@inproceedings{doerrich2026_116855,
  title = {Stylizing ViT : Anatomy-Preserving Instance Style Transfer for Domain Generalization},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Alle, Jonas and Ledig, Christian},
  year = {2026},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-116855x},
  url = {https://fis.uni-bamberg.de/handle/uniba/116855},
  language = {eng},
  abstract = {Deep learning models in medical image analysis often struggle with generalizability across domains and demographic groups due to data heterogeneity and scarcity. Traditional augmentation improves robustness, but fails under substantial domain shifts. Recent advances in stylistic augmentation enhance domain generalization by varying image styles but fall short in terms of style diversity or by introducing artifacts into the generated images. To address these limitations, we propose Stylizing ViT, a novel Vision Transformer encoder that utilizes weight-shared attention blocks for both self- and cross-attention. This design allows the same attention block to maintain anatomical consistency through self-attention while performing style transfer via cross-attention. We assess the effectiveness of our method for domain generalization by employing it for data augmentation on three distinct image classification tasks in the context of histopathology and dermatology. Results demonstrate an improved robustness (up to +13\% accuracy) over the state of the art while generating perceptually convincing images without artifacts. Additionally, we show that Stylizing ViT is effective beyond training, achieving a 17\% performance improvement during inference when used for test-time augmentation. The source code is available at https://github.com/sdoerrich97/stylizing-vit .},
  keywords = {style transfer, domain generalization, crossattention, data augmentation, vision transformer}
}

@inproceedings{doerrich2026_117126,
  title = {Stylizing ViT : Anatomy-Preserving Instance Style Transfer for Domain Generalization},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Alle, Jonas and Ledig, Christian},
  year = {2026},
  booktitle = {2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI)},
  publisher = {IEEE},
  address = {Piscataway, NJ},
  doi = {10.1109/isbi61048.2026.11515498},
  url = {https://fis.uni-bamberg.de/handle/uniba/117126},
  language = {eng},
  abstract = {Deep learning models in medical image analysis often struggle with generalizability across domains and demographic groups due to data heterogeneity and scarcity. Traditional augmentation improves robustness, but fails under substantial domain shifts. Recent advances in stylistic augmentation enhance domain generalization by varying image styles but fall short in terms of style diversity or by introducing artifacts into the generated images. To address these limitations, we propose Stylizing ViT, a novel Vision Transformer encoder that utilizes weight-shared attention blocks for both self- and cross-attention. This design allows the same attention block to maintain anatomical consistency through self-attention while performing style transfer via cross-attention. We assess the effectiveness of our method for domain generalization by employing it for data augmentation on three distinct image classification tasks in the context of histopathology and dermatology. Results demonstrate an improved robustness (up to +13\% accuracy) over the state of the art while generating perceptually convincing images without artifacts. Additionally, we show that Stylizing ViT is effective beyond training, achieving a 17\% performance improvement during inference when used for test-time augmentation. The source code is available at https://github.com/sdoerrich97/stylizing-vit .},
  keywords = {style transfer, domain generalization, crossattention, data augmentation, vision transformer}
}

@inproceedings{pan2026_113148,
  title = {An End-to-End AI Pipeline for Wood Knot Detection to Enhance Structural Assessment in Historic Timber Structures},
  author = {Pan, Junquan and Chizhova, Maria and Ebener, Frank and Luhmann, Thomas and Ledig, Christian and Maiwald, Ferdinand and Eißing, Thomas},
  year = {2026},
  pages = {267-274},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-113148x},
  url = {https://fis.uni-bamberg.de/handle/uniba/113148},
  language = {eng},
  abstract = {The accurate detection and assessment of wood surface defects in historic timber structures, particularly knots, is essential for effective conservation and strengthening planning. However, the application of automated visual grading methods to aged timber remains underexplored due to the irregular texture, weathering and lack of relevant datasets. In this study, we propose an end-to-end deep learning-based pipeline that integrates wood surface segmentation, perspective correction, and knot detection to estimate structural grading factors. A dedicated raw data collection of over 10,000 high-resolution images of historic timber surfaces was compiled using both DSLR cameras and mobile devices, resulting in multiple datasets with approximately 3,000 annotated samples. Three model families, YOLO, Detectron2 and DeepLabV3, were evaluated under different experimental setups. Beyond model benchmarking, we further compared the AI-derived results with expert manual measurements. The model for segmentation of timber surface achieved a mean IoU of over 0.85 and the model for detection of historical wood knots reached F1 scores of up to 0.9. The structural assessment factors estimated by the AI pipeline achieved a Pearson correlation coefficient of 0.641 compared to manual measurements, indicating a moderate level of consistency in knot factor estimation. This research highlights the potential of vision-based AI systems in supporting structural diagnosis and conservation of heritage timber elements.},
  keywords = {Photogrammetry, Deep Learning, Historic Timber Structures, Wood Knot Detection, Structural Assessment}
}

@article{kucerenko2025_111618,
  title = {Incorporating label uncertainty during the training of convolutional neural networks improves performance for the discrimination between certain and inconclusive cases in dopamine transporter SPECT},
  author = {Kucerenko, Aleksej and Buddenkotte, Thomas and Apostolova, Ivayla and Klutmann, Susanne and Ledig, Christian and Buchert, Ralph},
  year = {2025},
  pages = {1535-1548},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-111618x},
  url = {https://fis.uni-bamberg.de/handle/uniba/111618},
  language = {eng},
  abstract = {Purpose: Deep convolutional neural networks (CNN) hold promise for assisting the interpretation of dopamine transporter (DAT)-SPECT. For improved communication of uncertainty to the user it is crucial to reliably discriminate certain from inconclusive cases that might be misclassified by strict application of a predefined decision threshold on the CNN output. This study tested two methods to incorporate existing label uncertainty during the training to improve the utility of the CNN sigmoid output for this task.

Methods: Three datasets were used retrospectively: a “development” dataset (n = 1740) for CNN training, validation and testing, two independent out-of-distribution datasets (n = 640, 645) for testing only. In the development dataset, binary classification based on visual inspection was performed carefully by three well-trained readers. A ResNet-18 architecture was trained for binary classification of DAT-SPECT using either a randomly selected vote (“random vote training”, RVT), the proportion of “reduced” votes ( “average vote training”, AVT) or the majority vote (MVT) across the three readers as reference standard. Balanced accuracy was computed separately for “inconclusive” sigmoid outputs (within a predefined interval around the 0.5 decision threshold) and for “certain” (non-inconclusive) sigmoid outputs.

Results: The proportion of “inconclusive” test cases that had to be accepted to achieve a given balanced accuracy in the “certain” test case was lower with RVT and AVT than with MVT in all datasets (e.g., 1.9\% and 1.2\% versus 2.8\% for 98\% balanced accuracy in “certain” test cases from the development dataset). In addition, RVT and AVT resulted in slightly higher balanced accuracy in all test cases independent of their certainty (97.3\% and 97.5\% versus 97.0\% in the development dataset).

Conclusion: Making between-readers-discrepancy known to CNN during the training improves the utility of their sigmoid output to discriminate certain from inconclusive cases that might be misclassified by the CNN when the predefined decision threshold is strictly applied. This does not compromise on overall accuracy.},
  keywords = {Deep learning, Convolutional neural network, Uncertainty, FP-CIT, Dopamine transporter, SPECT}
}

@article{disalvo2025_109531,
  title = {An Embedding is Worth a Thousand Noisy Labels},
  author = {Di Salvo, Francesco and Doerrich, Sebastian and Ledig, Christian and Rieger, Ines},
  year = {2025},
  pages = {1-19},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-109531x},
  url = {https://fis.uni-bamberg.de/handle/uniba/109531},
  language = {eng},
  abstract = {The performance of deep neural networks scales with dataset size and label quality, rendering the efficient mitigation of low-quality data annotations crucial for building robust and cost-effective systems. Existing strategies to address label noise exhibit severe limitations due to computational complexity and application dependency. In this work, we propose WANN, a Weighted Adaptive Nearest Neighbor approach that builds on self-supervised feature representations obtained from foundation models. To guide the weighted voting scheme, we introduce a reliability score, which measures the likelihood of a data label being correct. WANN outperforms reference methods, including a linear layer trained with robust loss functions, on diverse datasets of varying size and under various noise types and severities. WANN also exhibits superior generalization on imbalanced data compared to both Adaptive-NNs (ANN) and fixed k-NNs. Furthermore, the proposed weighting scheme enhances supervised dimensionality reduction under noisy labels. This yields a significant boost in classification performance with 10x and 100x smaller image embeddings, minimizing latency and storage requirements. Our approach, emphasizing efficiency and explainability, emerges as a simple, robust solution to overcome inherent limitations of deep neural network training.},
  keywords = {Adaptive kNN, Label noise, Robustness, Foundation models}
}

@inproceedings{disalvo2025_109532,
  title = {MedMNIST-C : Comprehensive benchmark and improved classifier robustness by simulating realistic image corruptions},
  author = {Di Salvo, Francesco and Doerrich, Sebastian and Ledig, Christian},
  year = {2025},
  pages = {1-11},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-109532x},
  url = {https://fis.uni-bamberg.de/handle/uniba/109532},
  language = {eng},
  abstract = {The integration of neural-network-based systems into clinical practice is limited by challenges related to domain generalization and robustness. The computer vision community established benchmarks such as ImageNet-C as a fundamental prerequisite to measure progress towards those challenges. Similar datasets are largely absent in the medical imaging community which lacks a comprehensive benchmark that spans across imaging modalities and applications. To address this gap, we create and open-source MedMNIST-C, a benchmark dataset based on the MedMNIST+ collection covering 12 datasets and 9 imaging modalities. We simulate task and modality-specific image corruptions of varying severity to comprehensively evaluate the robustness of established algorithms against real-world artifacts and distribution shifts. We further provide quantitative evidence that our simple-to-use artificial corruptions allow for highly performant, lightweight data augmentation to enhance model robustness. Unlike traditional, generic augmentation strategies, our approach leverages domain knowledge, exhibiting significantly higher robustness when compared to widely adopted methods. By introducing MedMNIST-C and open-sourcing the corresponding library allowing for targeted data augmentations, we contribute to the development of increasingly robust methods tailored to the challenges of medical imaging.},
  keywords = {Robustness, Corruptions, Data augmentation}
}

@inproceedings{doerrich2025_109529,
  title = {Unsupervised Feature Orthogonalization for Learning Distortion-Invariant Representations},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Ledig, Christian},
  year = {2025},
  pages = {1-13},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-109529x},
  url = {https://fis.uni-bamberg.de/handle/uniba/109529},
  language = {eng},
  abstract = {This study introduces unORANIC+, a novel method that integrates unsupervised feature orthogonalization with the ability of a Vision Transformer to capture both local and global relationships for improved robustness and generalizability. The streamlined architecture of unORANIC+ effectively separates anatomical and image-specific attributes, resulting in robust and unbiased latent representations that allow the model to demonstrate excellent performance across various medical image analysis tasks and diverse datasets. Extensive experimentation demonstrates unORANIC+’s reconstruction proficiency, corruption resilience, as well as capability to revise existing image distortions. Additionally, the model exhibits notable aptitude in downstream tasks such as disease classification and corruption detection. We confirm its adaptability to diverse datasets of varying image sources and sample sizes which positions the method as a promising algorithm for advanced medical image analysis, particularly in resourceconstrained environments lacking large, tailored datasets. The source code is available at github.com/sdoerrich97/unoranic-plus.},
  keywords = {feature orthogonalization, robustness, corruption revision, unsupervised learning, generalization}
}

@inproceedings{disalvo2025_109530,
  title = {Privacy-preserving datasets by capturing feature distributions with Conditional VAEs},
  author = {Di Salvo, Francesco and Tafler, David E. and Doerrich, Sebastian and Ledig, Christian},
  year = {2025},
  pages = {1-14},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-109530x},
  url = {https://fis.uni-bamberg.de/handle/uniba/109530},
  language = {eng},
  abstract = {Large and well-annotated datasets are essential for advancing deep learning applications, however often costly or impossible to obtain by a single entity. In many areas, including the medical domain, approaches relying on data sharing have become critical to address those challenges. While effective in increasing dataset size and diversity, data sharing raises significant privacy concerns. Commonly employed anonymization methods based on the k-anonymity paradigm often fail to preserve data diversity, affecting model robustness. This work introduces a novel approach using Conditional Variational Autoencoders (CVAEs) trained on feature vectors extracted from large pre-trained vision foundation models. Foundation models effectively detect and represent complex patterns across diverse domains, allowing the CVAE to faithfully capture the embedding space of a given data distribution to generate (sample) a diverse, privacy-respecting, and potentially unbounded set of synthetic feature vectors. Our method notably outperforms traditional approaches in both medical and natural image domains, exhibiting greater dataset diversity and higher robustness against perturbations while preserving sample privacy. These results underscore the potential of generative models to significantly impact deep learning applications in data-scarce and privacy-sensitive environments.},
  keywords = {Generative models, Privacy, Foundation models, Robustness}
}

@inproceedings{pan2025_110553,
  title = {An End-to-End AI Pipeline for Wood Knot Detection to Enhance Structural Assessment in Historic Timber Structures},
  author = {Pan, Junquan and Chizhova, Maria and Ebener, Frank and Luhmann, Thomas and Ledig, Christian and Maiwald, Ferdinand and Eißing, Thomas},
  year = {2025},
  volume = {X-M-2-2025},
  pages = {267-274},
  publisher = {Copernicus Publications},
  address = {Katlenburg-Lindau},
  issn = {2194-9050},
  doi = {10.5194/isprs-annals-x-m-2-2025-267-2025},
  url = {https://fis.uni-bamberg.de/handle/uniba/110553},
  language = {eng},
  abstract = {The accurate detection and assessment of wood surface defects in historic timber structures, particularly knots, is essential for effective conservation and strengthening planning. However, the application of automated visual grading methods to aged timber remains underexplored due to the irregular texture, weathering and lack of relevant datasets. In this study, we propose an end-to-end deep learning-based pipeline that integrates wood surface segmentation, perspective correction, and knot detection to estimate structural grading factors. A dedicated raw data collection of over 10,000 high-resolution images of historic timber surfaces was compiled using both DSLR cameras and mobile devices, resulting in multiple datasets with approximately 3,000 annotated samples. Three model families, YOLO, Detectron2 and DeepLabV3, were evaluated under different experimental setups. Beyond model benchmarking, we further compared the AI-derived results with expert manual measurements. The model for segmentation of timber surface achieved a mean IoU of over 0.85 and the model for detection of historical wood knots reached F1 scores of up to 0.9. The structural assessment factors estimated by the AI pipeline achieved a Pearson correlation coefficient of 0.641 compared to manual measurements, indicating a moderate level of consistency in knot factor estimation. This research highlights the potential of vision-based AI systems in supporting structural diagnosis and conservation of heritage timber elements.},
  keywords = {Photogrammetry, Deep Learning, Historic Timber Structures, Wood Knot Detection, Structural Assessment}
}

@article{disalvo2025_109535,
  title = {An Embedding is Worth a Thousand Noisy Labels},
  author = {Di Salvo, Francesco and Doerrich, Sebastian and Rieger, Ines and Ledig, Christian},
  year = {2025},
  journal = {Transactions on machine learning research},
  number = {3},
  pages = {1-19},
  publisher = {OpenReview.net},
  address = {Amherst, Massachusetts},
  issn = {2835-8856},
  url = {https://fis.uni-bamberg.de/handle/uniba/109535},
  language = {eng},
  abstract = {The performance of deep neural networks scales with dataset size and label quality, rendering the efficient mitigation of low-quality data annotations crucial for building robust and cost-effective systems. Existing strategies to address label noise exhibit severe limitations due to computational complexity and application dependency. In this work, we propose WANN, a Weighted Adaptive Nearest Neighbor approach that builds on self-supervised feature representations obtained from foundation models. To guide the weighted voting scheme, we introduce a reliability score, which measures the likelihood of a data label being correct. WANN outperforms reference methods, including a linear layer trained with robust loss functions, on diverse datasets of varying size and under various noise types and severities. WANN also exhibits superior generalization on imbalanced data compared to both Adaptive-NNs (ANN) and fixed k-NNs. Furthermore, the proposed weighting scheme enhances supervised dimensionality reduction under noisy labels. This yields a significant boost in classification performance with 10x and 100x smaller image embeddings, minimizing latency and storage requirements. Our approach, emphasizing efficiency and explainability, emerges as a simple, robust solution to overcome inherent limitations of deep neural network training.},
  keywords = {Adaptive kNN, Label noise, Robustness, Foundation models}
}

@article{doerrich2025_107030,
  title = {Rethinking model prototyping through the MedMNIST+ dataset collection},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Brockmann, Julius and Ledig, Christian},
  year = {2025},
  pages = {1-15},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-1070309},
  url = {https://fis.uni-bamberg.de/handle/uniba/107030},
  language = {eng},
  abstract = {The integration of deep learning based systems in clinical practice is often impeded by challenges rooted in limited and heterogeneous medical datasets. In addition, the field has increasingly prioritized marginal performance gains on a few, narrowly scoped benchmarks over clinical applicability, slowing down meaningful algorithmic progress. This trend often results in excessive fine-tuning of existing methods on selected datasets rather than fostering clinically relevant innovations. In response, this work introduces a comprehensive benchmark for the MedMNIST+ dataset collection, designed to diversify the evaluation landscape across several imaging modalities, anatomical regions, classification tasks and sample sizes. We systematically reassess commonly used Convolutional Neural Networks (CNNs) and Vision Transformer (ViT) architectures across distinct medical datasets, training methodologies, and input resolutions to validate and refine existing assumptions about model effectiveness and development. Our findings suggest that computationally efficient training schemes and modern foundation models offer viable alternatives to costly end-to-end training. Additionally, we observe that higher image resolutions do not consistently improve performance beyond a certain threshold. This highlights the potential benefits of using lower resolutions, particularly in prototyping stages, to reduce computational demands without sacrificing accuracy. Notably, our analysis reaffirms the competitiveness of CNNs compared to ViTs, emphasizing the importance of comprehending the intrinsic capabilities of different architectures. Finally, by establishing a standardized evaluation framework, we aim to enhance transparency, reproducibility, and comparability within the MedMNIST+ dataset collection as well as future research. Code is available at github.com/sdoerrich97/rethinking-model-prototyping-MedMNISTPlus.},
  keywords = {Benchmarking, Prototyping recommendations, Medical image classification, Standardized evaluation framework, Foundation models}
}

@inproceedings{dorrich2025_106934,
  title = {unORANIC : Unsupervised Orthogonalization of Anatomy and Image-Characteristic Features},
  author = {Dörrich, Sebastian and Di Salvo, Francesco and Ledig, Christian},
  year = {2025},
  pages = {1-11},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-1069345},
  url = {https://fis.uni-bamberg.de/handle/uniba/106934},
  language = {eng},
  abstract = {We introduce unORANIC, an unsupervised approach that uses an adapted loss function to drive the orthogonalization of anatomy and image-characteristic features. The method is versatile for diverse modalities and tasks, as it does not require domain knowledge, paired data samples, or labels. During test time unORANIC is applied to potentially corrupted images, orthogonalizing their anatomy and characteristic components, to subsequently reconstruct corruption-free images, showing their domain-invariant anatomy only. This feature orthogonalization further improves generalization and robustness against corruptions. We confirm this qualitatively and quantitatively on 5 distinct datasets by assessing unORANIC’s classification accuracy, corruption detection and revision capabilities. Our approach shows promise for enhancing the generalizability and robustness of practical applications in medical image analysis. The source code is available at github.com/sdoerrich97/unORANIC.},
  keywords = {feature orthogonalization, robustness, corruption revision, unsupervised learning, generalization}
}

@article{doerrich2025_107047,
  title = {Rethinking model prototyping through the MedMNIST+ dataset collection},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Brockmann, Julius and Ledig, Christian},
  year = {2025},
  journal = {Scientific reports},
  volume = {15},
  number = {1, 7669},
  pages = {1-15},
  publisher = {Springer Nature},
  address = {London},
  issn = {2045-2322},
  doi = {10.1038/s41598-025-92156-9},
  url = {https://fis.uni-bamberg.de/handle/uniba/107047},
  language = {eng},
  abstract = {The integration of deep learning based systems in clinical practice is often impeded by challenges rooted in limited and heterogeneous medical datasets. In addition, the field has increasingly prioritized marginal performance gains on a few, narrowly scoped benchmarks over clinical applicability, slowing down meaningful algorithmic progress. This trend often results in excessive fine-tuning of existing methods on selected datasets rather than fostering clinically relevant innovations. In response, this work introduces a comprehensive benchmark for the MedMNIST+ dataset collection, designed to diversify the evaluation landscape across several imaging modalities, anatomical regions, classification tasks and sample sizes. We systematically reassess commonly used Convolutional Neural Networks (CNNs) and Vision Transformer (ViT) architectures across distinct medical datasets, training methodologies, and input resolutions to validate and refine existing assumptions about model effectiveness and development. Our findings suggest that computationally efficient training schemes and modern foundation models offer viable alternatives to costly end-to-end training. Additionally, we observe that higher image resolutions do not consistently improve performance beyond a certain threshold. This highlights the potential benefits of using lower resolutions, particularly in prototyping stages, to reduce computational demands without sacrificing accuracy. Notably, our analysis reaffirms the competitiveness of CNNs compared to ViTs, emphasizing the importance of comprehending the intrinsic capabilities of different architectures. Finally, by establishing a standardized evaluation framework, we aim to enhance transparency, reproducibility, and comparability within the MedMNIST+ dataset collection as well as future research. Code is available at github.com/sdoerrich97/rethinking-model-prototyping-MedMNISTPlus.},
  keywords = {Benchmarking, Prototyping recommendations, Medical image classification, Standardized evaluation framework, Foundation models}
}

@inproceedings{disalvo2024_109551,
  title = {Privacy-preserving datasets by capturing feature distributions with Conditional VAEs},
  author = {Di Salvo, Francesco and Tafler, David and Doerrich, Sebastian and Ledig, Christian},
  year = {2024},
  booktitle = {Proceedings : The 35th British Machine Vision Conference 2024},
  pages = {1-14},
  url = {https://fis.uni-bamberg.de/handle/uniba/109551},
  language = {eng},
  abstract = {Large and well-annotated datasets are essential for advancing deep learning applications, however often costly or impossible to obtain by a single entity. In many areas, including the medical domain, approaches relying on data sharing have become critical to address those challenges. While effective in increasing dataset size and diversity, data sharing raises significant privacy concerns. Commonly employed anonymization methods based on the k-anonymity paradigm often fail to preserve data diversity, affecting model robustness. This work introduces a novel approach using Conditional Variational Autoencoders (CVAEs) trained on feature vectors extracted from large pre-trained vision foundation models. Foundation models effectively detect and represent complex patterns across diverse domains, allowing the CVAE to faithfully capture the embedding space of a given data distribution to generate (sample) a diverse, privacy-respecting, and potentially unbounded set of synthetic feature vectors. Our method notably outperforms traditional approaches in both medical and natural image domains, exhibiting greater dataset diversity and higher robustness against perturbations while preserving sample privacy. These results underscore the potential of generative models to significantly impact deep learning applications in data-scarce and privacy-sensitive environments.},
  keywords = {Generative models, Privacy, Foundation models, Robustness}
}

@inproceedings{doerrich2024_109550,
  title = {Unsupervised Feature Orthogonalization for Learning Distortion-Invariant Representations},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Ledig, Christian},
  year = {2024},
  booktitle = {Workshop Proceedings : International Workshop on Robust Recognition in the Open World at The 35th British Machine Vision Conference},
  pages = {1-13},
  url = {https://fis.uni-bamberg.de/handle/uniba/109550},
  language = {eng},
  abstract = {This study introduces unORANIC+, a novel method that integrates unsupervised feature orthogonalization with the ability of a Vision Transformer to capture both local and global relationships for improved robustness and generalizability. The streamlined architecture of unORANIC+ effectively separates anatomical and image-specific attributes, resulting in robust and unbiased latent representations that allow the model to demonstrate excellent performance across various medical image analysis tasks and diverse datasets. Extensive experimentation demonstrates unORANIC+’s reconstruction proficiency, corruption resilience, as well as capability to revise existing image distortions. Additionally, the model exhibits notable aptitude in downstream tasks such as disease classification and corruption detection. We confirm its adaptability to diverse datasets of varying image sources and sample sizes which positions the method as a promising algorithm for advanced medical image analysis, particularly in resourceconstrained environments lacking large, tailored datasets. The source code is available at github.com/sdoerrich97/unoranic-plus.},
  keywords = {feature orthogonalization, robustness, corruption revision, unsupervised learning, generalization}
}

@inproceedings{disalvo2024_109539,
  title = {MedMNIST-C : Comprehensive benchmark and improved classifier robustness by simulating realistic image corruptions},
  author = {Di Salvo, Francesco and Doerrich, Sebastian and Ledig, Christian},
  year = {2024},
  pages = {1-11},
  doi = {10.48550/arXiv.2406.17536},
  url = {https://fis.uni-bamberg.de/handle/uniba/109539},
  language = {eng},
  abstract = {The integration of neural-network-based systems into clinical practice is limited by challenges related to domain generalization and robustness. The computer vision community established benchmarks such as ImageNet-C as a fundamental prerequisite to measure progress towards those challenges. Similar datasets are largely absent in the medical imaging community which lacks a comprehensive benchmark that spans across imaging modalities and applications. To address this gap, we create and open-source MedMNIST-C, a benchmark dataset based on the MedMNIST+ collection covering 12 datasets and 9 imaging modalities. We simulate task and modality-specific image corruptions of varying severity to comprehensively evaluate the robustness of established algorithms against real-world artifacts and distribution shifts. We further provide quantitative evidence that our simple-to-use artificial corruptions allow for highly performant, lightweight data augmentation to enhance model robustness. Unlike traditional, generic augmentation strategies, our approach leverages domain knowledge, exhibiting significantly higher robustness when compared to widely adopted methods. By introducing MedMNIST-C and open-sourcing the corresponding library allowing for targeted data augmentations, we contribute to the development of increasingly robust methods tailored to the challenges of medical imaging.},
  keywords = {Robustness, Corruptions, Data augmentation}
}

@article{kucerenko2024_106141,
  title = {Incorporating label uncertainty during the training of convolutional neural networks improves performance for the discrimination between certain and inconclusive cases in dopamine transporter SPECT},
  author = {Kucerenko, Aleksej and Buddenkotte, Thomas and Apostolova, Ivayla and Klutmann, Susanne and Ledig, Christian and Buchert, Ralph},
  year = {2024},
  journal = {European Journal of Nuclear Medicine and Molecular Imaging},
  volume = {52},
  number = {4},
  pages = {1535-1548},
  publisher = {Springer},
  address = {Heidelberg [u.a.]},
  issn = {1619-7089},
  doi = {10.1007/s00259-024-06988-0},
  url = {https://fis.uni-bamberg.de/handle/uniba/106141},
  language = {eng},
  abstract = {Purpose: Deep convolutional neural networks (CNN) hold promise for assisting the interpretation of dopamine transporter (DAT)-SPECT. For improved communication of uncertainty to the user it is crucial to reliably discriminate certain from inconclusive cases that might be misclassified by strict application of a predefined decision threshold on the CNN output. This study tested two methods to incorporate existing label uncertainty during the training to improve the utility of the CNN sigmoid output for this task.

Methods: Three datasets were used retrospectively: a “development” dataset (n = 1740) for CNN training, validation and testing, two independent out-of-distribution datasets (n = 640, 645) for testing only. In the development dataset, binary classification based on visual inspection was performed carefully by three well-trained readers. A ResNet-18 architecture was trained for binary classification of DAT-SPECT using either a randomly selected vote (“random vote training”, RVT), the proportion of “reduced” votes ( “average vote training”, AVT) or the majority vote (MVT) across the three readers as reference standard. Balanced accuracy was computed separately for “inconclusive” sigmoid outputs (within a predefined interval around the 0.5 decision threshold) and for “certain” (non-inconclusive) sigmoid outputs.

Results: The proportion of “inconclusive” test cases that had to be accepted to achieve a given balanced accuracy in the “certain” test case was lower with RVT and AVT than with MVT in all datasets (e.g., 1.9\% and 1.2\% versus 2.8\% for 98\% balanced accuracy in “certain” test cases from the development dataset). In addition, RVT and AVT resulted in slightly higher balanced accuracy in all test cases independent of their certainty (97.3\% and 97.5\% versus 97.0\% in the development dataset).

Conclusion: Making between-readers-discrepancy known to CNN during the training improves the utility of their sigmoid output to discriminate certain from inconclusive cases that might be misclassified by the CNN when the predefined decision threshold is strictly applied. This does not compromise on overall accuracy.},
  keywords = {Deep learning, Convolutional neural network, Uncertainty, FP-CIT, Dopamine transporter, SPECT}
}

@inproceedings{doerrich2024_104846,
  title = {Self-supervised Vision Transformer are Scalable Generative Models for Domain Generalization},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Ledig, Christian},
  year = {2024},
  pages = {1-10},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-1048464},
  url = {https://fis.uni-bamberg.de/handle/uniba/104846},
  language = {eng},
  abstract = {Despite notable advancements, the integration of deep learning (DL) techniques into impactful clinical applications, particularly in the realm of digital histopathology, has been hindered by challenges associated with achieving robust generalization across diverse imaging domains and characteristics. Traditional mitigation strategies in this field such as data augmentation and stain color normalization have proven insufficient in addressing this limitation, necessitating the exploration of alternative methodologies. To this end, we propose a novel generative method for domain generalization in histopathology images. Our method employs a generative, self-supervised Vision Transformer to dynamically extract characteristics of image patches and seamlessly infuse them into the original images, thereby creating novel, synthetic images with diverse attributes. By enriching the dataset with such synthesized images, we aim to enhance its holistic nature, facilitating improved generalization of DL models to unseen domains. Extensive experiments conducted on two distinct histopathology datasets demonstrate the effectiveness of our proposed approach, outperforming the state of the art substantially, on the Camelyon17-wilds challenge dataset (+2\%) and on a second epithelium-stroma dataset (+26\%). Furthermore, we emphasize our method's ability to readily scale with increasingly available unlabeled data samples and more complex, higher parametric architectures. Source code is available at https://github.com/sdoerrich97/vits-are-generative-models.},
  keywords = {domain generalization, self-supervised learning, feature orthogonalization, generative image synthesis}
}

@inproceedings{doerrich2024_105289,
  title = {Self-supervised Vision Transformer are Scalable Generative Models for Domain Generalization},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Ledig, Christian},
  year = {2024},
  booktitle = {Medical Image Computing and Computer Assisted Intervention – MICCAI 2024},
  pages = {644-654},
  publisher = {Springer Nature Switzerland},
  address = {Cham},
  isbn = {978-3-031-72117-5},
  doi = {10.1007/978-3-031-72117-5\_60},
  url = {https://fis.uni-bamberg.de/handle/uniba/105289},
  language = {eng},
  abstract = {Despite notable advancements, the integration of deep learning (DL) techniques into impactful clinical applications, particularly in the realm of digital histopathology, has been hindered by challenges associated with achieving robust generalization across diverse imaging domains and characteristics. Traditional mitigation strategies in this field such as data augmentation and stain color normalization have proven insufficient in addressing this limitation, necessitating the exploration of alternative methodologies. To this end, we propose a novel generative method for domain generalization in histopathology images. Our method employs a generative, self-supervised Vision Transformer to dynamically extract characteristics of image patches and seamlessly infuse them into the original images, thereby creating novel, synthetic images with diverse attributes. By enriching the dataset with such synthesized images, we aim to enhance its holistic nature, facilitating improved generalization of DL models to unseen domains. Extensive experiments conducted on two distinct histopathology datasets demonstrate the effectiveness of our proposed approach, outperforming the state of the art substantially, on the Camelyon17-wilds challenge dataset (+2\%) and on a second epithelium-stroma dataset (+26\%). Furthermore, we emphasize our method's ability to readily scale with increasingly available unlabeled data samples and more complex, higher parametric architectures. Source code is available at https://github.com/sdoerrich97/vits-are-generative-models.},
  keywords = {domain generalization, self-supervised learning, feature orthogonalization, generative image synthesis}
}

@inproceedings{doerrich2024_104752,
  title = {Integrating kNN with Foundation Models for Adaptable and Privacy-Aware Image Classification},
  author = {Doerrich, Sebastian and Archut, Tobias and Di Salvo, Francesco and Ledig, Christian},
  year = {2024},
  publisher = {Otto-Friedrich-Universität},
  address = {Bamberg},
  urn = {urn:nbn:de:bvb:473-irb-1047524},
  url = {https://fis.uni-bamberg.de/handle/uniba/104752},
  language = {eng},
  abstract = {Traditional deep learning models implicitly encode knowledge limiting their transparency and ability to adapt to data changes. Yet, this adaptability is vital for addressing user data privacy concerns. We address this limitation by storing embeddings of the underlying training data independently of the model weights, enabling dynamic data modifications without retraining. Specifically, our approach integrates the k-Nearest Neighbor (k-NN) classifier with a vision-based foundation model, pre-trained self-supervised on natural images, enhancing interpretability and adaptability. We share open-source implementations of a previously unpublished baseline method as well as our performance-improving contributions. Quantitative experiments confirm improved classification across established benchmark datasets and the method’s applicability to distinct medical image classification tasks. Additionally, we assess the method’s robustness in continual learning and data removal scenarios. The approach exhibits great promise for bridging the gap between foundation models’ performance and challenges tied to data privacy. The source code is available at github.com/TobArc.},
  keywords = {k-NN classifier, continual learning, transfer learning, few-shot classification, explainability}
}

@inproceedings{doerrich2024_104176,
  title = {Integrating kNN with Foundation Models for Adaptable and Privacy-Aware Image Classification},
  author = {Doerrich, Sebastian and Archut, Tobias and Di Salvo, Francesco and Ledig, Christian},
  year = {2024},
  booktitle = {2024 IEEE International Symposium on Biomedical Imaging (ISBI)},
  pages = {1-5},
  publisher = {IEEE},
  doi = {10.1109/isbi56570.2024.10635560},
  url = {https://fis.uni-bamberg.de/handle/uniba/104176},
  language = {eng},
  abstract = {Traditional deep learning models implicitly encode knowledge limiting their transparency and ability to adapt to data changes. Yet, this adaptability is vital for addressing user data privacy concerns. We address this limitation by storing embeddings of the underlying training data independently of the model weights, enabling dynamic data modifications without retraining. Specifically, our approach integrates the k-Nearest Neighbor (k-NN) classifier with a vision-based foundation model, pre-trained self-supervised on natural images, enhancing interpretability and adaptability. We share open-source implementations of a previously unpublished baseline method as well as our performance-improving contributions. Quantitative experiments confirm improved classification across established benchmark datasets and the method’s applicability to distinct medical image classification tasks. Additionally, we assess the method’s robustness in continual learning and data removal scenarios. The approach exhibits great promise for bridging the gap between foundation models’ performance and challenges tied to data privacy. The source code is available at github.com/TobArc.},
  keywords = {k-NN classifier, continual learning, transfer learning, few-shot classification, explainability}
}

@inproceedings{doerrich2023_91842,
  title = {unORANIC : Unsupervised Orthogonalization of Anatomy and Image-Characteristic Features},
  author = {Doerrich, Sebastian and Di Salvo, Francesco and Ledig, Christian},
  year = {2023},
  booktitle = {Machine Learning in Medical Imaging : 14th International Workshop, MLMI 2023, Held in Conjunction with MICCAI 2023, Vancouver, BC, Canada, October 8, 2023 ; Proceedings, Part I},
  pages = {62-71},
  publisher = {Springer Nature Switzerland},
  address = {Cham},
  isbn = {978-3-031-45672-5},
  issn = {1611-3349},
  doi = {10.1007/978-3-031-45673-2\_7},
  url = {https://fis.uni-bamberg.de/handle/uniba/91842},
  language = {eng},
  abstract = {We introduce unORANIC, an unsupervised approach that uses an adapted loss function to drive the orthogonalization of anatomy and image-characteristic features. The method is versatile for diverse modalities and tasks, as it does not require domain knowledge, paired data samples, or labels. During test time unORANIC is applied to potentially corrupted images, orthogonalizing their anatomy and characteristic components, to subsequently reconstruct corruption-free images, showing their domain-invariant anatomy only. This feature orthogonalization further improves generalization and robustness against corruptions. We confirm this qualitatively and quantitatively on 5 distinct datasets by assessing unORANIC’s classification accuracy, corruption detection and revision capabilities. Our approach shows promise for enhancing the generalizability and robustness of practical applications in medical image analysis. The source code is available at github.com/sdoerrich97/unORANIC.},
  keywords = {feature orthogonalization, robustness, corruption revision, unsupervised learning, generalization}
}

@article{huizinga2021_54056,
  title = {Differences Between MR Brain Region Segmentation Methods: Impact on Single-Subject Analysis},
  author = {Huizinga, W. and Poot, D. H. J. and Vinke, E. J. and Wenzel, F. and Bron, E. E. and Toussaint, N. and Ledig, Christian and Vrooman, H. and Ikram, M. A. and Niessen, W. J. and Vernooij, M. W. and Klein, S.},
  year = {2021},
  journal = {Frontiers in big data},
  volume = {4},
  number = {577164},
  publisher = {Frontiers Media},
  address = {Lausanne},
  issn = {2624-909X},
  doi = {10.3389/fdata.2021.577164},
  url = {https://fis.uni-bamberg.de/handle/uniba/54056},
  language = {eng}
}

@incollection{wolterink2020_55000,
  title = {Deep learning: Generative adversarial networks and adversarial methods},
  author = {Wolterink, Jelmer M. and Kamnitsas, Konstantinos and Ledig, Christian and Išgum, Ivana},
  year = {2020},
  booktitle = {Handbook of medical image computing and computer assisted intervention},
  pages = {547-574},
  publisher = {Academic Press},
  address = {London},
  edition = {1},
  isbn = {9780128165867},
  doi = {10.1016/B978-0-12-816176-0.00028-4},
  url = {https://fis.uni-bamberg.de/handle/uniba/55000},
  language = {eng},
  keywords = {Medizin, Bildverarbeitung}
}

@article{thomas2020_54994,
  title = {Integrative Analysis of Circulating Metabolite Profiles and Magnetic Resonance Imaging Metrics in Patients with Traumatic Brain Injury},
  author = {Thomas, Ilias and Dickens, Alex M. and Posti, Jussi P. and Mohammadian, Mehrbod and Ledig, Christian and Takala, Riikka S. K. and Hyötyläinen, Tuulia and Tenovuo, Olli and Orešič, Matej},
  year = {2020},
  journal = {International journal of molecular sciences},
  volume = {21},
  number = {4},
  issn = {1422-0067},
  doi = {10.3390/ijms21041395},
  url = {https://fis.uni-bamberg.de/handle/uniba/54994},
  language = {eng}
}

@article{biffi2020_54993,
  title = {Explainable Anatomical Shape Analysis Through Deep Hierarchical Generative Models},
  author = {Biffi, Carlo and Cerrolaza, Juan J. and Tarroni, Giacomo and Bai, Wenjia and De Marvao, Antonio and Oktay, Ozan and Ledig, Christian and Le Folgoc, Loic and Kamnitsas, Konstantinos and Doumou, Georgia and Duan, Jinming and Prasad, Sanjay K. and Cook, Stuart A and O'Regan, Declan P. and Rueckert, Daniel},
  year = {2020},
  journal = {IEEE transactions on medical imaging},
  volume = {39},
  number = {6},
  pages = {2088-2099},
  issn = {1558-254X},
  doi = {10.1109/TMI.2020.2964499},
  url = {https://fis.uni-bamberg.de/handle/uniba/54993},
  language = {eng}
}

@article{ostberg2020_54992,
  title = {Volume Change in Frontal Cholinergic Structures After Traumatic Brain Injury and Cognitive Outcome},
  author = {Östberg, Anna and Ledig, Christian and Katila, Ari and Maanpää, Henna-Riikka and Posti, Jussi P. and Takala, Riikka and Tallus, Jussi and Glocker, Ben and Rueckert, Daniel and Tenovuo, Olli},
  year = {2020},
  journal = {Frontiers in neurology},
  volume = {11},
  number = {832},
  issn = {1664-2295},
  doi = {10.3389/fneur.2020.00832},
  url = {https://fis.uni-bamberg.de/handle/uniba/54992},
  language = {eng}
}

@article{jones2020_54057,
  title = {Assessment of a deep-learning system for fracture detection in musculoskeletal radiographs},
  author = {Jones, Rebecca M. and Sharma, Anuj and Hotchkiss, Robert and Sperling, John W. and Hamburger, Jackson and Ledig, Christian and O'Toole, Robert and Gardner, Michael and Venkatesh, Srivas and Roberts, Matthew M. and Sauvestre, Romain and Shatkhin, Max and Gupta, Anant and Chopra, Sumit and Kumaravel, Manickam and Daluiski, Aaron and Plogger, Will and Nascone, Jason and Potter, Hollis G. and Lindsey, Robert V.},
  year = {2020},
  journal = {NPJ digital medicine},
  volume = {3},
  number = {144},
  publisher = {Macmillan Publishers Limited},
  address = {Basingstoke},
  issn = {2398-6352},
  doi = {10.1038/s41746-020-00352-w},
  url = {https://fis.uni-bamberg.de/handle/uniba/54057},
  language = {eng}
}

@article{wang2019_54995,
  title = {Quantitative assessment of myelination patterns in preterm neonates using T2-weighted MRI},
  author = {Wang, Siying and Ledig, Christian and Hajnal, Joseph V. and Counsell, Serena J. and Schnabel, Julia A. and Deprez, Maria},
  year = {2019},
  journal = {Scientific reports},
  volume = {9},
  number = {1},
  issn = {2045-2322},
  doi = {10.1038/s41598-019-49350-3},
  url = {https://fis.uni-bamberg.de/handle/uniba/54995},
  language = {eng}
}

@misc{deprez2018_59140,
  title = {Segmentation of Myelin-like Signals on Clinical MR Images for Age Estimation in Preterm Infants},
  author = {Deprez, Maria and Wang, Siying and Ledig, Christian and Hajnal, Joseph V. and Counsell, Serena J. and Schnabel, Julia A.},
  year = {2018},
  doi = {10.1101/357749},
  url = {https://fis.uni-bamberg.de/handle/uniba/59140},
  language = {eng}
}

@misc{wolterink2018_59134,
  title = {Generative adversarial networks and adversarial methods in biomedical image analysis},
  author = {Wolterink, Jelmer M. and Kamnitsas, Konstantinos and Ledig, Christian and Išgum, Ivana},
  year = {2018},
  pages = {1-29},
  publisher = {arXiv},
  doi = {10.48550/ARXIV.1810.10352},
  url = {https://fis.uni-bamberg.de/handle/uniba/59134},
  language = {eng}
}

@article{tolonen2018_54997,
  title = {Data-Driven Differential Diagnosis of Dementia Using Multiclass Disease State Index Classifier},
  author = {Tolonen, Antti and Rhodius-Meester, Hanneke F. M. and Bruun, Marie and Koikkalainen, Juha and Barkhof, Frederik and Lemstra, Afina W. and Koene, Teddy and Scheltens, Philip and Teunissen, Charlotte E. and Tong, Tong and Guerrero, Ricardo and Schuh, Andreas and Ledig, Christian and Baroni, Marta and Rueckert, Daniel and Soininen, Hilkka and Remes, Anne M. and Waldemar, Gunhild and Hasselbalch, Steen G. and Mecocci, Patrizia and Van der Flier, Wiesje M. and Lötjönen, Jyrki},
  year = {2018},
  journal = {Frontiers in aging neuroscience},
  volume = {10},
  issn = {1663-4365},
  doi = {10.3389/fnagi.2018.00111},
  url = {https://fis.uni-bamberg.de/handle/uniba/54997},
  language = {eng}
}

@article{ledig2018_54996,
  title = {Structural brain imaging in Alzheimer's disease and mild cognitive impairment: biomarker analysis and shared morphometry database},
  author = {Ledig, Christian and Schuh, Andreas and Guerrero, Ricardo and Heckemann, Rolf A. and Rueckert, Daniel},
  year = {2018},
  journal = {Scientific reports},
  volume = {8},
  number = {1},
  issn = {2045-2322},
  doi = {10.1038/s41598-018-29295-9},
  url = {https://fis.uni-bamberg.de/handle/uniba/54996},
  language = {eng}
}

@incollection{ledig2016_55001,
  title = {Semantic Parsing of Brain MR Images},
  author = {Ledig, Christian and Rueckert, Daniel},
  year = {2016},
  pages = {307-335},
  publisher = {Elsevier},
  address = {Amsterdam},
  isbn = {9780128025819},
  doi = {10.1016/B978-0-12-802581-9.00014-7},
  url = {https://fis.uni-bamberg.de/handle/uniba/55001},
  language = {deu}
}

@inproceedings{lotjonen2012_58622,
  title = {Hippocampal atrophy rate using an expectation maximization classifier with a disease-specific prior},
  author = {Lötjönen, Jyrki and Wolz, Robin and Koikkalainen, Juha and Manna, Valeria and Ledig, Christian and Thurfjell, Lennart and Lundqvist, Roger and Waldemar, Gunhild and Soininen, Hilkka and Rueckert, Daniel},
  year = {2012},
  pages = {1164-1167},
  doi = {10.1109/ISBI.2012.6235767},
  url = {https://fis.uni-bamberg.de/handle/uniba/58622},
  language = {eng}
}

