Software Technology, TUD Dresden University of TechnologyDepartment of Internal Medicine I, University Hospital Carl Gustav Carus, TUD Dresden University of TechnologyAbstract High-quality image data is essential for training deep learning (DL) classifiers, yet data sharing is often limited by privacy concerns. We hypothesized that generative adversarial networks (GANs) could synthesize bone marrow smear (BMS) images suitable for classifier training. BMS from 1251 patients with acute myeloid leukemia (AML), 51 patients with acute promyelocytic leukemia (APL), and 236 stem cell donors were digitized, and synthetic images were generated using StyleGAN2-Ada. In a blinded visual Turing test, eight hematologists achieved 63% accuracy in identifying synthetic images, confirming high image quality. DL classifiers trained on real data achieved AUROCs of 0.99 across AML, APL, and donor classifications, with performance remaining above 0.95 even when incrementally substituting real data for synthetic samples. Adding synthetic data to real training data offered performance gains for an exceptionally rare disease (APL). Our study demonstrates the usability of synthetic BMS data for training highly accurate image classifiers in microscopy.https://doi.org/10.1038/s41746-025-01563-9
spellingShingle Jan-Niklas Eckardt
Ishan Srivastava
Zizhe Wang
Susann Winter
Tim Schmittmann
Sebastian Riechert
Miriam Eva Helena Gediga
Anas Shekh Sulaiman
Martin M. K. Schneider
Freya Schulze
Christian Thiede
Katja Sockel
Frank Kroschinsky
Christoph Röllig
Martin Bornhäuser
Karsten Wendt
Jan Moritz Middeke
Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models
npj Digital Medicine
title Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models
title_full Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models
title_fullStr Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models
title_full_unstemmed Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models
title_short Synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models
title_sort synthetic bone marrow images augment real samples in developing acute myeloid leukemia microscopy classification models
url https://doi.org/10.1038/s41746-025-01563-9
work_keys_str_mv AT janniklaseckardt syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT ishansrivastava syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT zizhewang syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT susannwinter syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT timschmittmann syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT sebastianriechert syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT miriamevahelenagediga syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT anasshekhsulaiman syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT martinmkschneider syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT freyaschulze syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT christianthiede syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT katjasockel syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT frankkroschinsky syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT christophrollig syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT martinbornhauser syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT karstenwendt syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels
AT janmoritzmiddeke syntheticbonemarrowimagesaugmentrealsamplesindevelopingacutemyeloidleukemiamicroscopyclassificationmodels