A5000341866- AI in cancer detection
- Radiomics and Machine Learning in Medical Imaging
- Cell Image Analysis Techniques
- Digital Imaging for Blood Diseases
- Generative Adversarial Networks and Image Synthesis
- Colorectal Cancer Screening and Detection
- Cancer Genomics and Diagnostics
- Cervical Cancer and HPV Research
- Video Analysis and Summarization
- Image Retrieval and Classification Techniques
- Single-cell and spatial transcriptomics
- Biomedical Text Mining and Ontologies
- Explainable Artificial Intelligence (XAI)
- Natural Language Processing Techniques
- Genetics, Bioinformatics, and Biomedical Research
- Computational Drug Discovery Methods
- Gene expression and cancer classification
- Machine Learning and Data Classification
- Molecular Biology Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Topic Modeling
- Artificial Intelligence in Healthcare and Education
- Advanced Image Processing Techniques
- Machine Learning and Algorithms
- Cancer-related molecular mechanisms research
Helmholtz Zentrum München
2022-2024
Technical University of Munich
2022-2024
Fresenius (Germany)
2023-2024
Institute of Bioinformatics and Systems Biology
2021
Deep learning (DL) can accelerate the prediction of prognostic biomarkers from routine pathology slides in colorectal cancer (CRC). However, current approaches rely on convolutional neural networks (CNNs) and have mostly been validated small patient cohorts. Here, we develop a new transformer-based pipeline for end-to-end biomarker by combining pre-trained transformer encoder with network patch aggregation. Our approach substantially improves performance, generalizability, data efficiency,...
Foundation models are reshaping computational pathology by enabling transfer learning, where pre-trained on vast datasets can be adapted for downstream diagnostic, prognostic, and therapeutic response tasks. Despite these advances, foundation still limited in their ability to encode the entire gigapixel whole-slide images without additional training often lack complementary multimodal data. Here, we introduce Threads, a slide-level model capable of generating universal representations any...
To handle the large scale of whole slide images in computational pathology, most approaches first tessellate into smaller patches, extract features from these and finally aggregate feature vectors with weakly-supervised learning. The performance this workflow strongly depends on quality extracted features. Recently, foundation models computer vision showed that leveraging huge amounts data through supervised or self-supervised learning improves generalizability for a variety tasks. In study,...
Abstract Histopathology is considered the reference standard for diagnosing presence and nature of many malignancies, including cancer. However, analyzing tissue samples writing pathology reports time-consuming, labor-intensive, non-standardized. To address this problem, we present HistoGPT, first vision language model that simultaneously generates from multiple images. It was trained on more than 15,000 whole slide images over 6,000 dermatology patients with corresponding reports. The...
Spatial transcriptomics (ST) enables interrogating the molecular composition of tissue with ever-increasing resolution, depth, and sensitivity. However, costs, rapidly evolving technology, lack standards have constrained computational methods in ST to narrow tasks small cohorts. In addition, underlying morphology as reflected by H&E-stained whole slide images (WSIs) encodes rich information often overlooked studies. Here, we introduce HEST-1k, a collection 1,108 spatial transcriptomic...
Machine learning models are typically evaluated by computing similarity with reference annotations and trained maximizing such. Especially in the biomedical domain, subjective suffer from low inter-and intra-rater reliability. Since only reflect one interpretation of real world, this can lead to sub-optimal predictions even though model achieves high scores. Here, theoretical concept Peak Ground Truth (PGT) is introduced. PGT marks point beyond which an increase annotation stops translating...
Latent diffusion models (LDMs) have emerged as a state-of-the-art image generation method, outperforming previous Generative Adversarial Networks (GANs) in terms of training stability and quality. In computational pathology, generative are valuable for data sharing augmentation. However, the impact LDM-generated images on histopathology tasks compared to traditional GANs has not been systematically studied. We trained three LDMs styleGAN2 model histology tiles from nine colorectal cancer...
The field of computational pathology has been transformed with recent advances in foundation models that encode histopathology region-of-interests (ROIs) into versatile and transferable feature representations via self-supervised learning (SSL). However, translating these advancements to address complex clinical challenges at the patient slide level remains constrained by limited data disease-specific cohorts, especially for rare conditions. We propose TITAN, a multimodal whole model...
Annotating microscopy images for nuclei segmentation by medical experts is laborious and time-consuming. To leverage the few existing annotations, also across multiple modalities, we propose a novel microscopy-style augmentation technique based on generative adversarial network (GAN). Unlike other style transfer methods, it can not only deal with different cell assay types lighting conditions, but imaging such as bright-field fluorescence microscopy. Using disentangled representations...
Background: Deep learning (DL) can extract predictive and prognostic biomarkers from routine pathology slides in colorectal cancer. For example, a DL test for the diagnosis of microsatellite instability (MSI) CRC has been approved 2022. Current approaches rely on convolutional neural networks (CNNs). Transformer are outperforming CNNs replacing them many applications, but have not used biomarker prediction cancer at large scale. In addition, most trained small patient cohorts, which limits...
Abstract Recent progress in computational pathology has been driven by deep learning. While code and data availability are essential to reproduce findings from preceding publications, ensuring a learning model’s reusability is more challenging. For that, the codebase should be well-documented easy integrate existing workflows, models robust towards noise generalizable different sources. Strikingly, only few algorithms have reused other researchers so far, let alone employed clinical setting....
Vision Transformers (ViTs) and Swin (Swin) are currently state-of-the-art in computational pathology. However, domain experts still reluctant to use these models due their lack of interpretability. This is not surprising, as critical decisions need be transparent understandable. The most common approach understanding transformers visualize attention. attention maps ViTs often fragmented, leading unsatisfactory explanations. Here, we introduce a novel architecture called the B-cos Transformer...
Große Sprachmodelle (Large Language Models, LLMs) sind Tools der künstlichen Intelligenz (KI), die speziell für Verarbeitung und Erzeugung von Text trainiert sind. LLMs erregten erhebliche öffentliche Aufmerksamkeit, nachdem ChatGPT OpenAI im November 2022 öffentlich zugänglich gemacht wurde. können Fragen beantworten, Texte zusammenfassen, paraphrasieren übersetzen, zwar auf einer Ebene, menschlichen Fähigkeiten kaum zu unterscheiden ist. Die Möglichkeit, aktiv mit Modellen wie...
In hematology, computational models offer significant potential to improve diagnostic accuracy, streamline workflows, and reduce the tedious work of analyzing single cells in peripheral blood or bone marrow smears. However, clinical adoption has been hampered by lack generalization due large batch effects, small dataset sizes, poor performance transfer learning from natural images. To address these challenges, we introduce DinoBloom, first foundation model for cell images utilizing a...
Hematoxylin- and eosin (H&E) stained whole-slide images (WSIs) are the foundation of diagnosis cancer. In recent years, development deep learning-based methods in computational pathology enabled prediction biomarkers directly from WSIs. However, accurately linking tissue phenotype to at scale remains a crucial challenge for democratizing complex precision oncology. This protocol describes practical workflow solid tumor associative modeling (STAMP), enabling WSIs using learning. The STAMP is...