Nadine S. Schaadt’s research while affiliated with Hannover Medical School and other places

Predicting the biological behavior and time to recurrence (TTR) of high-grade diffuse gliomas (HGG) after maximum safe neurosurgical resection and combined radiation and chemotherapy plays a pivotal role in planning clinical follow-up, selecting potentially necessary second-line treatment and improving the quality of life for patients diagnosed with a malignant brain tumor. The current standard-of-care (SoC) for HGG includes follow-up neuroradiological imaging to detect recurrence as early as possible and relies on several clinical, neuropathological, and radiological prognostic factors, which have limited accuracy in predicting TTR. In this study, using an in-silico analysis, we aim to improve predictive power for TTR by considering the role of (i) prognostically relevant information available through diagnostics used in the current SoC, (ii) advanced image-based information not currently part of the standard diagnostic workup, such as tumor-normal tissue interface (edge) features and quantitative data specific to biopsy positions within the tumor, and (iii) information on tumor-associated macrophages. In particular, we introduced a state-of-the-art spatio-temporal model of tumor-immune interactions, emphasizing the interplay between macrophages and glioma cells. This model serves as a synthetic reality for assessing the predictive value of various features. We generated a cohort of virtual patients based on our mathematical model. Each patient's dataset includes simulated T1Gd and Fluid-attenuated inversion recovery (FLAIR) MRI volumes. T1-weighted imaging highlights anatomical structures with high contrast, providing clear detail on brain morphology, whereas FLAIR suppresses fluid signals, improving the visualization of lesions near fluid-filled spaces, which is particularly helpful for identifying peritumoral edema. Additionally, we simulated results on macrophage density and proliferative activity, either in a specified part of the tumor, namely the tumor core or edge ("localized"), or unspecified ("non-localized"). To enhance the realism of our synthetic data, we imposed different levels of noise. Our findings reveal that macrophage density at the tumor edge contributed to a high predictive value of feature importance for the selected regression model. Moreover, there are lower MSE values for the "localized" biopsy in prediction accuracy toward recurrence post-resection compared with "non-localized" specimens in the noisy data. In conclusion, the results show that localized biopsies provided more information about tumor behavior, especially at the interface of tumor and normal tissue (Edge). High-grade diffuse gliomas (HGG) include the most common types of primary malignant brain tumors in adults, namely glioblastoma multiforme (GBM), astrocytoma grade 3/grade 4 (A°4/A°3, and oligodendroglioma grade 3 (O°3) 1. Glioblastoma is associated with the most unfavorable prognosis and poor significant therapeutic advances in the past few decades 2 , and A°4/A°3 with better prognosis but a similar need for aggressive therapy and almost inevitable recurrence 3-6. HGG shares their potential for diffuse invasion of adjacent CNS structures and their pheno-typical plasticity. In contrast, phenotypic plasticity is recognized as one of the hallmarks of cancer 7 and can be associated with tissue invasion in many A full list of affiliations appears at the end of the paper.

In recent years, there has been remarkable progress in the field of digital pathology, driven by the ability to model complex tissue patterns using advanced deep-learning algorithms. However, the robustness of these models is often severely compromised in the presence of data shifts (e.g., different stains, organs, centers, etc.). Alternatively, continual learning (CL) techniques aim to reduce the forgetting of past data when learning new data with distributional shift conditions. Specifically, rehearsal-based CL techniques, which store some past data in a buffer and then replay it with new data, have proven effective in medical image analysis tasks. However, privacy concerns arise as these approaches store past data, prompting the development of our novel Generative Latent Replay-based CL (GLRCL) approach. GLRCL captures the previous distribution through Gaussian Mixture Models instead of storing past samples, which are then utilized to generate features and perform latent replay with new data. We systematically evaluate our proposed framework under different shift conditions in histopathology data, including stain and organ shift. Our approach significantly outperforms popular buffer-free CL approaches and performs similarly to rehearsal-based CL approaches that require large buffers causing serious privacy violations.

In recent years, the diagnosis of gliomas has become increasingly complex. Analysis of glioma histopathology images using artificial intelligence (AI) offers new opportunities to support diagnosis and outcome prediction. To give an overview of the current state of research, this review examines 83 publicly available research studies that have proposed AI-based methods for whole-slide histopathology images of human gliomas, covering the diagnostic tasks of subtyping (23/83), grading (27/83), molecular marker prediction (20/83), and survival prediction (29/83). All studies were reviewed with regard to methodological aspects as well as clinical applicability. It was found that the focus of current research is the assessment of hematoxylin and eosin-stained tissue sections of adult-type diffuse gliomas. The majority of studies (52/83) are based on the publicly available glioblastoma and low-grade glioma datasets from The Cancer Genome Atlas (TCGA) and only a few studies employed other datasets in isolation (16/83) or in addition to the TCGA datasets (15/83). Current approaches mostly rely on convolutional neural networks (63/83) for analyzing tissue at 20x magnification (35/83). A new field of research is the integration of clinical data, omics data, or magnetic resonance imaging (29/83). So far, AI-based methods have achieved promising results, but are not yet used in real clinical settings. Future work should focus on the independent validation of methods on larger, multi-site datasets with high-quality and up-to-date clinical and molecular pathology annotations to demonstrate routine applicability.

In digital pathology, deep learning (DL) models for tasks such as segmentation or tissue classification are known to suffer from domain shifts due to different staining techniques. Stain adaptation aims to reduce the generalization error between different stains by training a model on source stains that generalizes to target stains. Despite the abundance of target stain data, a key challenge is the lack of annotations. To address this, we propose a joint training between artificially labeled and unlabeled data including all available stained images called Unsupervised Latent Stain Adaption (ULSA). Our method uses stain translation to enrich labeled source images with synthetic target images in order to increase supervised signals. Moreover, we leverage unlabeled target stain images using stain-invariant feature consistency learning. With ULSA we present a semi-supervised strategy for efficient stain adaption without access to annotated target stain data. Remarkably, ULSA is task agnostic in patch-level analysis for whole slide images (WSIs). Through extensive evaluation on external datasets, we demonstrate that ULSA achieves state-of-the-art (SOTA) performance in kidney tissue segmentation and breast cancer classification across a spectrum of staining variations. Our findings suggest that ULSA is an important framework towards stain adaption in digital pathology.

Crowdsourcing has been used in computational pathology to generate cell and cell nuclei annotations for machine learning. Herein, we broaden its scope to the previously unsolved challenging task of glioma cell detection. This requires multiplexed immunofluorescence microscopy due to diffuse invasiveness and exceptional similarity between glioma cells and reactive astrocytes. In four pilot experiments, we iteratively developed a task design enabling high-quality annotations by crowdworkers on Amazon Mechanical Turk. We applied majority or weighted vote and validated them against ground truth in the final setting. On the base of a YOLO convolutional neural network architecture, we used these consensus labels for training with different image representations regarding colors, intensities, and immmunohistochemical marker combinations. A crowd of 712 workers defined aggregated point annotations in 235 images with an average F1$F_1$ score of 0.627 for majority vote. The networks resulted in acceptable F1$F_1$ scores up to 0.69 for YOLOv8 on average and indicated first evidence for transferability to images lacking tumor markers, especially in IDH-wildtype glioblastoma. Our work confirms feasibility of crowdsourcing to generate labels suitable for training of machine learning tools in the challenging and clinically relevant use case of glioma microenvironment.