@article{bertram_performance_2026,

title = {Performance evaluation of deep learning models for image analysis: Considerations for visual assessment and statistical metrics},

author = {Christof A. Bertram and Jonas Ammeling and Alexander Bartel and Gillian Beamer and Marc Aubreville},

url = {https://journals.sagepub.com/doi/10.1177/03009858261461760},

doi = {10.1177/03009858261461760},

issn = {0300-9858, 1544-2217},

year  = {2026},

date = {2026-07-01},

urldate = {2026-07-07},

journal = {Veterinary Pathology},

pages = {03009858261461760},

abstract = {Deep learning-based automated image analysis (DL-AIA) has been shown to outperform trained pathologists in tasks related to feature quantification. Related to these capabilities, the use of DL-AIA tools is currently extending from proof-of-principle studies to routine applications, including the evaluation of patient samples (diagnostic pathology), regulatory safety assessment (toxicologic pathology), and recurrent research tasks. To ensure that DL-AIA applications are safe and reliable, it is critical to conduct a thorough and objective generalization performance assessment to evaluate an algorithm’s ability to accurately predict patterns of interest and possibly evaluate model robustness (ie, the algorithm’s capacity to maintain predictive accuracy on images from different sources). In this article, we review the practices for performance assessment in veterinary pathology publications by which 2 main approaches were identified: (1) exclusive visual performance assessment (ie, eyeballing algorithmic predictions) plus validation of the model’s application utilizing secondary performance indices and (2) statistical performance assessment (alongside other methods), which requires creation of a test set with ground truth labels that is a hold-out from model development. This article compares the strengths and weaknesses of statistical and visual performance assessment methods. Furthermore, we discuss relevant considerations for rigorous statistical performance evaluation including metric selection, test data set image composition, ground truth label quality, resampling methods such as bootstrapping, statistical comparison of multiple models, and evaluation of model stability. It is our conclusion that visual and statistical evaluation have complementary strengths and a combined approach provides the greatest insight into the DL model’s performance and sources of error.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{bertram_data_2026,

title = {Data set creation for supervised deep learning–based analysis of microscopic images: Review of important considerations and recommendations},

author = {Christof A. Bertram and Viktoria Weiss and Jonas Ammeling and F. Maria Schabel and Taryn A. Donovan and Frauke Wilm and Christian Marzahl and Katharina Breininger and Marc Aubreville},

url = {https://journals.sagepub.com/doi/10.1177/03009858261457959},

doi = {10.1177/03009858261457959},

issn = {0300-9858, 1544-2217},

year  = {2026},

date = {2026-06-01},

urldate = {2026-06-30},

journal = {Veterinary Pathology},

pages = {03009858261457959},

abstract = {Supervised deep learning (DL) receives great interest for automated analysis of microscopic images with an increasing body of literature supporting its potential. The development and testing of those DL models rely heavily on the availability of high-quality, large-scale data sets. However, creating such data sets is a complex and resource-intensive process, often hindered by challenges such as time constraints, domain variability, and risks of bias in image collection and label creation. This review provides a comprehensive guide to the critical steps in data set creation, including (1) image acquisition, (2) selection of annotation software, and (3) annotation creation. For image acquisition, besides ensuring a sufficiently large number, it is important to address sources of image variability (domain shifts), such as those related to slide preparation and digitization, that could lead to algorithmic errors if not adequately represented in the training data. For annotations, key quality criteria are the 3 “C”s: correctness, consistency, and completeness. For mitigation of annotation bias of a single annotator, this review explores advanced annotation methods (eg, computer-assisted annotations). To support data set creators, a standard operating procedure is provided as supplemental material, summarizing all important considerations for data set creation. Furthermore, this article underscores the importance of open data sets in driving innovation and enhancing reproducibility of DL research. By addressing the challenges and offering practical recommendations, this review aims to advance the creation and availability of high-quality, large-scale data sets, ultimately contributing to the development of generalizable and robust DL models for pathology applications.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{banerjee_reporting_2026,

title = {Reporting transparency in veterinary pathology deep learning: A systematic review of reproducibility-critical details},

author = {Sweta Banerjee and Christof A. Bertram and Viktoria Weiss and Jonas Ammeling and Thomas Conrad and Nils Porsche and Robert Klopfleisch and Christoph Stroblberger and Christopher Kaltenecker and Katharina Breininger and Marc Aubreville},

url = {https://journals.sagepub.com/doi/10.1177/03009858261459452},

doi = {10.1177/03009858261459452},

issn = {0300-9858, 1544-2217},

year  = {2026},

date = {2026-06-01},

urldate = {2026-06-30},

journal = {Veterinary Pathology},

pages = {03009858261459452},

abstract = {Whereas reproducibility of studies is a prerequisite for trustworthy deep learning (DL) in veterinary histopathology and microscopy, the actual degree of methodological transparency that exists in the literature remains uncertain. We performed a Preferred Reporting Items for Systematic Reviews and Meta-Analyses-guided systematic review to quantify the degree to which supervised DL and supervised machine learning studies report reproducibility-critical details. Using a veterinary-journal-restricted Boolean search executed in PubMed and Scopus, we screened 180 unique records and included 50 primary research articles for full-text analysis. Based on a recently published guideline for the development of DL models in veterinary pathology, we extracted information for each study across 5 dimensions: (1) study and task characterization, (2) data transparency, (3) experimental design and data-leakage control, (4) model and training details, and (5) performance evaluation and reporting. Among the included studies, private data sets predominated, with 90% of studies relying on private data. Sharing of code was uncommon (3%). Key training details such as augmentation and hyperparameters were often incompletely reported; augmentation was not reported in 56% of studies, and key hyperparameters were absent in 40% of studies. It was often not clear whether patient-level stratification (necessary to avoid data leakage) was performed. In summary, these results highlight major deficits in the reporting of details and experimental design necessary for reproducing DL results in veterinary histopathology. This review provides a practical baseline and reporting roadmap to support more transparent and reproducible research in veterinary computational pathology.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{rosbach_stuck_2026,

title = {Stuck on Suggestions: Automation Bias, the Anchoring Effect, and the Factors That Shape Them in Computational Pathology},

author = {Emely Rosbach and Jonas Ammeling and Jonathan Ganz and Christof A. Bertram and Thomas Conrad and Andreas Riener and Marc Aubreville},

doi = {https://doi.org/10.59275/j.melba.2026-87b1},

year  = {2026},

date = {2026-03-01},

journal = {Machine Learning for Biomedical Imaging},

volume = {3},

number = {MELBA–BVM 2025 Special Issue},

pages = {126–147},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{ivan_subphase-labeled_2026,

title = {A Subphase-Labeled Mitotic Dataset for AI-powered Cell Division Analysis},

author = {Zsanett Zsofia Ivan and Dominik Hirling and Istvan Grexa and Jonas Ammeling and Csaba Molnar and Tamas Micsik and Katalin Dobra and Levente Kuthi and Farkas Sukosd and Janos Fillinger and Judit Moldvay and Erika Toth and Marc Aubreville and Vivien Miczan and Peter Horvath},

url = {https://www.nature.com/articles/s41597-026-07007-7},

doi = {10.1038/s41597-026-07007-7},

issn = {2052-4463},

year  = {2026},

date = {2026-03-01},

urldate = {2026-03-21},

journal = {Scientific Data},

abstract = {Abstract 

 Mitosis detection represents a critical task in digital pathology, as it plays an important role in the tumor grading and prognosis of patients. Manual determination is a labor-intensive task for practitioners with high interobserver variability, thus, automation is a priority. There has been substantial progress towards creating robust mitosis detection algorithms, primarily driven by the Mitosis Domain Generalization (MIDOG) challenges. Also, there has been growing interest in the molecular characterization of mitosis to achieve a more comprehensive understanding of its underlying mechanisms in a subphase-specific manner. We introduce a new mitotic figure dataset annotated with subphase information based on the MIDOG++ dataset as well as a previously unrepresented tumor domain to enhance the diversity and applicability. We envision a new perspective for domain generalization by improving model performance with subtyping mitosis, complemented with an atypical mitotic class. Our work has implications in two main areas: subtyping information can provide helpful information in mitosis detection, while also providing promising new directions in answering biological questions, such as molecular analysis of subphases.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{ammeling_benchmarking_2026,

title = {Benchmarking Foundation Models for Mitotic Figure Classification},

author = {Jonas Ammeling and Jonathan Ganz and Emely Rosbach and Ludwig Lausser and Christof A. Bertram and Katharina Breininger and Marc Aubreville},

url = {https://melba-journal.org/2026:003},

doi = {10.59275/j.melba.2026-a3eb},

issn = {2766-905X},

year  = {2026},

date = {2026-01-01},

urldate = {2026-02-01},

journal = {Machine Learning for Biomedical Imaging},

volume = {3},

number = {MELBA–BVM 2025 Special Issue},

pages = {38–55},

abstract = {The performance of deep learning models is known to scale with data quantity and diversity. In pathology, as in many other medical imaging domains, the availability of labeled images for a specific task is often limited. Self-supervised learning techniques have enabled the use of vast amounts of unlabeled data to train large-scale neural networks, i.e., foundation models, that can address the limited data problem by providing semantically rich feature vectors that can generalize well to new tasks with minimal training effort increasing model performance and robustness. In this work, we investigate the use of foundation models for mitotic figure classification. The mitotic count, which can be derived from this classification task, is an independent prognostic marker for specific tumors and part of certain tumor grading systems. In particular, we investigate the data scaling laws on multiple current foundation models and evaluate their robustness to unseen tumor domains. Next to the commonly used linear probing paradigm, we also adapt the models using low-rank adaptation (LoRA) of their attention mechanisms. We compare all models against end-to-end-trained baselines, both CNNs and Vision Transformers. Our results demonstrate that LoRA-adapted foundation models provide superior performance to those adapted with standard linear probing, reaching performance levels close to 100 % data availability with only 10 % of training data. Furthermore, LoRA-adaptation of the most recent foundation models almost closes the out-of-domain performance gap when evaluated on unseen tumor domains. However, full fine-tuning of traditional architectures still yields competitive performance.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{banerjee_benchmarking_2026,

title = {Benchmarking Deep Learning and Vision Foundation Models for Atypical vs. Normal Mitosis Classification with Cross-Dataset Evaluation},

author = {Sweta Banerjee and Viktoria Weiss and Taryn A. Donovan and Rutger H. J. Fick and Thomas Conrad and Jonas Ammeling and Nils Porsche and Robert Klopfleisch and Christopher Kaltenecker and Katharina Breininger and Marc Aubreville and Christof A Bertram},

doi = {https://doi.org/10.59275/j.melba.2026-6c1g},

year  = {2026},

date = {2026-01-01},

journal = {Machine Learning for Biomedical Imaging},

number = {MELBA–BVM 2025 Special Issue},

pages = {115–125},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{ammeling_investigation_2025,

title = {Investigation of Class Separability within Object Detection Models in Histopathology},

author = {Jonas Ammeling and Jonathan Ganz and Frauke Wilm and Katharina Breininger and Marc Aubreville},

url = {https://ieeexplore.ieee.org/document/10965484/},

doi = {10.1109/TMI.2025.3560134},

issn = {0278-0062, 1558-254X},

year  = {2025},

date = {2025-08-01},

urldate = {2025-04-16},

journal = {IEEE Transactions on Medical Imaging},

volume = {44},

number = {8},

pages = {3162–3174},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{ammeling_interdisciplinary_2025,

title = {An interdisciplinary perspective on AI-supported decision making in medicine},

author = {Jonas Ammeling and Marc Aubreville and Alexis Fritz and Angelika Kießig and Sebastian Krügel and Matthias Uhl},

url = {https://linkinghub.elsevier.com/retrieve/pii/S0160791X24003397},

doi = {10.1016/j.techsoc.2024.102791},

issn = {0160791X},

year  = {2025},

date = {2025-06-01},

urldate = {2024-12-08},

journal = {Technology in Society},

volume = {81},

pages = {102791},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{ammeling_interdisciplinary_2024,

title = {An Interdisciplinary Perspective on AI-Supported Decision Making in Medicine},

author = {Jonas Ammeling and Marc Aubreville and Alexis Fritz and Angelika Kießig and Sebastian Krügel and Matthias Uhl},

url = {https://linkinghub.elsevier.com/retrieve/pii/S0160791X24003397},

doi = {10.1016/j.techsoc.2024.102791},

issn = {0160791X},

year  = {2024},

date = {2024-12-01},

urldate = {2024-12-06},

journal = {Technology in Society},

pages = {102791},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{ganz_information_2024,

title = {Information mismatch in PHH3-assisted mitosis annotation leads to interpretation shifts in H&E slide analysis},

author = {Jonathan Ganz and Christian Marzahl and Jonas Ammeling and Emely Rosbach and Barbara Richter and Chloé Puget and Daniela Denk and Elena A. Demeter and Flaviu A. Tăbăran and Gabriel Wasinger and Karoline Lipnik and Marco Tecilla and Matthew J. Valentine and Michael J. Dark and Niklas Abele and Pompei Bolfa and Ramona Erber and Robert Klopfleisch and Sophie Merz and Taryn A. Donovan and Samir Jabari and Christof A. Bertram and Katharina Breininger and Marc Aubreville},

url = {https://www.nature.com/articles/s41598-024-77244-6},

doi = {10.1038/s41598-024-77244-6},

issn = {2045-2322},

year  = {2024},

date = {2024-11-01},

urldate = {2024-11-04},

journal = {Scientific Reports},

volume = {14},

number = {1},

pages = {26273},

abstract = {Abstract 

 The count of mitotic figures (MFs) observed in hematoxylin and eosin (H&E)-stained slides is an important prognostic marker, as it is a measure for tumor cell proliferation. However, the identification of MFs has a known low inter-rater agreement. In a computer-aided setting, deep learning algorithms can help to mitigate this, but they require large amounts of annotated data for training and validation. Furthermore, label noise introduced during the annotation process may impede the algorithms’ performance. Unlike H&E, where identification of MFs is based mainly on morphological features, the mitosis-specific antibody phospho-histone H3 (PHH3) specifically highlights MFs. Counting MFs on slides stained against PHH3 leads to higher agreement among raters and has therefore recently been used as a ground truth for the annotation of MFs in H&E. However, as PHH3 facilitates the recognition of cells indistinguishable from H&E staining alone, the use of this ground truth could potentially introduce an interpretation shift and even label noise into the H&E-related dataset, impacting model performance. This study analyzes the impact of PHH3-assisted MF annotation on inter-rater reliability and object level agreement through an extensive multi-rater experiment. Subsequently, MF detectors, including a novel dual-stain detector, were evaluated on the resulting datasets to investigate the influence of PHH3-assisted labeling on the models’ performance. We found that the annotators’ object-level agreement significantly increased when using PHH3-assisted labeling (F1: 0.53 to 0.74). However, this enhancement in label consistency did not translate to improved performance for H&E-based detectors, neither during the training phase nor the evaluation phase. Conversely, the dual-stain detector was able to benefit from the higher consistency. This reveals an information mismatch between the H&E and PHH3-stained images as the cause of this effect, which renders PHH3-assisted annotations not well-aligned for use with H&E-based detectors. Based on our findings, we propose an improved PHH3-assisted labeling procedure.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{aubreville_prediction_2024,

title = {Prediction of tumor board procedural recommendations using large language models},

author = {Marc Aubreville and Jonathan Ganz and Jonas Ammeling and Emely Rosbach and Thomas Gehrke and Agmal Scherzad and Stephan Hackenberg and Miguel Goncalves},

url = {https://doi.org/10.1007/s00405-024-08947-9},

doi = {10.1007/s00405-024-08947-9},

issn = {1434-4726},

year  = {2024},

date = {2024-09-01},

journal = {European Archives of Oto-Rhino-Laryngology},

abstract = {Multidisciplinary tumor boards are meetings where a team of medical specialists, including medical oncologists, radiation oncologists, radiologists, surgeons, and pathologists, collaborate to determine the best treatment plan for cancer patients. While decision-making in this context is logistically and cost-intensive, it has a significant positive effect on overall cancer survival.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{ganz_re-identification_2024,

title = {Re-identification from histopathology images},

author = {Jonathan Ganz and Jonas Ammeling and Samir Jabari and Katharina Breininger and Marc Aubreville},

url = {https://linkinghub.elsevier.com/retrieve/pii/S1361841524002603},

doi = {10.1016/j.media.2024.103335},

issn = {13618415},

year  = {2024},

date = {2024-09-01},

urldate = {2024-09-20},

journal = {Medical Image Analysis},

pages = {103335},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{aubreville_domain_2024,

title = {Domain generalization across tumor types, laboratories, and species — Insights from the 2022 edition of the Mitosis Domain Generalization Challenge},

author = {Marc Aubreville and Nikolas Stathonikos and Taryn A. Donovan and Robert Klopfleisch and Jonas Ammeling and Jonathan Ganz and Frauke Wilm and Mitko Veta and Samir Jabari and Markus Eckstein and Jonas Annuscheit and Christian Krumnow and Engin Bozaba and Sercan Çayır and Hongyan Gu and Xiang ‘Anthony’ Chen and Mostafa Jahanifar and Adam Shephard and Satoshi Kondo and Satoshi Kasai and Sujatha Kotte and V. G. Saipradeep and Maxime W. Lafarge and Viktor H. Koelzer and Ziyue Wang and Yongbing Zhang and Sen Yang and Xiyue Wang and Katharina Breininger and Christof A. Bertram},

url = {https://linkinghub.elsevier.com/retrieve/pii/S136184152400080X},

doi = {10.1016/j.media.2024.103155},

issn = {13618415},

year  = {2024},

date = {2024-05-01},

urldate = {2024-03-27},

journal = {Medical Image Analysis},

volume = {94},

pages = {103155},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{krugel_perceived_2024-1,

title = {Perceived responsibility in AI-supported medicine},

author = {Sebastian Krügel and Jonas Ammeling and Marc Aubreville and Alexis Fritz and Angelika Kießig and Matthias Uhl},

url = {https://link.springer.com/10.1007/s00146-024-01972-6},

doi = {10.1007/s00146-024-01972-6},

issn = {0951-5666, 1435-5655},

year  = {2024},

date = {2024-05-01},

urldate = {2024-05-01},

journal = {AI & SOCIETY},

abstract = {Abstract 

 In a representative vignette study in Germany with 1,653 respondents, we investigated laypeople’s attribution of moral responsibility in collaborative medical diagnosis. Specifically, we compare people’s judgments in a setting in which physicians are supported by an AI-based recommender system to a setting in which they are supported by a human colleague. It turns out that people tend to attribute moral responsibility to the artificial agent, although this is traditionally considered a category mistake in normative ethics. This tendency is stronger when people believe that AI may become conscious at some point. In consequence, less responsibility is attributed to human agents in settings with hybrid diagnostic teams than in settings with human-only diagnostic teams. Our findings may have implications for behavior exhibited in contexts of collaborative medical decision making with AI-based as opposed to human recommenders because less responsibility is attributed to agents who have the mental capacity to care about outcomes.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{krugel_perceived_2024,

title = {Perceived responsibility in AI-supported medicine},

author = {Sebastian Krügel and Jonas Ammeling and Marc Aubreville and Alexis Fritz and Angelika Kießig and Matthias Uhl},

url = {https://link.springer.com/10.1007/s00146-024-01972-6},

doi = {10.1007/s00146-024-01972-6},

issn = {0951-5666, 1435-5655},

year  = {2024},

date = {2024-05-01},

urldate = {2024-05-23},

journal = {AI & SOCIETY},

abstract = {Abstract 

 In a representative vignette study in Germany with 1,653 respondents, we investigated laypeople’s attribution of moral responsibility in collaborative medical diagnosis. Specifically, we compare people’s judgments in a setting in which physicians are supported by an AI-based recommender system to a setting in which they are supported by a human colleague. It turns out that people tend to attribute moral responsibility to the artificial agent, although this is traditionally considered a category mistake in normative ethics. This tendency is stronger when people believe that AI may become conscious at some point. In consequence, less responsibility is attributed to human agents in settings with hybrid diagnostic teams than in settings with human-only diagnostic teams. Our findings may have implications for behavior exhibited in contexts of collaborative medical decision making with AI-based as opposed to human recommenders because less responsibility is attributed to agents who have the mental capacity to care about outcomes.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{aubreville_comprehensive_2023,

title = {A comprehensive multi-domain dataset for mitotic figure detection},

author = {Marc Aubreville and Frauke Wilm and Nikolas Stathonikos and Katharina Breininger and Taryn A. Donovan and Samir Jabari and Mitko Veta and Jonathan Ganz and Jonas Ammeling and Paul J. Van Diest and Robert Klopfleisch and Christof A. Bertram},

url = {https://www.nature.com/articles/s41597-023-02327-4},

doi = {10.1038/s41597-023-02327-4},

issn = {2052-4463},

year  = {2023},

date = {2023-07-01},

urldate = {2023-07-26},

journal = {Scientific Data},

volume = {10},

number = {1},

pages = {484},

abstract = {Abstract 

 The prognostic value of mitotic figures in tumor tissue is well-established for many tumor types and automating this task is of high research interest. However, especially deep learning-based methods face performance deterioration in the presence of domain shifts, which may arise from different tumor types, slide preparation and digitization devices. We introduce the MIDOG++ dataset, an extension of the MIDOG 2021 and 2022 challenge datasets. We provide region of interest images from 503 histological specimens of seven different tumor types with variable morphology with in total labels for 11,937 mitotic figures: breast carcinoma, lung carcinoma, lymphosarcoma, neuroendocrine tumor, cutaneous mast cell tumor, cutaneous melanoma, and (sub)cutaneous soft tissue sarcoma. The specimens were processed in several laboratories utilizing diverse scanners. We evaluated the extent of the domain shift by using state-of-the-art approaches, observing notable differences in single-domain training. In a leave-one-domain-out setting, generalizability improved considerably. This mitotic figure dataset is the first that incorporates a wide domain shift based on different tumor types, laboratories, whole slide image scanners, and species.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}