Introduction
Medical imaging data is a cornerstone of modern clinical practice and research. However, this data is frequently "trapped" within a hospital’s internal Picture Archive and Communications Systems (PACS). For clinicians and researchers, accessing these images for advanced analysis typically requires specialized local software, high-end hardware, and manual data transfers. These hurdles not only slow down research but also compromise data security and the reproducibility of results.
CHIPS (Cloud Healthcare Image Processing Service) was developed to address these challenges. It is a web-based, containerized platform designed to bridge the gap between siloed hospital databases and powerful cloud computing resources. By providing a seamless, end-to-end workflow from data collection to 3D visualization, CHIPS enables users to securely pull medical images, run complex analysis pipelines, and collaborate in real-time—all within a standard web browser.
System Architecture
The core innovation of CHIPS lies in its distributed, microservice-oriented architecture. The system is designed around the principle of compute portability: the ability to move analysis algorithms to where the data resides, rather than moving massive datasets to the algorithms.
As shown in the architecture diagram, the system is divided into three distinct zones:
- PACS Node: Located within the hospital firewall, this node handles secure authentication and performs on-the-fly anonymization of DICOM data before it leaves the clinical environment.
- Web-Entry Point: This serves as the central hub for data hosting, user management, and the main CHIPS web interface.
- Compute Node: These are remote cloud resources where "plugins"—the actual processing code—are executed.
To ensure consistency across these diverse environments, CHIPS utilizes pervasive containerization. Every component, from the backend database to individual analysis tools like brain segmentation pipelines, is housed in Docker containers. This ensures that a tool will run identically regardless of whether it is deployed on a local server or a public cloud provider like AWS or Google Cloud.
The "Feed" Metaphor and Workflow
Traditional medical imaging software often relies on complex file-and-folder structures that can be difficult to navigate during multi-step research projects. CHIPS introduces a "feed-based" workflow, inspired by social media timelines. Each processing step applied to a dataset creates a new branch in a logical tree, providing a persistent and intuitive record of the research history.
This interface allows users to track the progress of various stages, such as PACS retrieval, pre-processing, and curvature analysis. By leveraging modern browser standards (HTML5 and WebGL) and the XTK toolkit, CHIPS provides high-performance 3D rendering of medical volumes directly in the browser. This eliminates the need for client-side installations and allows for real-time collaboration; multiple users can interact with the same 3D model simultaneously, facilitating remote consultations and joint research efforts.
Key Insights and Results
CHIPS transforms a collection of images into a searchable "Big Data" repository. During ingestion, the system extracts non-private DICOM tags (such as scanner type and acquisition parameters) into a structured database. This metadata mining allows researchers to perform large-scale clinical informatics that were previously impossible with siloed data.
The platform has been successfully deployed at Boston Children’s Hospital, demonstrating its ability to securely bridge internal networks and external cloud nodes. It effectively orchestrates long-running, resource-intensive pipelines, such as FreeSurfer for MRI analysis, and returns organized, shareable results to the user.
Conclusion and Future Work
CHIPS represents a significant step toward making advanced medical image processing more accessible and collaborative. By combining containerization with a web-centric approach, it removes the hardware and software barriers that often hinder clinical research.
Future development will focus on expanding the system's "Big Data" capabilities, including more advanced data mining frameworks for DICOM metadata. Additionally, the authors plan to streamline the integration of machine learning pipelines, allowing users to build and deploy AI models directly within the CHIPS environment to further accelerate medical discovery.
Summary updated