Ramona Woitek

Nuclei instance segmentation in hematoxylin and eosin (H&E)-stained images plays an important role in automated histological image analysis, with various applications in downstream tasks. While several machine learning and deep learning approaches have been proposed for nuclei instance segmentation, most research in this field focuses on developing new segmentation algorithms and benchmarking them on a limited number of arbitrarily selected public datasets. In this work, rather than focusing on model development, we focused on the datasets used for this task. Based on an extensive literature review, we identified manually annotated, publicly available datasets of H&E-stained images for nuclei instance segmentation and standardized them into a unified input and annotation format. Using two state-of-the-art segmentation models, one based on convolutional neural networks (CNNs) and one based on a hybrid CNN and vision transformer architecture, we systematically evaluated and ranked these datasets based on their nuclei instance segmentation performance. Furthermore, we proposed a unified test set (NucFuse-test) for fair cross-dataset evaluation and a unified training set (NucFuse-train) for improved segmentation performance by merging images from multiple datasets. To the best of our knowledge, this is the first study to systematically benchmark and rank publicly available datasets for nuclei instance segmentation. By evaluating and ranking the datasets, performing comprehensive analyses, generating fused datasets, conducting external validation, and making our implementation publicly available, we provided a new protocol for training, testing, and evaluating nuclei instance segmentation models on H&E-stained histological images.

With the advancement of computers and digitalization, many medical records and data, including medical images, can now be stored and analyzed using semi- or fully automated computerized approaches

Nuclei instance segmentation in hematoxylin and eosin (H&E)-stained images plays an important role in automated histological image analysis, with various applications in downstream tasks. While several machine learning and deep learning approaches have been proposed for nuclei instance segmentation, most research in this field focuses on developing new segmentation algorithms and benchmarking them on a limited number of arbitrarily selected public datasets. In this work, rather than focusing on model development, we focused on the datasets used for this task. Based on an extensive literature review, we identified manually annotated, publicly available datasets of H&E-stained images for nuclei instance segmentation and standardized them into a unified input and annotation format. Using two state-of-the-art segmentation models, one based on convolutional neural networks (CNNs) and one based on a hybrid CNN and vision transformer architecture, we systematically evaluated and ranked these datasets based on their nuclei instance segmentation performance. Furthermore, we proposed a unified test set (NucFuse-test) for fair cross-dataset evaluation and a unified training set (NucFuse-train) for improved segmentation performance by merging images from multiple datasets. By evaluating and ranking the datasets, performing comprehensive analyses, generating fused datasets, conducting external validation, and making our implementation publicly available, we provided a new benchmark for training, testing, and evaluating nuclei instance segmentation models on H&E-stained histological images.

Despite recent therapeutic advances, renal cell carcinoma (RCC) has a high mortality rate. The development of new neoadjuvant strategies requires reliable companion biomarkers of early and successful response to treatment. A change in tumor size is a late measure of response and novel targeted imaging-based biomarkers may be more accurate for treatment response prediction. Here we evaluated the potential of hyperpolarized carbon-13 MRI (HP ¹³ C-MRI), following the injection of hyperpolarized ¹³ C-pyruvate, to assess response to neoadjuvant treatment in four patients with clear cell RCC as an exploratory outcome within a prospective clinical trial. The change in the tumor lactate-to-pyruvate ratio (LAC/PYR) following treatment varied across the patients: mean percentage change ± S.D. = +6 ± 27%. The largest decrease in LAC/PYR ratio was in the patient treated with cediranib monotherapy (-21%), followed by a smaller reduction in a patient receiving the combination of cediranib and olaparib (-14%). An increase in the LAC/PYR ratio post-treatment was observed in the second patient receiving combination treatment (+ 21%), with the largest increase in the patient receiving olaparib monotherapy (+ 35%). Metabolic changes were observed following treatment in the absence of significant changes in tumor size. In summary, HP ¹³ C-MRI successfully captured heterogeneous metabolic responses to cediranib and olaparib therapy, revealing both increases and decreases in tumor lactate labelling, independent of any morphologic change. Finally, this is the first study to evaluate the potential of clinical HP ¹³ C-MRI to assess early treatment response in renal cancer by using a range of therapeutics. Key take home message: The study provides preliminary evidence supporting HP ¹³ C-MRI as a promising imaging biomarker for evaluating early metabolic changes in renal cell carcinoma following neoadjuvant therapy. Clinical Trials Registry: NCT03741426, Registration date: 13 November 2018.

Histopathological image classification using computational methods such as fine-tuned convolutional neural networks (CNNs) has gained significant attention in recent years. Graph neural networks (GNNs) have also emerged as strong alternatives, often employing CNNs or vision transformers (ViTs) as node feature extractors. However, as these models are usually pre-trained on small-scale natural image datasets, their performance in histopathology tasks can be limited. The introduction of foundation models trained on large-scale histopathological data now enables more effective feature extraction for GNNs. In this work, we integrate recently developed foundation models as feature extractors within a lightweight GNN and compare their performance with standard fine-tuned CNN and ViT models. Furthermore, we explore a prediction fusion approach that combines the outputs of the best-performing GNN and fine-tuned model to evaluate the benefits of complementary representations. Results demonstrate that GNNs utilizing foundation model features outperform those trained with CNN or ViT features and achieve performance comparable to standard fine-tuned CNN and ViT models. The highest overall performance is obtained with the proposed prediction fusion strategy. Evaluated on three publicly available datasets, the best fusion achieved F1-scores of 98.04%, 96.51%, and 98.28%, and balanced accuracies of 98.03%, 96.50%, and 97.50% on PanNuke, BACH, and BreakHis, respectively.

Objective: Sperm DNA fragmentation (SDF) tests such as Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) and Sperm Chromatin Structure Assay, which is reported by indices such as the DNA fragmentation Index (DFI), are valuable tools in the assessment of male infertility. However, they are underutilized due to their limited accessibility and high cost. Identifying candidates who are unlikely to benefit from SDF testing could optimize clinical decision-making and reduce unnecessary expenses. This study aims to predict DFI and TUNEL values using routinely available semen parameters and age, which are easily accessible and cost-effective. Additionally, it seeks to develop a clinical decision support system (CDSS) to assist in guiding recommendations for SDF testing. Methods: Utilizing a newly proposed dataset of 10,000 infertile men, we develop a machine learning pipeline that includes pre-processing, feature engineering, and the development of models for predicting DFI and TUNEL values. This pipeline enables the establishment of a comprehensive benchmark and introduces a novel composite feature called the Unhealthy Sperm Score (USS), which serves as a risk indicator of sperm dysfunction. Additionally, a CDSS was constructed to identify patients with low DNA fragmentation risk (DFI < 30 and TUNEL < 20) based on their sperm profile and age. Results: The USS score demonstrated the highest correlation with both DFI and TUNEL, outperforming single semen parameters, and proved to be the primary driver for SDF predictions. Additionally, the developed CDSS achieved high precision in identifying low-risk individuals (94.83% for DFI and 96.15% for TUNEL). %We tested the tool on 1,000 cases and successfully bypassed 624 patients who may safely defer DNA fragmentation testing. Conclusion: Our work offers a benchmark dataset on predicting SDF tests (DFI and TUNEL), and a first-of-its-kind CDSS model for separating low-risk patients from high-risk ones, introducing new avenues for cost-effective, personalized male infertility care. The dataset and implementation code are publicly available at: https://github.com/HealMaDe/SDF_prediction

Objectives: This study aimed to analyze variations in the sella turcica (ST) concerning its size, shape, and bridging, providing first reference values in Austrian individuals. Additionally, it assessed associations between these morphological and demographic parameters and their correlation with patients’ skeletal class. Methods: 208 lateral cephalometric radiographs (154 female, 54 male; age 8–58 years) from DPU Dental Clinic (Austria) were included. Size, skeletal class, shape, age, and gender of ST were tested for significance in correlation using, (M)ANOVA, and chi-square. Results: Linear dimensions of ST ranged from 11.1 to 12.9 mm across readers, with a standard deviation of 2.0–2.2 mm. Normal ST (49.76%) and round ST (58.77%) were the most frequent. ST bridging was detected in 6.97%. Skeletal class I appeared most frequently (54.8%). Statistical significance was observed between age, gender, and ST length, with further significant age effects on ST shape. Moreover, age showed significant modification of ST shape, while skeletal parameters appeared unaffected by other ST parameters. Conclusions: These preliminary findings define normal ST dimensions in an Austrian population, offering reference values for clinical interpretation and broadening the available European data. Clear associations between morphological and demographic parameters were detected. Additionally, these findings may contribute to diagnostic and therapeutic strategies in orthodontics and craniofacial pathology. Future studies employing cone beam computed tomography (CBCT) along a larger sample size could enhance the generalizability of these findings.