Core expertise

We understand how to derive depth, camera motion, and anatomical geometry from monocular and stereo endoscopic video. This allows you to transform endoscopic images into spatial information for measurement, mapping, and guidance.

We combine feature tracking, depth information, and registration methods to follow moving tissue and align preoperative models with live video. This helps you connect surgical plans and segmented anatomy with the intraoperative view.

We have proven models for training surgical phase classification algorithms and anatomy segmentation models based on our customers specific requirements.

We know how to optimize vision pipelines for live video and integrate them with cameras, computing platforms, and surgical systems, including support for Nvidia Jetson/IGX and Holoscan-based platforms. This enables you to move from individual algorithms to a responsive endoscopic application.

We developed algorithms that can robustly deal with differences in image contrast, resolution, field of view, and acquisition geometry. This allows us to align data any modality, including CT, MRI, Ultrasound, X-ray, PET, and endoscopic imaging.

We develop dedicated image filtering, segmentation, detection, and feature extraction workflows with state-of-the-art AI models. Often, these models exploit aspects of ultrasound physics and the inherent modality-specific artifacts.

We have been reconstructing 3D ultrasound since 2014 both with and without dedicated tracking equipment. We won the TUS-REC 2025 challenge on trackerless 3D ultrasound and have adopted this technology multiple times for medical devices. Our powerful live compounding algorithms allow on-the-fly reconstruction.

We shape the state-of-art in multi-modal ultrasound registration. Dedicated conventional and AI-based similarity metrics allow for blazingly fast global registrations, even when custom deformation models are involved. We can stitch and decompress, compensate for motion, and even recover probe trajectories from pair-wise acquisitions.

We cover the complete reconstruction chain, from projection preprocessing (flat-field correction, log conversion, dead-pixel inpainting) to GPU-accelerated analytical and iterative reconstruction (MLEM, SART, SQS, CG, TV-regularized). Our experience with curved detectors, half-scan, biplanar, and C-arm geometries allows us to adapt the pipeline to your imaging system and application.

We combine GPU-based forward projection with multi-resolution intensity registration to align X-ray or fluoroscopy images with CT volumes and anatomical meshes. Different initialization methods and dedicated support for EOS biplanar systems help us address a wide range of clinical setups.

We understand how detector geometry, system configuration, and acquisition conditions affect image quality and can flexibly calibrate for diverse detector types and vendors. Our methods cover C-arm distortion correction, geometric self-calibration, and the reduction of metal, beam-hardening, and ring artifacts.

We use GPU-accelerated forward projection, spectral X-ray modelling, and realistic noise simulation to generate representative synthetic images. This provides suitable data for applications such as model training, fiducial detection, and systematic data augmentation.

We combine deep knowledge of medical imaging data with experience across segmentation, classification, detection, tracking, registration, and quantitative analysis. This allows us to to select and develop approaches suited to your specific clinical and technical challenge and to automate many steps of your workflow.

Over the years, we have collected and curated numerous medical imaging datasets across a wide range of applications. We support dataset design, annotation workflows, and annotation bootstrapping, including efficient labeling with ImFusion Labels or other tools. We can also customize and fine-tune our off-the-shelf models for your data, helping you achieve strong performance with less development effort.

We understand how model architecture, training data, validation methods, and computing constraints affect performance. Our extensive experience with AI frameworks, inference engines, and deployment environments enable us to integrate models in the most efficient way, whether it runs in the cloud or on surgical devices.

We understand the traceability, reproducibility, validation, and documentation requirements for AI in regulated medical products. With that, we anticipate certification needs early and avoid costly changes later in development.

We developed algorithms that can robustly deal with differences in image contrast, resolution, field of view, and acquisition geometry. This allows us to align data any modality, including CT, MRI, Ultrasound, X-ray, PET, and endoscopic imaging.

We model registration with the combination of rigid, affine, non-linear and dense deformation that better reflects how anatomy and imaging conditions change between images. This enables us to account for patient positioning, tissue deformation, anatomical movement and any change cause by the imaging setup.

We combine image volumes, projection images, surfaces, landmarks, and point clouds across 2D/2D, 3D/3D, and 2D/3D workflows. This supports applications ranging from longitudinal analysis to intraoperative guidance.

We bring experience in GPU acceleration, automatic initialization, and learning-based registration. This helps you achieve reliable alignment within the performance and usability requirements of your clinical application.

We understand the anatomical variation and imaging characteristics that make automated analysis challenging and are aware of the typical challenges that might otherwise stop a naive approach in its tracks. This allows us to quickly develop robust approaches relevant to your application.

We combine label maps, surface meshes, landmarks, and relationships between multiple structures in coherent anatomical models. This gives you consistent, reusable data for downstream tasks such as measurements, planning, visualization, and navigation. Our architecture is built in a modular way that allows you to create complex pipelines by chaining together different independent steps.

We have experience adapting anatomical analysis to different modalities and clinical fields, including CT and MRI applications for abdominal, cardiac, orthopedic, spinal, and cranial anatomy. In many cases, we have anatomy-specific libraries that can be integrated into a product. This enables us to shape the analysis around your data, workflow, and product requirements.

We manage streams with different formats, frame rates, timestamps, and latencies. This keeps images, tracking information, and sensor measurements correctly aligned throughout the workflow.

We apply image processing, AI models, reconstruction, and registration directly to incoming data. GPU acceleration supports responsive visualization and analysis during time-critical procedures.

We record one or several streams together, including mixed image and tracking sources, and replay them with original timing so the same pipelines run live or offline. The same stack connects local cameras and medical devices and exchanges streams with external systems over OpenIGTLink, network protocols, and ROS.

We connect robotic systems with cameras, tracking devices, imaging equipment, and force or tactile sensors. This provides the robot with the spatial information required to interact with the patient, instruments, and its environment.

We determine the geometric relationships between robots, sensors, images, tools, and anatomical structures. This allows planned actions in medical images to be transferred accurately to the physical setup.

We work with robot kinematics, trajectory planning, collision checking, and constrained movement. This helps you develop safe and predictable robot behaviour for the available workspace and clinical task.

We combine imaging, perception, robot control, and information about the current procedural step in one software architecture. This allows robotic actions to respond to the clinical workflow rather than operating as isolated movements.

We integrate optical, electromagnetic, robotic devices, and other sensors providing real-time positional data with medical instruments, imaging devices, and software applications. This allows you to combine live position data with the anatomical and procedural information needed during an intervention.

We determine the spatial relationships between trackers, instruments, imaging devices, and the patient. Landmark-, surface-, and image-based registration methods connect preoperative plans with the physical anatomy in the operating room.

We develop 2D and 3D visualizations that show instruments, anatomy, targets, trajectories, and safety-relevant structures together. Guidance views such as tunnel, bullseye, and tool-oriented displays can be adapted to the procedure and user needs.

We combine preoperative plans, intraoperative images, and live tracking data in a shared 3D scene. This gives you a consistent spatial basis for navigation throughout the procedure.