Duration: | 2022 - current |
Technologies: | Python, Django, Docker, Postgres, RQ, PyTorch, MONAI, JavaScript, Cornerstone3D |
Collaborators: | Roy Wiggins, James O'Callaghan |
Website: | gravis-imaging.org |
GRAVIS is a web-based open-source visualization and annotation tool designed for use with large, multi-dimensional DICOM datasets. It was designed to display GRASP datasets reconstructed at high time resolutions without undue latency or processing delays.
Digital subtraction angiography (DSA) with direct catheterization is currently the gold standard for mapping abnormalities of the brain’s vascular system, such as arteriovenous malformations, dural arteriovenous fistulae, or aneurysms. The invasiveness of DSA, however, introduces well-recognized risks to the fragile cerebral vasculature. Existing non-invasive alternatives based on CT or MR imaging achieve only limited temporal and spatial resolution. Our group developed a novel solution for dynamic MR angiography using the golden-angle radial acquisition (GRASP) technique that promises to overcome these limitations and permits 1mm whole-brain angiograms from gadolinium bolus passage with sub-second temporal resolution.
The amount of 4D image data generated with this new technique can easily exceed 4GB per case, which poses a major challenge for existing viewing software and severely complicates the translation into clinical practice. To address this challenge, a dedicated web-based visualization software called “GRAVIS” has been developed, which has been specifically designed to streamline the processing, visualization, and analysis of large multi-dimensional datasets, aiming to enable routine clinical use of the GRASP MRA technology. In addition to brain MRA, the GRAVIS software has also found application for radio-surgical treatment planning of brain tumors, such as metastases and schwannomas. To this end, the GRAVIS Onco pipeline has been developed, which provides automatic multi-parametric analysis of the datasets. After reading cases, findings and analysis data can be transferred to a regular PACS for archiving.
GRAVIS has been written in Python, HTML, and JavaScript. The backend runs on the Ubuntu 22.04 operating system and uses the Django web framework and RQ queuing library to serve requests and coordinate processing steps. Processing operations can be implemented either natively in Python or by invoking Docker containers, providing the flexibility to integrate advanced algorithms such AI models developed in PyTorch or MONAI. GRAVIS uses the PostgreSQL database for case management and NGINX webserver for front-end communication.
On the front-end side, GRAVIS uses the Cornerstone3D JavaScript DICOM library to render cases. Due to the web-based architecture and server-side execution of processing steps, cases can be viewed on regular desktop computers without need for large memory or powerful CPUs. Moreover, due to GRAVIS’s unique adaptive load mechanism, cases open without significant wait time, which previously posed a major obstacle for clinical utilization of the GRASP MRA technique. GRAVIS precalculates most computationally-demanding preprocessing steps, including AI-based object segmentation and multi-planar reformation of datasets. Readers can be informed about the readiness of cases through configurable messaging services.
GRAVIS is based solely on open-source libraries and has been released as open-source software, aiming to offer a community solution for processing dynamic MR angiography and similarly data-intensive applications. Furthermore, GRAVIS can be used as development platform for other clinical applications requiring visualization of large-scale or multi-dimensional DICOM datasets.