Projects

Image: Scidraw | info

TIMELY: Time-series Integration across Modalities for Evaluation of Latent DYnamics

TIMELY provides the first comprehensive benchmark for multimodal biological time-series data, addressing the lack of standardized, high-quality datasets for modeling complex dynamical systems. It fosters the development of statistical and foundation models tailored to the analytical needs of research in biomedicine and neuroscience.

BSIC 2023 contribution by Sebastian Dupraz (AG Bradke), DZNE; title: Stranger in the mirror

Image: Sebastian Dupraz (AG Bradke), DZNE | info

Spatio-temporal inverse approaches for EEG/MEG reconstruction of neural networks in the human brain

This project aims to develop novel methods for reconstructing brain activity from dynamic EEG and MEG measurements. By using realistic, individualized finite element models and advanced regularization techniques, including machine learning, we seek to solve this inverse problem in real patient settings, ultimately improving the diagnosis and treatment of medication-resistant focal epilepsy.

Visual for UQOB; Creating a benchmark dataset for object-detection and Uncertainty Quantification (UQ) in a multi-rater setting, to address annotation variability and AI model evaluation.

info

UQOB – Uncertainty Quantification in Object-detection Benchmark

Creating a benchmark dataset for object-detection and Uncertainty Quantification (UQ) in a multi-rater setting, to address annotation variability and AI model evaluation.

Visusal for SCHEMA;Metastases represent a significant exacerbation of tumor severity. If one could predict the likelihood of tumors metastasizing, this could inform treatment decisions to avoid or delay this outcome. SCHEMA develops a benchmark dataset of primary tumor samples and metadata on whether the tumor has metastasized at different time points after sampling. With this dataset, a challenge for machine learning scientists will be defined to build prognostic models for likelihood of tumors metastasizing, promoting innovation in prognostic modeling for a clinically relevant task.

Image: Hellmut Augustin, DKFZ (BSIC 2021 contribution) | info

SCHEMA – profiling Spatial Cancer HEterogeneity across modalities to benchmark Metastasis risk prediction

SCHEMA creates a benchmark dataset linking tumor samples with metastasis outcomes to enable machine-learning models that predict metastasis risk and support clinical decision-making.

info

RenewBench – A Global Benchmark for Renewable Energy Generation

Renewable energy’s variability makes grid management complex. RenewBench aims to provide standardized, high-quality data to advance trustworthy AI models and accelerate the transition to sustainable energy.

Visual for Pero; Addressing the lack of standardized, FAIR benchmark datasets in perovskite photovoltaics. Pero enables reproducible AI models for efficiency prediction, material classification, and defect detection, which are critical for industrial scaling of sustainable energy technologies.

Image: Photo: Markus Breig, KIT; illustration: Felix Laufer, KIT | info

Pero – Unlocking ML Potential: Benchmark Datasets on Perovskite Thin Film Processing

Addressing the lack of standardized, FAIR benchmark datasets in perovskite photovoltaics. Pero enables reproducible AI models for efficiency prediction, material classification, and defect detection, which are critical for industrial scaling of sustainable energy technologies.

Visual for NeuroHarmonize; The benchmark addresses the lack of harmonized, reproducible, and privacy-preserving multimodal datasets for Alzheimer’s disease (AD). Current AI models struggle with fragmented and non-standardized data, which limits their generalizability and clinical deployment. NeuroHarmonize creates a FAIR-compliant, decentralized benchmarking framework to accelerate reliable, transparent, and collaborative AI for AD diagnosis, prognosis, and long-term monitoring.

Image: Georg Kislinger, Martina Schifferer, Christian Haass & Maryam Khojasteh-Farat, DZNE (BSIC 2021 contribution) | info

NeuroHarmonize – A Benchmark Decentralized Data Harmonization Workflow for AI-Driven Alzheimer’s Disease Management

The benchmark addresses the lack of harmonized, reproducible, and privacy-preserving multimodal datasets for Alzheimer’s disease (AD). NeuroHarmonize creates a FAIR-compliant, decentralized benchmarking framework to accelerate reliable, transparent, and collaborative AI for AD diagnosis, prognosis, and long-term monitoring.

Visual for GRIDMARK; Transforming energy systems toward climate neutrality: Distribution grids have the potential to be catalysts for the energy transition. Unfortunately, most Distribution System Operators lack the resources to fully monitor their systems. Therefore, there is an urgent need for more high-quality data, particularly to develop and test machine learning models.

info

GRIDMARK – Generating Reproducible Insights through Data Benchmarking for AI in Energy Systems

Transforming energy systems toward climate neutrality: Distribution grids have the potential to be catalysts for the energy transition. Unfortunately, most Distribution System Operators lack the resources to fully monitor their systems. Therefore, there is an urgent need for more high-quality data, particularly to develop and test machine learning models.

Visual for ForestUNLOCK; Building the first consistent multi-modal single tree benchmark for forest structure and carbon stock assessments of the northern boreal forest

Image: Open white spruce forest with glacier in background in the Chugach Mountains, Alaska, US ©Stefan Kruse, AWI | info

ForestUNLOCK: A multi-modal Multiscale Benchmark Dataset for AI-Driven Boreal Forest Monitoring and Carbon Accounting

Building the first consistent multi-modal single tree benchmark for forest structure and carbon stock assessments of the northern boreal forest

Image: Ingmar Nitze, AWI (BSIC 2021 contribution) | info

BASE: Benchmarking Agro-environmental database for Sustainable agriculture intensification

Building a BASE dataset enables robust predictions of yield potential, resource efficiency, and sustainability thresholds, driving climate resilience and sustainable agricultural intensification

Image: FZJ

AMOEBE: lArge-scale Multi-mOdal Microbial livE-cell imaging BEnchmark

Building a large-scale, FAIR benchmark for AI-driven analysis of microbial communities using time-lapse microscopy to advance understanding of microbial dynamics, ecosystem stability, and their role in health and biotechnology.

Manual microscopic biodiversity monitoring is time consuming and expert requiring thus limits the potential for biodiversity monitoring and hence to recognize risks climate and environmental changes on biodiversity related to crucial ecosystem functions.

Image: AIMBIS | info

AIMBIS – Artificial Intelligence for Microscopic Biodiversity Screening

Manual microscopic biodiversity monitoring is time-consuming and requires expert knowledge, limiting the potential for biodiversity monitoring, hence to recognize the impacts of climate and environmental change on crucial ecosystem functions.

Visual for ADD-ON; ADD-ON addresses the lack of reliable data for predicting how microbial enzymes assemble peptide-based natural products. By enabling accurate AI-driven structure prediction, it accelerates the discovery of new bioactive compounds and ultimately supports efforts to combat antimicrobial resistance.

Image: ADD-ON | info

ADD-ON: Adenylation Domain Database and Online Benchmarking Platform

ADD-ON addresses the lack of reliable data for predicting how microbial enzymes assemble peptide-based natural products. By enabling accurate AI-driven structure prediction, it accelerates the discovery of new bioactive compounds and ultimately supports efforts to combat antimicrobial resistance.

Image: Stefan Kruse | info

POINTR

Mapping Boreal Forest Change Using 3D Radar and Point Cloud Data

Global warming is reshaping northern boreal forests. This project maps forest structure and ecosystem services using high-resolution radar satellite monitoring combined with 3D imaging data.

This image shows simulation results comparing MWF mapping with single inversions of "T2-decay" data, "T2*-decay" data, and with joint inversion of both. The joint inversion is significantly closer to ground truth, as the second row shows.

info

JIMM2

3D Myelin Mapping with AI and Uncertainty Quantification

Changes in brain myelin are linked to many neurological diseases. This project aims to improve myelin water imaging, enabling more accurate and accessible diagnostics.

info

BestMeta

Behavioral Standard Metadata

Developing metadata standards and FAIR analysis pipelines for Video Tracking Assays (VTAs) in toxicology and medical sciences

Image: FOMIA | info

FOMIA

A Foundation Model for Microscopy Image Analysis

This project will develop a foundation model trained on a large and diverse dataset of microscopy images to facilitate the adaptation of artificial intelligence methods to biological image analysis.

Image: PlastoView | info

PlastoView

Microplastic Detection with the PlastiScope

Water quality is essential for ecosystems and human health, yet it's increasingly threatened by microplastics. This project develops image-based methods for detecting both plankton and microplastics using a new low-cost, mobile system.

info

cryoFocal

3D Reconstruction from defocused cryo-EM images

This project explores how defocused images recorded with an electron microscope can be used to reconstruct the 3D structure of molecules inside cells. This method aims to enable faster and more cost-effective structural analysis of molecules to accelerate understanding of their functions and to design drugs against them.

visual for third-party funded project UMDISTO

info

UMDISTO: Unsupervised Model Discovery

The project aims to develop novel methods for unsupervised multi-matching to map cellular-level correspondences in organisms like C. elegans.

Image: ChatGPT. Prompt: MRI image of the knee with a touch of AI

The Human Radiome Project (THRP)

The Human Radiome Project (THRP) aims to drive a paradigm shift in medical research, providing novel insights into human health and disease through the power of AI. By integrating diverse radiological data, it seeks to enable groundbreaking advancements in personalized medicine, enhancing diagnostic accuracy and improving patient care.

Microcosmos of the Ocean by Klas Ove Möller, Hereon

Image: Klas Ove Möller, Hereon | info

AqQua

AqQua aims to build the first foundational pelagic imaging model using billions of aquatic images worldwide. These images, spanning species from plankton, will help an AI classify species, extract traits, and estimate carbon content, offering key insights into biodiversity, ecosystem health, and the biological carbon pump's role in climate regulation.

Image: NicoElNino on Shutterstock | info

Synergy Unit

The Synergy Unit amplifies the Helmholtz Foundation Model Initiative's impact by developing AI principles for diverse fields. Collaborating with HFMI projects, it focuses on knowledge sharing, community building, and representation to ensure the initiative's lasting influence.

decorative visual; blue-futuristic-stream-data-communication-flying

Image: Envato Elements | info

FONDA: Dependability, Adaptability and Uncertainty Quantification for Data Analysis Workflows in Large-Scale Biomedical Image Analysis

The project aims to enhance infrastructures for machine learning (ML)-intensive DAWs in advanced biomedical imaging applications.

Image: Tim Roith | info

COMFORT

COMFORT aims to achieve breakthroughs in developing compact, flexible, and robust machine learning models for image, audio, and network data. In doing so, its application-oriented research program will advance the mathematical understanding of machine learning at the intersection of effectiveness and robustness.

GLAM, third-party funded project, Helmholtz Imaging

info

GLAM: Generative lung architecture modeling

This project is developing generative methods for designing bio-printable lung tissues across a spectrum of disease severity in the specific context of mouse and human lung disease.

info

Fast-EMI

Deep-learning assisted fast in situ 4D electron microscope imaging

A novel imaging approach combining electron microscopy and deep learning has been established. This method enables adaptive tracking of atomic defects, accelerating material development for the energy transition.

info

BrainShapes

Laplace-Beltrami shape descriptors of brain structures: Comparative optimization and genetic dissection

The project explores the 3D structure of the human brain by creating a digital 'map' of the brain and examining its unique genetic properties, potentially linking genetic variations to brain disorders.

info

CLARITY

CineMR-guided ML-driven Breathing Models for Adaptive Radiotherapy

Dose-escalated radiotherapy of lung cancers requires precise monitoring of lesions and nearby organs at risk. Current methods are able to track ultra-central lesions but neglect their deforming vicinity, risking unacceptable toxicity to aortico-pulmonary structures. AI-based anomaly detection and generative AI models can address both requirements in real-time.

info

3DforestSIF

Understanding the solar-induced fluorescence (SIF) signal of natural, complex tree canopies

3DforestSIF seeks to correct airborne solar-induced fluorescence (SIF) data from forests for canopy structural and illumination effects, providing valuable insights for the early detection of forest stress.

Image: X-BRAIN | info

X-BRAIN

Cross-modality representation learning for brain analysis and data integration

This project aims to develop AI methods that support the integration of multimodal imaging data into human brain atlases, thereby advancing the analysis of brain structure in both health and disease.

Image: Aaron Christian Banze | info

HYPER-AMPLIFAI

Advancing Visual Foundation Models for Multi-/Hyperspectral Image Analysis in Agriculture/Forestry

The project aims to make advanced AI models accessible for Hyperspectral Earth Observation, reducing computational demands, and improving environmental assessments through user-friendly interfaces.

Decorative image; Third-party funded project MDLMA; this project aims to optimize biodegradable Mg-based implants by analyzing vast multimodal biomedical image data

info

Multi-task Deep Learning for Large-scale Multimodal Biomedical Image Analysis (MDLMA)

The MDLMA project aims to optimize biodegradable Mg-based implants by analyzing vast multimodal biomedical image data. Through deep learning, it seeks to improve implant properties by understanding the interplay of microstructure, mechanics, biology, and degradation. The project will develop multi-task deep learning methods for enhanced image analysis, offering a unified framework for broader application in biomedical imaging.

info

Deep Learning based Regularization for Inverse Problems

This project aims to investigate the construction of regularization methods for ill-posed inverse problems based on deep learning and their theoretical foundations. Specific objectives include the development of robust and interpretable results, requiring the initial development of new concepts of robustness and interpretability in this context.

info

Foundations of Supervised Deep Learning for Inverse Problems

Recently, deep learning methods have excelled at various data processing tasks including the solution of ill-posed inverse problems. The goal of this project is to contribute to the theoretical foundation for truly understanding deep networks as regularization techniques which can reestablish a continuous dependence of the solution on the data.

info

SFB Transregio 154 – C06: Transport metrics for analysis and optimization of network problems

SFB TRR 154 is a project of the German Research Foundation (DFG) and combines integer-continuous methods, model adaptation, and numerical simulation, to analyze and optimize gas markets, infrastructure, and control of networks. The third funding period specifically focuses on the transition from natural gas to hydrogen.

info

QGRIS: Quantitative Gamma-Ray Imaging System

Compton cameras are used for the radiological characterization of nuclear power plants. In this project, a suitable camera system is designed, and the associated algorithms for image reconstruction and nuclide characterization are implemented as user software.

info

Bayesian Computations for Large-scale (Nonlinear) Inverse Problems in Imaging

During research stays with the collaborating group at Caltech, we aim to investigate various aspects of statistical inverse problems. This includes inquiries into particle- and PDE-based sampling methods, as well as robust regularization using neural networks.

Image: Jonas Baumann, HIPS | info

ImageTox

Automated image-based Detection of Early Toxicity Events in Zebrafish Larvae

ImageTox wants to establish an automated image-based system to assess zebrafish larval development. This will allow for a fast and unbiased evaluation of pathophysiological events during toxicological studies. To achieve this, the imaging process has to be optimized and a reliable model for sequence recognition based on deep learning has to be developed.

info

HighLine

High Image Quality for Lines in MRI: From Roots to Angiograms

MR images of roots and vessels are very similar: both display thin, line-like objects. The aim of the project is to increase image quality of both kind of MR data by exploiting their similarity. HighLine aims at obtaining high quality images in reduced scan time to lower patient burden and increase patient and plant throughput by adapting state-of-the-art 3D image enhancement methods, and developing new deep-learning based methods.

Image: Johannes Seiffarth, FZ Jülich | info

EMSIG

Event-driven Microscopy for Smart Microfluidic Single-cell Analysis

Microfluidic live-cell imaging (MLCI) unlocks spatio-temporal insights into population heterogeneity emerging from a single cell. EMSIG brings smart live-event detection capabilities to MLCI to facilitate the adaptive optimization of biological event resolution and autonomously counteracting deteriorating image qualities.

Image: DIPLO | info

DIPLO

Paving the way from in situ plankton image data to a Digital Twin Ocean

This project will develop a user-friendly software platform to analyze plankton images independent of the instrument with which images were collected. This will help to compare data and create a common database, which is a critical step towards an image-based ecosystem component of a “Digital Twin Ocean”.

Image: Hailong He, Helmholtz Munich | info

Deep4OM

Deep learning powered optoacoustic mesoscopy for non-invasive diagnostics of skin diseases

Deep4OM aims to develop a deep learning-based framework for optoacoustic mesoscopy image analysis, enabling quantification of human skin biomarkers for non-invasive skin disease diagnosis. Deep4OM has the potential to change the landscape of non-invasive skin imaging, and could significantly promote the diagnostic and prognostic applications of RSOM in clinical routine.

info

BENIGN

Biocompatible and Efficient Nanocrystals for Shortwave Infrared Imaging

The BENIGN project aims to enable non-invasive molecular imaging with cellular resolution in vivo at depths of several millimeters. This will be achieved using light from the shortwave infrared (SWIR) range (1000-2000 nm), which has less scattering and autofluorescence compared to the visible and near-infrared spectral range. Bright and targeted imaging agents are needed to fully exploit this range. The project will develop a new approach using lanthanide-based core-shell structures that emit light in the 1500-2000 nm range.

info

AutoCoast

Automatic detection of coastline change and causal linkage with natural and human drivers

Coastal erosion enhanced by climate change has become an increasing global threat, which requires rapid detection and reliable risk assessment. AutoCoast aims to provide advanced and reliable remote sensing-based AI tools to quantify coastline change rate at high-resolution and unravel the linkage between coastline change rate and natural and anthropogenic drivers at regional to global scale.

Image: Xun Xu, Hereon | info

AIOrganoid

Artificial Intelligence Assisted-Imaging for Creating High-yield, High-fidelity Human Lung Organoid

AIOrganoid will apply cutting-edge imaging techniques and develop novel AI-based solutions to facilitate human lung organoid formation with high yield and fidelity, bridging the gap between cell biology and computational imaging.

Image: WeMonitor | info

WeMonitor

Satellite-based Earth observation to detect natural hazards

Satellite imagery makes it possible to detect spatio-temporal anomalies on the Earth's surface, including natural hazards such as landslides, deforestation, or the emergence of large waste dump sites. This project aims to use artificial intelligence to detect these changes at an early stage and to be able to monitor their progress.

Image: Ehsan Faridi, IEK-13, Forschungszentrum Juelich GmbH

UTILE

Autonomous image analysis to accelerate energy materials discovery and integration

Research into green materials for clean energy generation is moving at full speed – yet still requires a long time to complete. This project is working on an open source image processing application that uses artificial intelligence to drive the analysis and management of image data from experiments across the energy materials community.

Image: DLR

TerraByte-DNN2Sim

On the trail of the mystery of the laws of calving

Researchers still face a mystery when it comes to the laws by which glaciers calve. This project aims to use satellite imagery, artificial intelligence, mathematical optimisation and a new data processing pipeline to track the movements of glacier fronts in Antarctica to get closer to solving the mystery.

Image: Paul Kamm, Helmholtz-Zentrum Berlin

Avanti

X-ray tomoscopy of dynamic manufacturing processes

How can the manufacturing processes of materials be mapped at the smallest level? How do you train an artificial intelligence to analyze these processes automatically? That's the focus of the Avanti project, which aims to improve X-ray tomoscopy – the imaging and quantification of three-dimensional images of very fast-moving processes.

Image: SmartPhase project partners

SmartPhase

Fast, intelligent 3-D X-ray images for material examination

In order to be able to improve materials, it helps to take a look at their microstructures. This is because valuable information about their properties and behaviour can be found there – for example information about when which ageing processes begin. The aim of this project is to automate and accelerate access to this information with the help of a smart imaging technique.

SyNaToSe

Leveraging Cross-Domain Synergies for Efficient Machine Learning of Nanoscale Tomogram Segmentation

The aim is to develop an adaptable algorithm that can be used to perform different tasks in data and image analysis without needing to be trained with new, laboriously annotated images for each separate task.

UCS

Ultra Content Screening for Clinical Diagnostics and Deep Phenotyping

A method will be developed in which selected biomarkers in tumour and bone marrow cells from cancer patients will be examined and analysed automatically. The novel technology is based on ultra content screening technology, which allows detailed insights at the single cell level.

NImRLS

Neuroimaging Biomarkers for Restless Leg Syndrome

The aim is to develop a software solution that can analyse enormous amounts of data on tens of thousands of subjects from large-scale health studies. Using restless leg syndrome as an example, genomic data will be combined with neuroimaging data in order to identify new biomarkers with the help of machine learning methods.

Hyper 3D-AI

Artificial Intelligence for 3D multimodal point cloud classification

The aim is to develop an artificial intelligence that can achieve the fusion of two-dimensional data with three-dimensional information. Based on this, the software would simultaneously be able to recognise image characteristics as well as the spatial relationships between different objects.

SATOMI

Tackling the segmentation and tracking challenges of growing colonies and microbialdiversity

An artificial intelligence will observe the growth of bacteria: from microscope images of bacterial cultures taken at regular intervals, it will precisely track the development and division of individual cells – even when multiple bacterial species are cultivated together.

HIT Permafrost

The Hidden Image of Thawing Permafrost

The project aims to develop a method for determining just how extensively thaw processes have already progressed in permafrost regions. The machine learning approach to be developed will be used to analyse radar images from aircraft in order to learn more about the properties of the subsurface permafrost.

MultiSaT4SLOWS

Multi-Satellite imaging for Space-based Landslide Occurrence and Warning Service

In order to detect impending landslides before they occur and to enable reliable emergency mapping after a landslide, the researchers are combining optical data with radar data from satellites. Using machine learning methods, computers will be trained to recognise the tiniest of changes in things like sloping landscape surfaces.

Decorative image with blue, green, pink and yellow colors

info

BRLEMM

Breaking resolution limit of electron microscopy for magnetic materials

A new method will make it possible to take images of the magnetic properties of materials under the electron microscope and to correlate these properties with their atomic structure. In order to achieve high resolution, a special algorithm must be developed to compute the magnetic properties from the microscope data.

Image: DZNE

JIMM

Geophysical Joint Inversion for Accurate Brain Myelin Mapping

The aim of this project is to develop a method for clinically diagnosing neurodegenerative diseases. The content of myelin in the brain – a substance that becomes degraded in diseases – will be quantified using methods from geophysics in order to facilitate early detection and treatment.

SIM

Solar Image-based Modelling

The aim of the project is to develop an algorithm by which computers can automatically predict the space weather. This will make use of datasets of solar images that have been captured from space. The method could replace computationally demanding physics-based models and deliver space weather forecasts long before the effects of solar events are […]

AsoftXm

Advanced Soft-X-Ray Microscopy Solutions

The project aims to develop a method that will speed up the analysis of diffraction patterns that arise in UV and soft X-ray light microscopy, so that the structure of the studied sample can be calculated more efficiently. The method could make the three-dimensional study of nanomaterials considerably easier. There are times when researchers need […]

Project Gallery