An HTML5-Based Pure Website Solution for Rapidly Viewing and Processing Large-Scale 3D Medical Volume Reconstruction on Mobile Internet

Liang Qiao, Xin Chen, Ye Zhang, Jingna Zhang, Yi Wu, Ying Li, Xue-Mei Mo, Wei Chen, Bing Xie, Mingguo QiuInternational Journal of Telemedicine and Applications 2017 · DOI

TL;DR

This paper introduces a novel "Master-Slave" web-based system for viewing and processing large 3D medical scans on mobile devices. It achieves this by offloading heavy rendering to a server while using a lightweight 3D model on the client for interaction, enabling fast, high-quality visualization without special software or hardware.

What Problem Does This Paper Address?

Medical imaging, particularly 3D reconstructions from CT or MRI scans, generates massive datasets. Accessing and interacting with these large volumes on mobile devices, which are increasingly used for medical purposes, presents a significant challenge. Traditional methods either require powerful local hardware and specialized software (limiting mobile access) or involve sending raw data over the network, leading to slow loading times and poor user experience. Existing remote rendering solutions often suffer from high network load and can be clunky, requiring plugins or custom software, which hinders widespread adoption, especially in diverse network conditions. The authors aim to bridge this gap by providing a pure web solution that offers a smooth, native-like experience for 3D medical visualization on any device with a web browser.

Methods

The core innovation is the "Master-Slave" dual-channel interaction mode.

  • Master Volume (Server-side): This is the high-quality, full-resolution 3D medical volume data. The server handles the computationally intensive volume rendering (using ray-casting) based on user instructions.
  • Slave Model (Client-side): This is a significantly downsampled, simplified 3D geometric model (surface rendering) of the anatomical structure. It's generated by resampling the original scans and then using algorithms like Marching Cubes to extract contours. This model is lightweight (around 1MB) and runs directly in the browser via WebGL, providing a responsive interface for user interactions like rotation, panning, and clipping.

The interaction flow is as follows:

  1. The client downloads the Slave model.
  2. The user interacts with the Slave model on their device.
  3. These interactions (viewing angle, clipping parameters) are sent as "request instructions" to the server.
  4. The server uses these instructions to perform volume rendering on the Master volume.
  5. A 2D projection image of the high-quality Master volume is sent back to the client for display.

This approach separates the interactive navigation (handled by the lightweight Slave model on the client) from the high-fidelity rendering (handled by the powerful Master volume on the server), optimizing for both responsiveness and image quality.

Datasets

The authors used three groups of anonymized CT scans from the radiology department of Southwest Hospital of Chongqing for testing:

  1. Head and neck CT: 495 slices, 512x512 resolution, 249.4 MB.
  2. Trunk CT: 609 slices, 512x512 resolution, 306.5 MB.
  3. Thorax and mandible CT: 500 slices, 512x512 resolution, 251.6 MB.

The primary dataset used for demonstration and detailed analysis was the head and neck CT.

Key Insights and Ideas

  • Decoupling Interaction and Rendering: The most significant insight is the separation of user interaction (handled by a lightweight client-side model) from high-quality rendering (offloaded to the server). This is crucial for achieving responsiveness on resource-constrained mobile devices.
  • Pure Web Solution: By leveraging HTML5 and WebGL, the system avoids the need for any client-side installations or plugins, making it universally accessible through standard web browsers.
  • Master-Slave for Efficiency: The Master-Slave architecture effectively balances the need for detailed visualization with the constraints of mobile internet bandwidth and device processing power.
  • Interactive Clipping: The ability to perform interactive clipping directly on the 3D model on the client side, which then translates to server-side rendering, is a valuable feature for exploring internal anatomy.
  • Lightweight Slave Model: The successful compression of the 3D model to around 1MB while retaining sufficient detail for navigation is a key enabler for fast client-side performance.

Key Results

  • Performance on Low Bandwidth: The system demonstrated functionality even with a limited 100-KBps wireless bandwidth (an extreme test).
  • Fast Initial Loading: The first loading time for the entire application and initial data was less than 12 seconds.
  • Responsive Interaction: The response time for each behavior request (e.g., viewing angle change, clipping) to get the final high-quality image was approximately 1 second.
  • Low Bandwidth Consumption: Peak bandwidth usage for behavior requests was less than 50 KBps, and idle status bandwidth was negligible (<1 KBps).
  • High Frame Rates: The Slave model rendering on the client achieved frame rates of ≥40 FPS.
  • Multi-tasking Capability: The system could handle more than four independent tasks simultaneously, each with a large data size (249.4 MB).
  • User Experience: Radiologists reported positive feedback on imaging quality and interactive experience, with a good user experience similar to native applications.

Limitations and Future Work

  • Slave Model Detail: One radiologist found the default skeleton Slave model (1.04 MB) to be "a little rough," suggesting that users might prefer higher-detail Slave models (e.g., 5.26 MB skeleton or 913 MB skin models) if available, though this would increase initial download size.
  • Network Dependency: While optimized for low bandwidth, performance is still dependent on network conditions.
  • Future Work: The authors suggest that this platform could serve as a technological basis for online communication among doctors and patients and could be easily integrated into other telemedical systems. They also highlight its potential for medical education and remote consultation due to its low cost and accessibility.

Summary updated

Abstract

This study aimed to propose a pure web-based solution to serve users to access large-scale 3D medical volume anywhere with good user experience and complete details. A novel solution of the <i>Master-Slave</i> interaction mode was proposed, which absorbed advantages of remote volume rendering and surface rendering. On server side, we designed a message-responding mechanism to listen to interactive requests from clients (<i>Slave</i> model) and to guide <i>Master</i> volume rendering. On client side, we used HTML5 to normalize user-interactive behaviors on <i>Slave</i> model and enhance the accuracy of behavior request and user-friendly experience. The results showed that more than four independent tasks (each with a data size of 249.4 MB) could be simultaneously carried out with a 100-KBps client bandwidth (extreme test); the first loading time was <12 s, and the response time of each behavior request for final high quality image remained at approximately 1 s, while the peak value of bandwidth was <50-KBps. Meanwhile, the FPS value for each client was ≥40. This solution could serve the users by rapidly accessing the application via one URL hyperlink without special software and hardware requirement in a diversified network environment and could be easily integrated into other telemedical systems seamlessly.
BibTeX
@article{qiao2017html,
  author = {Liang Qiao and Xin Chen and Ye Zhang and Jingna Zhang and Yi Wu and Ying Li and Xue-Mei Mo and Wei Chen and Bing Xie and Mingguo Qiu},
  title = {{An HTML5-Based Pure Website Solution for Rapidly Viewing and Processing Large-Scale 3D Medical Volume Reconstruction on Mobile Internet}},
  journal = {International Journal of Telemedicine and Applications},
  year = {2017},
  doi = {10.1155/2017/4074137},
}
Figure