arXiv subjects

Thomas Walter

Publications and source records attributed to Thomas Walter.

16 recordsLinked to original sources

arXiv2026

Concept-based explanation of gene expression prediction from H&E images

Recent advances in pathology foundation models have enabled accurate prediction of spatial transcriptomics (ST) from routine H&E images. However, existing explainability methods for vision transformer (ViT)-based models are largely limited to local heatmaps and do not reveal how morphological concepts contribute to ST predictions. Here, we introduce an explainable framework that combines relevance propagation and concept discovery to link transcriptional programs to tissue morphology. We developed a ViT-based framework for virtual ST from H&E images that combines ViT-aware layer-wise relevance propagation with relaxed archetypal TopK sparse autoencoder-based concept discovery. This approach provides both local explanations and global insights into the morphological patterns associated with transcriptional programs. We applied the framework to colorectal cancer ST data from the HEST-1k cohort and evaluated its generalizability in TCGA COAD. Our architecture accurately predicts clinically relevant ST signatures and accompanying molecular phenotypes. Measured and predicted gene expression profiles reveal substantial spatial heterogeneity of the colorectal cancer subtypes iCMS2 and iCMS3 across a large number of samples. Spatially resolved and aggregated iCMS classification achieve weighted F1 scores of 0.872 and 0.819 (0.770 in TCGA COAD), respectively, and both stratify patient outcome. Beyond prediction, our framework establishes a relevance-based concept atlas linking molecular phenotypes to histopathological representations. Comparison of activation- with relevance-derived concepts demonstrates that relevances provide a more direct link between tissue morphology and downstream predictions. We establish a general strategy for concept-based explanation of spatial prediction, and our framework is readily applicable to a broad range of ViT-based pathology models.

cs.CV↗

arXiv2026

Mitosis Detection in the Wild: Multi-Tumor and Context-Aware Generalization in the MIDOG 2025 Challenge

Automated mitosis detection is a well-established task in computational pathology. While previous benchmarks focused on scanner-induced domain shift, clinical "real-world" application requires models to be robust across the vast variance to be expected in the histological landscape. The MItosis DOmain Generalization (MIDOG) 2025 challenge was designed to evaluate algorithmic performance across unprecedented biological and contextual diversity. We curated a test dataset of 365 cases, encompassing 12 distinct human, canine and feline tumor types, digitized across multiple scanning platforms. Moving beyond hand-selected hotspots, the challenge required detection also in random tissue areas (representative of the whole slide detection situation) and challenging areas (areas rich in hard negatives). In the second track, we introduced the classification of atypical mitotic figures (AMFs). There were 18 teams submitting to the detection track, with F1 scores ranging up to 0.740. In the AMF detection track, we had 21 submissions with balanced accuracy values up to 0.908. Our analysis reveals that while most models perform reliably in traditional hotspots, significant performance degradation occurs in challenging ROIs, where false positive rates tripled. Furthermore, performance varied significantly across the 12 tumor types, highlighting "blind spots" in current state-of-the-art architectures when encountering rare or highly pleomorphic malignancies. Moreover, we evaluated the effectiveness of ensembling and found a mean increases of 1.5 and 1.3 percentage points in F1 score and balanced accuracy, respectively. In contrast, TTA showed no relevant improvement. MIDOG 2025 demonstrates that "in the wild" mitosis detection remains a significant hurdle. The transition from hotspot-only evaluation to a multi-contextual framework provides a more realistic proxy for clinical reliability.

cs.CV↗

arXiv2026

LEMON: a foundation model for nuclear morphology in Computational Pathology

Computational pathology relies on effective representation learning to support cancer research and precision medicine. Although self-supervised learning has driven major progress at the patch and whole-slide image levels, representation learning at the single-cell level remains comparatively underexplored, despite its importance for characterizing cell types and cellular phenotypes. We introduce LEMON (Learning Embeddings from Morphology Of Nuclei), a self-supervised foundation model for scalable single-cell image representation learning. Trained on millions of cell images from diverse tissues and cancer types, LEMON learns robust and versatile morphological representations that support large-scale single-cell analyses in pathology. We evaluate LEMON on five benchmark datasets across a range of prediction tasks and show that it provides strong performance, highlighting its potential as a new paradigm for cell-level computational pathology. Model weights are available at https://huggingface.co/aliceblondel/LEMON.

cs.CV↗

arXiv2025

Robust Pan-Cancer Mitotic Figure Detection with YOLOv12

Mitotic figures represent a key histoprognostic feature in tumor pathology, providing crucial insights into tumor aggressiveness and proliferation. However, their identification remains challenging, subject to significant inter-observer variability, even among experienced pathologists. To address this issue, the MItosis DOmain Generalization (MIDOG) 2025 challenge marks the third edition of an international competition aiming to develop robust mitosis detection algorithms. In this paper, we present a mitotic figure detection approach based on the state-of-the-art YOLOv12 object detection architecture. Our method achieved an F1-score of 0.801 on the preliminary test set (hotspots only) and ranked second on the final test leaderboard with an F1-score of 0.7216 across complex and heterogeneous whole-slide regions, without relying on external data.

eess.IV↗

arXiv2025

ConvNeXt with Histopathology-Specific Augmentations for Mitotic Figure Classification

Accurate mitotic figure classification is crucial in computational pathology, as mitotic activity informs cancer grading and patient prognosis. Distinguishing atypical mitotic figures (AMFs), which indicate higher tumor aggressiveness, from normal mitotic figures (NMFs) remains challenging due to subtle morphological differences and high intra-class variability. This task is further complicated by domain shifts, including variations in organ, tissue type, and scanner, as well as limited annotations and severe class imbalance. To address these challenges in Track 2 of the MIDOG 2025 Challenge, we propose a solution based on the lightweight ConvNeXt architecture, trained on all available datasets (AMi-Br, AtNorM-Br, AtNorM-MD, and OMG-Octo) to maximize domain coverage. Robustness is enhanced through a histopathology-specific augmentation pipeline, including elastic and stain-specific transformations, and balanced sampling to mitigate class imbalance. A grouped 5-fold cross-validation strategy ensures reliable evaluation. On the preliminary leaderboard, our model achieved a balanced accuracy of 0.8961, ranking among the top entries. These results highlight that broad domain exposure combined with targeted augmentation strategies is key to building accurate and generalizable mitotic figure classifiers.

eess.IV↗

arXiv2025

Efficient Fine-Tuning of DINOv3 Pretrained on Natural Images for Atypical Mitotic Figure Classification

Atypical mitotic figures (AMFs) indicate abnormal cell division associated with poor prognosis. Their detection remains difficult due to low prevalence, subtle morphology, and inter-observer variability. The MItosis DOmain Generalization (MIDOG) 2025 challenge introduces a benchmark for AMF classification across multiple domains. In this work, we fine-tuned the recently published DINOv3-H+ vision transformer, pretrained on natural images, using low-rank adaptation (LoRA), training only 1.3M parameters. We combine this with extensive augmentation and a domain-weighted Focal Loss to better handle the strong domain heterogeneity in the dataset. Despite the large shift between natural images and histopathology, our fine-tuned DINOv3 transfers effectively, reaching first place on the final test set. These results highlight the advantages of DINOv3 pretraining and underline the efficiency and robustness of our fine-tuning strategy, yielding state-of-the-art results for the atypical mitosis classification challenge in MIDOG 2025. Our code is publicly available on GitHub.

eess.IV↗

arXiv2025

MIPHEI-ViT: Multiplex Immunofluorescence Prediction from H&E Images using ViT Foundation Models

Histopathological analysis is a cornerstone of cancer diagnosis, with Hematoxylin and Eosin (H&E) staining routinely acquired for every patient to visualize cell morphology and tissue architecture. On the other hand, multiplex immunofluorescence (mIF) enables more precise cell type identification via proteomic markers, but has yet to achieve widespread clinical adoption due to cost and logistical constraints. To bridge this gap, we introduce MIPHEI (Multiplex Immunofluorescence Prediction from H&E Images), a U-Net-inspired architecture that leverages a ViT pathology foundation model as encoder to predict mIF signals from H&E images using rich pretrained representations. MIPHEI targets a comprehensive panel of markers spanning nuclear content, immune lineages (T cells, B cells, myeloid), epithelium, stroma, vasculature, and proliferation. We train our model using the publicly available OrionCRC dataset of restained H&E and mIF images from colorectal cancer tissue, and validate it on five independent datasets: HEMIT, PathoCell, IMMUcan, Lizard and PanNuke. On OrionCRC test set, MIPHEI achieves accurate cell-type classification from H&E alone, with F1 scores of 0.93 for Pan-CK, 0.83 for alpha-SMA, 0.68 for CD3e, 0.36 for CD20, and 0.28 for CD68, substantially outperforming both a state-of-the-art baseline and a random classifier for most markers. Our results indicate that, for some molecular markers, our model captures the complex relationships between nuclear morphologies in their tissue context, as visible in H&E images and molecular markers defining specific cell types. MIPHEI offers a promising step toward enabling cell-type-aware analysis of large-scale H&E datasets, in view of uncovering relationships between spatial cellular organization and patient outcomes.

cs.CV↗

arXiv2025

UWB Anchor Based Localization of a Planetary Rover

Localization of an autonomous mobile robot during planetary exploration is challenging due to the unknown terrain, the difficult lighting conditions and the lack of any global reference such as satellite navigation systems. We present a novel approach for robot localization based on ultra-wideband (UWB) technology. The robot sets up its own reference coordinate system by distributing UWB anchor nodes in the environment via a rocket-propelled launcher system. This allows the creation of a localization space in which UWB measurements are employed to supplement traditional SLAM-based techniques. The system was developed for our involvement in the ESA-ESRIC challenge 2021 and the AMADEE-24, an analog Mars simulation in Armenia by the Austrian Space Forum (\"OWF).

cs.RO↗

arXiv2024

Interpretable Embeddings for Segmentation-Free Single-Cell Analysis in Multiplex Imaging

Multiplex Imaging (MI) enables the simultaneous visualization of multiple biological markers in separate imaging channels at subcellular resolution, providing valuable insights into cell-type heterogeneity and spatial organization. However, current computational pipelines rely on cell segmentation algorithms, which require laborious fine-tuning and can introduce downstream errors due to inaccurate single-cell representations. We propose a segmentation-free deep learning approach that leverages grouped convolutions to learn interpretable embedded features from each imaging channel, enabling robust cell-type identification without manual feature selection. Validated on an Imaging Mass Cytometry dataset of 1.8 million cells from neuroblastoma patients, our method enables the accurate identification of known cell types, showcasing its scalability and suitability for high-dimensional MI data.

eess.IV↗

arXiv2023

Simple and Efficient Confidence Score for Grading Whole Slide Images

Grading precancerous lesions on whole slide images is a challenging task: the continuous space of morphological phenotypes makes clear-cut decisions between different grades often difficult, leading to low inter- and intra-rater agreements. More and more Artificial Intelligence (AI) algorithms are developed to help pathologists perform and standardize their diagnosis. However, those models can render their prediction without consideration of the ambiguity of the classes and can fail without notice which prevent their wider acceptance in a clinical context. In this paper, we propose a new score to measure the confidence of AI models in grading tasks. Our confidence score is specifically adapted to ordinal output variables, is versatile and does not require extra training or additional inferences nor particular architecture changes. Comparison to other popular techniques such as Monte Carlo Dropout and deep ensembles shows that our method provides state-of-the art results, while being simpler, more versatile and less computationally intensive. The score is also easily interpretable and consistent with real life hesitations of pathologists. We show that the score is capable of accurately identifying mispredicted slides and that accuracy for high confidence decisions is significantly higher than for low-confidence decisions (gap in AUC of 17.1% on the test set). We believe that the proposed confidence score could be leveraged by pathologists directly in their workflow and assist them on difficult tasks such as grading precancerous lesions.

eess.IV↗

arXiv2023

PointFISH -- learning point cloud representations for RNA localization patterns

Subcellular RNA localization is a critical mechanism for the spatial control of gene expression. Its mechanism and precise functional role is not yet very well understood. Single Molecule Fluorescence in Situ Hybridization (smFISH) images allow for the detection of individual RNA molecules with subcellular accuracy. In return, smFISH requires robust methods to quantify and classify RNA spatial distribution. Here, we present PointFISH, a novel computational approach for the recognition of RNA localization patterns. PointFISH is an attention-based network for computing continuous vector representations of RNA point clouds. Trained on simulations only, it can directly process extracted coordinates from experimental smFISH images. The resulting embedding allows scalable and flexible spatial transcriptomics analysis and matches performance of hand-crafted pipelines.

q-bio.QM↗

arXiv2023

Learning with minimal effort: leveraging in silico labeling for cell and nucleus segmentation

Deep learning provides us with powerful methods to perform nucleus or cell segmentation with unprecedented quality. However, these methods usually require large training sets of manually annotated images, which are tedious and expensive to generate. In this paper we propose to use In Silico Labeling (ISL) as a pretraining scheme for segmentation tasks. The strategy is to acquire label-free microscopy images (such as bright-field or phase contrast) along fluorescently labeled images (such as DAPI or CellMask). We then train a model to predict the fluorescently labeled images from the label-free microscopy images. By comparing segmentation performance across several training set sizes, we show that such a scheme can dramatically reduce the number of required annotations.

eess.IV↗

arXiv2022

Giga-SSL: Self-Supervised Learning for Gigapixel Images

Whole slide images (WSI) are microscopy images of stained tissue slides routinely prepared for diagnosis and treatment selection in medical practice. WSI are very large (gigapixel size) and complex (made of up to millions of cells). The current state-of-the-art (SoTA) approach to classify WSI subdivides them into tiles, encodes them by pre-trained networks and applies Multiple Instance Learning (MIL) to train for specific downstream tasks. However, annotated datasets are often small, typically a few hundred to a few thousand WSI, which may cause overfitting and underperforming models. Conversely, the number of unannotated WSI is ever increasing, with datasets of tens of thousands (soon to be millions) of images available. While it has been previously proposed to use these unannotated data to identify suitable tile representations by self-supervised learning (SSL), downstream classification tasks still require full supervision because parts of the MIL architecture is not trained during tile level SSL pre-training. Here, we propose a strategy of slide level SSL to leverage the large number of WSI without annotations to infer powerful slide representations. Applying our method to The Cancer-Genome Atlas, one of the most widely used data resources in cancer research (16 TB image data), we are able to downsize the dataset to 23 MB without any loss in predictive power: we show that a linear classifier trained on top of these embeddings maintains or improves previous SoTA performances on various benchmark WSI classification tasks. Finally, we observe that training a classifier on these representations with tiny datasets (e.g. 50 slides) improved performances over SoTA by an average of +6.3 AUC points over all downstream tasks.

cs.CV↗

arXiv2021

Prototyping and Evaluation of Infrastructure-Assisted Transition of Control for Cooperative Automated Vehicles

Automated driving is now possible in diverse road and traffic conditions. However, there are still situations that automated vehicles cannot handle safely and efficiently. In this case, a Transition of Control (ToC) is necessary so that the driver takes control of the driving. Executing a ToC requires the driver to get full situation awareness of the driving environment. If the driver fails to get back the control in a limited time, a Minimum Risk Maneuver (MRM) is executed to bring the vehicle into a safe state (e.g., decelerating to full stop). The execution of ToCs requires some time and can cause traffic disruption and safety risks that increase if several vehicles execute ToCs/MRMs at similar times and in the same area. This study proposes to use novel C-ITS traffic management measures where the infrastructure exploits V2X communications to assist Connected and Automated Vehicles (CAVs) in the execution of ToCs. The infrastructure can suggest a spatial distribution of ToCs, and inform vehicles of the locations where they could execute a safe stop in case of MRM. This paper reports the first field operational tests that validate the feasibility and quantify the benefits of the proposed infrastructure-assisted ToC and MRM management. The paper also presents the CAV and roadside infrastructure prototypes implemented and used in the trials. The conducted field trials demonstrate that infrastructure-assisted traffic management solutions can reduce safety risks and traffic disruptions.

eess.SY↗

arXiv2020

Neural network fast-classifies biological images using features selected after their random-forests-importance to power smart microscopy

Artificial intelligence is nowadays used for cell detection and classification in optical microscopy, during post-acquisition analysis. The microscopes are now fully automated and next expected to be smart, to make acquisition decisions based on the images. It calls for analysing them on the fly. Biology further imposes training on a reduced dataset due to cost and time to prepare the samples and have the datasets annotated by experts. We propose here a real-time image processing, compliant with these specifications by balancing accurate detection and execution performance. We characterised the images using a generic, high-dimensional feature extractor. We then classified the images using machine learning for the sake of understanding the contribution of each feature in decision and execution time. We found that the non-linear-classifier random forests outperformed Fisher's linear discriminant. More importantly, the most discriminant and time-consuming features could be excluded without any significant loss in accuracy, offering a substantial gain in execution time. It suggests a feature-group redundancy likely related to the biology of the observed cells. We offer a method to select fast and discriminant features. In our assay, a 79.6 $\pm$ 2.4 % accurate classification of a cell took 68.7 $\pm$ 3.5 ms (mean $\pm$ SD, 5-fold cross-validation nested in 10 bootstrap repeats), corresponding to 14 cells per second, dispatched into 8 phases of the cell cycle using 12 feature-groups and operating a consumer market ARM-based embedded system. Interestingly, a simple neural network offered similar performances paving the way to faster training and classification, using parallel execution on a general-purpose graphic processing unit. Finally, this strategy is also usable for deep neural networks paving the way to optimising these algorithms for smart microscopy.

q-bio.QM↗

arXiv2014

Assessment of algorithms for mitosis detection in breast cancer histopathology images

The proliferative activity of breast tumors, which is routinely estimated by counting of mitotic figures in hematoxylin and eosin stained histology sections, is considered to be one of the most important prognostic markers. However, mitosis counting is laborious, subjective and may suffer from low inter-observer agreement. With the wider acceptance of whole slide images in pathology labs, automatic image analysis has been proposed as a potential solution for these issues. In this paper, the results from the Assessment of Mitosis Detection Algorithms 2013 (AMIDA13) challenge are described. The challenge was based on a data set consisting of 12 training and 11 testing subjects, with more than one thousand annotated mitotic figures by multiple observers. Short descriptions and results from the evaluation of eleven methods are presented. The top performing method has an error rate that is comparable to the inter-observer agreement among pathologists.

cs.CV↗