Flensburg University of Applied Sciences

Atypical mitosis marks a deviation in the cell division process that has been shown be an independent prognostic marker for tumor malignancy. However, atypical mitosis classification remains challenging due to low prevalence, at times subtle morphological differences from normal mitotic figures, low inter-rater agreement among pathologists, and class imbalance in datasets. Building on the Atypical Mitosis dataset for Breast Cancer (AMi-Br), this study presents a comprehensive benchmark comparing deep learning approaches for automated atypical mitotic figure (AMF) classification, including end-to-end fine-tuned deep learning models, foundation models with linear probing, and foundation models fine-tuned with low-rank adaptation (LoRA). For rigorous evaluation, we further introduce two new held-out AMF datasets - AtNorM-Br, a dataset of mitotic figures from the TCGA breast cancer cohort, and AtNorM-MD, a multi-domain dataset of mitotic figures from a subset of the MIDOG++ training set. We found average balanced accuracy values of up to 0.8135, 0.7788, and 0.7723 on the in-domain AMi-Br and the out-of-domain AtNorm-Br and AtNorM-MD datasets, respectively. Our work shows that atypical mitotic figure classification, while being a challenging problem, can be effectively addressed through the use of recent advances in transfer learning and model fine-tuning techniques. We make all code and data used in this paper available in this github repository: https://github.com/DeepMicroscopy/AMi-Br_Benchmark

Object detection is one of the most common tasks in histopathological image analysis and generalization is a key requirement for the clinical applicability of deep object detection models. However, traditional evaluation metrics often fail to provide insights into why models fail on certain test cases, especially in the presence of domain shifts. In this work, we propose a novel quantitative method for assessing the discriminative power of a model’s latent space. Our approach, applicable to all object detection models with known local correspondences such as the popular RetinaNet, FCOS, or YOLO approaches, allows tracing discrimination across layers and coordinates. We motivate, adapt, and evaluate two suitable metrics, the generalized discrimination value and the Hellinger distance, and incorporate them into our approach. Through empirical validation on real-world histopathology datasets, we demonstrate the effectiveness of our method in capturing model discrimination properties and providing insights for architectural optimization. This work contributes to bridging the gap between model performance evaluation and understanding the underlying mechanisms influencing model behavior.

For large-scale floating wind turbines, load and response validation enables a closed-loop design process, which helps to validate the simulation models and algorithms and provides a basis for improving subsequent designs. For this purpose, the characteristics of the freestream inflow wind are vital. The nacelle lidar system is easy to install, cost-effective, and capable of measuring the freestream wind in front of the rotor of a floating turbine. However, compared to bottom-fixed turbines, measurement errors in nacelle lidar caused by the motion of the floating platform require accurate motion measurements and complex reconstruction algorithms for motion compensation. In this paper, we investigate the potential of using nacelle lidar for load and response validation of floating turbines through numerical simulations. We propose a method for generating turbulent wind fields considering rotor-averaged wind speed estimated by simulated-lidar system as a constraint. This method does not require motion velocity measurements and allows for simplified wind field reconstruction. Based on aero-hydro-servo-elastic integrated simulations, this study demonstrates the potential of a nine-beam nacelle lidar system for load validation, including both sample-averaged statistical properties and one-to-one comparisons.

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.

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.