Real world federated learning with a knowledge distilled transformer for cardiac CT imaging
Abstract Federated learning is a renowned technique for utilizing decentralized data while preserving privacy. However, real-world applications often face challenges like partially labeled datasets, where only a few locations have certain expert annotations, leaving large portions of unlabeled data...
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-02-01
|
| Series: | npj Digital Medicine |
| Online Access: | https://doi.org/10.1038/s41746-025-01434-3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1823861599027331072 |
|---|---|
| author | Malte Tölle Philipp Garthe Clemens Scherer Jan Moritz Seliger Andreas Leha Nina Krüger Stefan Simm Simon Martin Sebastian Eble Halvar Kelm Moritz Bednorz Florian André Peter Bannas Gerhard Diller Norbert Frey Stefan Groß Anja Hennemuth Lars Kaderali Alexander Meyer Eike Nagel Stefan Orwat Moritz Seiffert Tim Friede Tim Seidler Sandy Engelhardt |
| author_facet | Malte Tölle Philipp Garthe Clemens Scherer Jan Moritz Seliger Andreas Leha Nina Krüger Stefan Simm Simon Martin Sebastian Eble Halvar Kelm Moritz Bednorz Florian André Peter Bannas Gerhard Diller Norbert Frey Stefan Groß Anja Hennemuth Lars Kaderali Alexander Meyer Eike Nagel Stefan Orwat Moritz Seiffert Tim Friede Tim Seidler Sandy Engelhardt |
| author_sort | Malte Tölle |
| collection | DOAJ |
| description | Abstract Federated learning is a renowned technique for utilizing decentralized data while preserving privacy. However, real-world applications often face challenges like partially labeled datasets, where only a few locations have certain expert annotations, leaving large portions of unlabeled data unused. Leveraging these could enhance transformer architectures’ ability in regimes with small and diversely annotated sets. We conduct the largest federated cardiac CT analysis to date (n = 8, 104) in a real-world setting across eight hospitals. Our two-step semi-supervised strategy distills knowledge from task-specific CNNs into a transformer. First, CNNs predict on unlabeled data per label type and then the transformer learns from these predictions with label-specific heads. This improves predictive accuracy and enables simultaneous learning of all partial labels across the federation, and outperforms UNet-based models in generalizability on downstream tasks. Code and model weights are made openly available for leveraging future cardiac CT analysis. |
| format | Article |
| id | doaj-art-dccc8465471c47bd8fa210674ad4e336 |
| institution | Kabale University |
| issn | 2398-6352 |
| language | English |
| publishDate | 2025-02-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | npj Digital Medicine |
| spelling | doaj-art-dccc8465471c47bd8fa210674ad4e3362025-02-09T12:55:35ZengNature Portfolionpj Digital Medicine2398-63522025-02-018111410.1038/s41746-025-01434-3Real world federated learning with a knowledge distilled transformer for cardiac CT imagingMalte Tölle0Philipp Garthe1Clemens Scherer2Jan Moritz Seliger3Andreas Leha4Nina Krüger5Stefan Simm6Simon Martin7Sebastian Eble8Halvar Kelm9Moritz Bednorz10Florian André11Peter Bannas12Gerhard Diller13Norbert Frey14Stefan Groß15Anja Hennemuth16Lars Kaderali17Alexander Meyer18Eike Nagel19Stefan Orwat20Moritz Seiffert21Tim Friede22Tim Seidler23Sandy Engelhardt24DZHK (German Centre for Cardiovascular Research), partner site Heidelberg/MannheimClinic for Cardiology III, University Hospital MünsterDZHK (German Centre for Cardiovascular Research), partner site MunichDZHK (German Centre for Cardiovascular Research), partner site Hamburg/Kiel/LübeckDZHK (German Centre for Cardiovascular Research), partner site Lower SaxonyDZHK (German Centre for Cardiovascular Research), partner site BerlinDZHK (German Centre for Cardiovascular Research), partner site GreifswaldDZHK (German Centre for Cardiovascular Research), partner site RhineMainDepartment of Cardiology, Angiology and Pneumology, Heidelberg University HospitalDepartment of Cardiology, Angiology and Pneumology, Heidelberg University HospitalDepartment of Cardiology, Angiology and Pneumology, Heidelberg University HospitalDZHK (German Centre for Cardiovascular Research), partner site Heidelberg/MannheimDZHK (German Centre for Cardiovascular Research), partner site Hamburg/Kiel/LübeckClinic for Cardiology III, University Hospital MünsterDZHK (German Centre for Cardiovascular Research), partner site Heidelberg/MannheimDZHK (German Centre for Cardiovascular Research), partner site GreifswaldDZHK (German Centre for Cardiovascular Research), partner site BerlinDZHK (German Centre for Cardiovascular Research), partner site GreifswaldDZHK (German Centre for Cardiovascular Research), partner site BerlinDZHK (German Centre for Cardiovascular Research), partner site RhineMainClinic for Cardiology III, University Hospital MünsterDZHK (German Centre for Cardiovascular Research), partner site Hamburg/Kiel/LübeckDZHK (German Centre for Cardiovascular Research), partner site Lower SaxonyDZHK (German Centre for Cardiovascular Research), partner site Lower SaxonyDZHK (German Centre for Cardiovascular Research), partner site Heidelberg/MannheimAbstract Federated learning is a renowned technique for utilizing decentralized data while preserving privacy. However, real-world applications often face challenges like partially labeled datasets, where only a few locations have certain expert annotations, leaving large portions of unlabeled data unused. Leveraging these could enhance transformer architectures’ ability in regimes with small and diversely annotated sets. We conduct the largest federated cardiac CT analysis to date (n = 8, 104) in a real-world setting across eight hospitals. Our two-step semi-supervised strategy distills knowledge from task-specific CNNs into a transformer. First, CNNs predict on unlabeled data per label type and then the transformer learns from these predictions with label-specific heads. This improves predictive accuracy and enables simultaneous learning of all partial labels across the federation, and outperforms UNet-based models in generalizability on downstream tasks. Code and model weights are made openly available for leveraging future cardiac CT analysis.https://doi.org/10.1038/s41746-025-01434-3 |
| spellingShingle | Malte Tölle Philipp Garthe Clemens Scherer Jan Moritz Seliger Andreas Leha Nina Krüger Stefan Simm Simon Martin Sebastian Eble Halvar Kelm Moritz Bednorz Florian André Peter Bannas Gerhard Diller Norbert Frey Stefan Groß Anja Hennemuth Lars Kaderali Alexander Meyer Eike Nagel Stefan Orwat Moritz Seiffert Tim Friede Tim Seidler Sandy Engelhardt Real world federated learning with a knowledge distilled transformer for cardiac CT imaging npj Digital Medicine |
| title | Real world federated learning with a knowledge distilled transformer for cardiac CT imaging |
| title_full | Real world federated learning with a knowledge distilled transformer for cardiac CT imaging |
| title_fullStr | Real world federated learning with a knowledge distilled transformer for cardiac CT imaging |
| title_full_unstemmed | Real world federated learning with a knowledge distilled transformer for cardiac CT imaging |
| title_short | Real world federated learning with a knowledge distilled transformer for cardiac CT imaging |
| title_sort | real world federated learning with a knowledge distilled transformer for cardiac ct imaging |
| url | https://doi.org/10.1038/s41746-025-01434-3 |
| work_keys_str_mv | AT maltetolle realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT philippgarthe realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT clemensscherer realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT janmoritzseliger realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT andreasleha realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT ninakruger realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT stefansimm realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT simonmartin realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT sebastianeble realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT halvarkelm realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT moritzbednorz realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT florianandre realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT peterbannas realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT gerharddiller realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT norbertfrey realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT stefangroß realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT anjahennemuth realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT larskaderali realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT alexandermeyer realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT eikenagel realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT stefanorwat realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT moritzseiffert realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT timfriede realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT timseidler realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging AT sandyengelhardt realworldfederatedlearningwithaknowledgedistilledtransformerforcardiacctimaging |