000294877 001__ 294877
000294877 005__ 20241210182915.0
000294877 0247_ $$2doi$$a10.48550/ARXIV.2403.12834
000294877 037__ $$aDKFZ-2024-02587
000294877 1001_ $$0P:(DE-He78)47726a6f46da663d4584c1c7254a285a$$aGotkowski, Karol$$b0$$eFirst author$$udkfz
000294877 245__ $$aEmbarrassingly Simple Scribble Supervision for 3D Medical Segmentation
000294877 260__ $$barXiv$$c2024
000294877 3367_ $$0PUB:(DE-HGF)25$$2PUB:(DE-HGF)$$aPreprint$$bpreprint$$mpreprint$$s1733825332_31422
000294877 3367_ $$2ORCID$$aWORKING_PAPER
000294877 3367_ $$028$$2EndNote$$aElectronic Article
000294877 3367_ $$2DRIVER$$apreprint
000294877 3367_ $$2BibTeX$$aARTICLE
000294877 3367_ $$2DataCite$$aOutput Types/Working Paper
000294877 520__ $$aTraditionally, segmentation algorithms require dense annotations for training, demanding significant annotation efforts, particularly within the 3D medical imaging field. Scribble-supervised learning emerges as a possible solution to this challenge, promising a reduction in annotation efforts when creating large-scale datasets. Recently, a plethora of methods for optimized learning from scribbles have been proposed, but have so far failed to position scribble annotation as a beneficial alternative. We relate this shortcoming to two major issues: 1) the complex nature of many methods which deeply ties them to the underlying segmentation model, thus preventing a migration to more powerful state-of-the-art models as the field progresses and 2) the lack of a systematic evaluation to validate consistent performance across the broader medical domain, resulting in a lack of trust when applying these methods to new segmentation problems. To address these issues, we propose a comprehensive scribble supervision benchmark consisting of seven datasets covering a diverse set of anatomies and pathologies imaged with varying modalities. We furthermore propose the systematic use of partial losses, i.e. losses that are only computed on annotated voxels. Contrary to most existing methods, these losses can be seamlessly integrated into state-of-the-art segmentation methods, enabling them to learn from scribble annotations while preserving their original loss formulations. Our evaluation using nnU-Net reveals that while most existing methods suffer from a lack of generalization, the proposed approach consistently delivers state-of-the-art performance. Thanks to its simplicity, our approach presents an embarrassingly simple yet effective solution to the challenges of scribble supervision. Source code as well as our extensive scribble benchmarking suite will be made publicly available upon publication.
000294877 536__ $$0G:(DE-HGF)POF4-315$$a315 - Bildgebung und Radioonkologie (POF4-315)$$cPOF4-315$$fPOF IV$$x0
000294877 588__ $$aDataset connected to DataCite
000294877 650_7 $$2Other$$aComputer Vision and Pattern Recognition (cs.CV)
000294877 650_7 $$2Other$$aFOS: Computer and information sciences
000294877 7001_ $$0P:(DE-He78)6a78e3a44a8038881d941fb467eb4e19$$aLüth, Carsten$$b1$$udkfz
000294877 7001_ $$0P:(DE-He78)04a0b5a49db132d8f00cee326cb743ca$$aJäger, Paul F.$$b2$$udkfz
000294877 7001_ $$0P:(DE-He78)ce72738384a651b46bb9b3f92dde8230$$aZiegler, Sebastian$$b3$$udkfz
000294877 7001_ $$0P:(DE-He78)9491eaed26df05482b258b0cac49375e$$aKrämer, Lars$$b4$$udkfz
000294877 7001_ $$0P:(DE-He78)e35f6a9bd89b1c66d107df8a2325a758$$aDenner, Stefan$$b5$$udkfz
000294877 7001_ $$0P:(DE-He78)d2bf7126723ea8f6005ba141ea3c3e2c$$aXiao, Shuhan$$b6$$udkfz
000294877 7001_ $$0P:(DE-He78)6aeec5fe74dfe54da185a85629cfbf3a$$aDisch, Nico$$b7$$udkfz
000294877 7001_ $$0P:(DE-He78)33c74005e1ce56f7025c4f6be15321b3$$aMaier-Hein, Klaus H.$$b8$$udkfz
000294877 7001_ $$0P:(DE-He78)7ea9af59d03ec7deb982a0e0562358fa$$aIsensee, Fabian$$b9$$eLast author$$udkfz
000294877 773__ $$a10.48550/ARXIV.2403.12834
000294877 909CO $$ooai:inrepo02.dkfz.de:294877$$pVDB
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)47726a6f46da663d4584c1c7254a285a$$aDeutsches Krebsforschungszentrum$$b0$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)6a78e3a44a8038881d941fb467eb4e19$$aDeutsches Krebsforschungszentrum$$b1$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)04a0b5a49db132d8f00cee326cb743ca$$aDeutsches Krebsforschungszentrum$$b2$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)ce72738384a651b46bb9b3f92dde8230$$aDeutsches Krebsforschungszentrum$$b3$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)9491eaed26df05482b258b0cac49375e$$aDeutsches Krebsforschungszentrum$$b4$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)e35f6a9bd89b1c66d107df8a2325a758$$aDeutsches Krebsforschungszentrum$$b5$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)d2bf7126723ea8f6005ba141ea3c3e2c$$aDeutsches Krebsforschungszentrum$$b6$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)6aeec5fe74dfe54da185a85629cfbf3a$$aDeutsches Krebsforschungszentrum$$b7$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)33c74005e1ce56f7025c4f6be15321b3$$aDeutsches Krebsforschungszentrum$$b8$$kDKFZ
000294877 9101_ $$0I:(DE-588b)2036810-0$$6P:(DE-He78)7ea9af59d03ec7deb982a0e0562358fa$$aDeutsches Krebsforschungszentrum$$b9$$kDKFZ
000294877 9131_ $$0G:(DE-HGF)POF4-315$$1G:(DE-HGF)POF4-310$$2G:(DE-HGF)POF4-300$$3G:(DE-HGF)POF4$$4G:(DE-HGF)POF$$aDE-HGF$$bGesundheit$$lKrebsforschung$$vBildgebung und Radioonkologie$$x0
000294877 9141_ $$y2024
000294877 9201_ $$0I:(DE-He78)E230-20160331$$kE230$$lE230 Medizinische Bildverarbeitung$$x0
000294877 9201_ $$0I:(DE-He78)E290-20160331$$kE290$$lNWG Interaktives maschinelles Lernen$$x1
000294877 980__ $$apreprint
000294877 980__ $$aVDB
000294877 980__ $$aI:(DE-He78)E230-20160331
000294877 980__ $$aI:(DE-He78)E290-20160331
000294877 980__ $$aUNRESTRICTED