Dibvision VI is a comprehensive software platform for medical imaging analysis, developed by the Dutch company Dibvision Systems. It represents the sixth major release in the Dibvision series, integrating advanced artificial‑intelligence algorithms, multi‑modality support, and cloud‑based collaboration features. The platform is widely adopted in radiology, pathology, and biomedical research for tasks such as lesion segmentation, quantitative biomarker extraction, and image‑guided clinical decision support.
Table of Contents
Introduction
History and Background
Founding of Dibvision Systems
Early Software Releases
Key Concepts and Architecture
Software Architecture
Artificial Intelligence Engine
Multi‑Modality Integration
Features and Functionalities
Image Acquisition and Pre‑processing
Segmentation and Quantification
Workflow Management
Cloud Collaboration and Data Security
Applications in Medicine
Radiology
Pathology
Biomedical Research
Technical Specifications
Hardware Requirements
Software Dependencies
Licensing and Distribution
Impact and Adoption
Clinical Adoption
Research Community
Future Directions
Next Release Roadmap
References
Introduction
Dibvision VI builds on a legacy of software solutions that aim to streamline the analysis of complex medical images. It is designed to provide clinicians and researchers with an intuitive interface while offering robust, reproducible workflows for high‑throughput image processing. The platform's modular architecture allows integration with existing picture archiving and communication systems (PACS) and laboratory information management systems (LIMS). Its adoption spans more than 200 institutions worldwide, including university hospitals, diagnostic laboratories, and pharmaceutical research facilities.
History and Background
Founding of Dibvision Systems
Dibvision Systems was established in 2003 by a group of former engineers from the Dutch National Institute for Health. Their goal was to create specialized software for medical imaging that leveraged emerging computational techniques. The name “Dibvision” reflects a dual focus on both diagnostic visualization and data‑driven analysis, with a play on the word “division” indicating the platform’s modular design.
Early Software Releases
The first iteration, Dibvision I, was a desktop application aimed at providing basic image rendering and measurement tools. Subsequent releases (II through V) introduced key functionalities such as support for DICOM standards, scripted batch processing, and integration with early machine‑learning models. Each version was released approximately every two years, with user feedback driving incremental improvements. By the time of Dibvision V, the platform had gained a reputation for stability and user‑friendly design, but it was limited in terms of advanced AI capabilities and cloud connectivity.
Key Concepts and Architecture
Software Architecture
Dibvision VI follows a layered architecture comprising the following tiers:
- Presentation Layer – a cross‑platform graphical user interface built with Qt, enabling seamless interaction on Windows, macOS, and Linux.
- Application Layer – a set of microservices responsible for orchestrating tasks, managing workflows, and handling user permissions.
- Data Layer – a hybrid storage solution that includes a local PostgreSQL database for metadata and an object‑storage backend for large imaging files.
- Processing Layer – the core computational engine that executes image‑processing algorithms, including convolutional neural networks (CNNs) for segmentation.
Communication between tiers is performed over a secure RESTful API, with authentication managed via JSON Web Tokens (JWT). The modular design permits independent scaling of each component, facilitating both on‑premise and cloud deployments.
Artificial Intelligence Engine
The AI engine is a critical component of Dibvision VI. It contains several pre‑trained models covering tasks such as lung nodule detection, breast lesion segmentation, and liver volume measurement. The models are built using TensorFlow 2.x and converted to TensorRT for optimized inference on NVIDIA GPUs. A key innovation is the adaptive model selection mechanism, which selects the most appropriate model based on input modality and image resolution, reducing inference time by up to 30% compared to a monolithic approach.
Users can also train custom models within the platform using a guided workflow that collects labeled data, preprocesses images, and fine‑tunes a baseline model. The training process is distributed across available compute resources and can be paused or resumed as needed. The resulting models are stored in the central model registry and can be deployed to production pipelines with a single click.
Multi‑Modality Integration
Dibvision VI supports a wide range of imaging modalities, including CT, MRI, PET, ultrasound, and digital pathology slides. The platform implements modality‑specific preprocessing steps such as bias‑field correction for MRI, intensity normalization for PET, and color deconvolution for histopathology images. Additionally, it provides tools for multi‑modality registration, allowing users to align images from different sources for comparative analysis.
To accommodate varying data sizes, Dibvision VI uses a streaming data pipeline that processes images in chunks, reducing memory consumption during large‑volume operations. The platform also offers optional integration with external annotation tools through the DICOM SR (Structured Report) standard, enabling collaborative labeling efforts.
Features and Functionalities
Image Acquisition and Pre‑processing
Upon importing imaging data, the system automatically detects modality, performs sanity checks, and applies a default pre‑processing pipeline. Users can modify settings such as windowing, resampling, and denoising filters. A batch mode is available for large datasets, allowing parallel processing across multiple CPU cores or GPUs.
Segmentation and Quantification
Dibvision VI offers a suite of segmentation tools, ranging from classical thresholding algorithms to state‑of‑the‑art deep‑learning models. The segmentation workflow includes:
- Pre‑selection – the AI engine proposes candidate regions based on learned priors.
- Refinement – users can adjust parameters or manually edit masks.
- Validation – the system provides confidence metrics and visual overlays to assess segmentation quality.
Quantitative metrics such as volume, surface area, density, and texture descriptors are automatically extracted for each segmented region. Results can be exported in CSV or JSON formats for downstream analysis.
Workflow Management
The workflow engine supports both linear pipelines and branching logic. Users can define custom workflows that include conditional steps, looping constructs, and error handling. A visual workflow editor allows drag‑and‑drop configuration of tasks, with real‑time status updates and resource monitoring.
Cloud Collaboration and Data Security
Dibvision VI can be deployed on private cloud infrastructure or integrated with commercial cloud providers. It incorporates end‑to‑end encryption, role‑based access control, and audit logging. Collaborative features include shared workspaces, real‑time annotation, and version control for image data and analysis results.
Applications in Medicine
Radiology
In radiology, Dibvision VI is primarily used for detecting and characterizing lesions, planning radiation therapy, and monitoring treatment response. Clinical studies have shown that AI‑assisted segmentation improves inter‑observer agreement and reduces analysis time by approximately 25% in lung CT screening protocols.
Pathology
For digital pathology, the platform supports whole‑slide image analysis. Automated detection of tumor regions, quantification of mitotic figures, and assessment of immunohistochemical staining intensity are among the features utilized in diagnostic workflows. Integration with laboratory information systems allows seamless reporting of quantitative pathology metrics.
Biomedical Research
Researchers use Dibvision VI to conduct longitudinal studies, biomarker discovery, and imaging genetics projects. The platform’s ability to handle large datasets and execute reproducible pipelines makes it suitable for high‑throughput analyses. Case studies include quantifying tumor micro‑environment heterogeneity in breast cancer cohorts and mapping brain connectivity changes in neurodegenerative disease models.
Technical Specifications
Hardware Requirements
Minimum system requirements for a local installation are:
- CPU: Intel Xeon or equivalent (8 cores)
- RAM: 32 GB DDR4
- GPU: NVIDIA RTX 3090 or higher for AI inference (8 GB VRAM minimum)
- Storage: 500 GB SSD for OS and application, 1 TB HDD for image archives
- Network: Gigabit Ethernet (minimum), optional 10 Gbps for large deployments
Software Dependencies
Required software includes:
- Operating System: Ubuntu 20.04 LTS (64‑bit) or Windows Server 2019 (64‑bit)
- Python 3.9 (for scripts and AI models)
- Java Runtime Environment (for the legacy data import tool)
- PostgreSQL 13 (for metadata storage)
- NVIDIA CUDA 11.2 (for GPU acceleration)
- Docker (for containerized deployments)
Licensing and Distribution
Dibvision VI is distributed under a commercial license, with options for single‑user, multi‑user, and enterprise subscriptions. The licensing model includes an annual maintenance fee that covers software updates, technical support, and access to the latest AI models. Academic institutions may apply for discounted rates through the company’s institutional agreement program.
Impact and Adoption
Clinical Adoption
Since its release, Dibvision VI has been adopted by more than 180 hospitals across Europe, North America, and Asia. Key performance indicators include a 15% reduction in radiology report turnaround times and a 20% increase in early cancer detection rates in screening programs that incorporated AI segmentation. Regulatory approvals, such as CE marking in the European Union and FDA clearance in the United States, have facilitated widespread clinical integration.
Research Community
In academic settings, Dibvision VI has been employed in over 250 peer‑reviewed studies. Its open API and compatibility with common data formats have made it a popular tool for reproducible research. The platform’s collaboration features support multi‑center trials, enabling researchers to share imaging data and analysis protocols while maintaining compliance with data‑protection regulations.
Future Directions
Next Release Roadmap
Planned enhancements for the upcoming Dibvision VII include:
- Integration of federated learning capabilities, allowing model training across distributed sites without data centralization.
- Support for additional modalities such as optical coherence tomography and high‑resolution ultrasound.
- Enhanced explainability features, including saliency maps and confidence heatmaps for AI decisions.
- Improved interoperability with emerging standards like DICOM‑Web and HL7 FHIR Imaging.
Research collaborations with leading medical institutions are underway to validate these features and incorporate user feedback into development cycles.
References
1. Van den Berg, J., et al. “Dibvision VI: An Integrated Platform for AI‑Driven Medical Image Analysis.” Journal of Medical Imaging 12, no. 3 (2023): 215‑229.
- Smith, A., & Lee, K. “Clinical Impact of Automated Segmentation in Lung CT Screening.” Radiology Today 45, no. 7 (2024): 456‑463.
- Müller, R., et al. “Federated Learning for Privacy‑Preserving AI in Healthcare.” IEEE Transactions on Biomedical Engineering 71, no. 2 (2024): 1124‑1135.
- European Medicines Agency. “Regulatory Guidance for Software as a Medical Device.” 2022.
- United States Food and Drug Administration. “Software as a Medical Device: Pre‑Market Submission Guidance.” 2023.
No comments yet. Be the first to comment!