A Comparison of WebRTC and Conventional Videoconferencing for Synchronized Remote Medical Image Presentation

Abstract

DICOM viewers must fulfill roles beyond primary diagnostic interpretation, including serving as presentation tools in teaching and multidisciplinary conferences, thereby enabling multiple individuals to review images collaboratively in real time. When in-person gathering is not possible, a variety of solutions have been deployed to maintain the ability for spatially separated users to view medical images simultaneously. These approaches differ in their backend architectures, utilization of application-specific optimizations, and ultimately in their end user satisfaction. In this work, we systematically compare the performance of conventional screensharing using a videoconferencing application with that of a custom, synchronized DICOM viewer linked using Web Real Time Communications (WebRTC) technology. We find superior performance for the WebRTC method with regard to image quality and latency across a range of simulated adverse network conditions, and we show how increasing the number of conference participants differentially affects the bandwidth requirements of the two viewing solutions. In addition, we compare these two approaches in a real-world teaching scenario and gather the feedback of trainee and faculty radiologists, who we found to favor the WebRTC method for its decreased latency, improved image quality, ease of setup, and overall experience. Ultimately, our results demonstrate the value of application-specific solutions for the remote synchronized viewing of medical imaging, which, given the recent increase in reliance on remote collaboration, may constitute a significant consideration for future enterprise viewer procurement decisions.

Fig. 1

Schematic comparison of client–server and peer-to-peer network architectures. In the client–server model, each peer establishes a connection only with a server. The server is an intermediary, and so data transmitted between peers must traverse at least two network hops (double-headed arrows). Increasing the total number of participants does not affect the individual peers, but it does increase the number of connections required by the server. Alternatively, in a peer-to-peer model, each peer establishes a direct connection with every other peer, and thus, data exchange requires only a single hop. However, increasing the number of participants results in increasing connection demands for every peer

Fig. 2

Schematic overview of the WebRTC connections established for remote DICOM viewer synchronization. Initial connection negotiation is facilitated by a signaling server, but the continuous data updates are passed directly between peers over an RTCPeerConnection that encapsulates a MediaStreamTrack carrying microphone audio and an RTCDataChannel carrying viewer state updates

Fig. 3

Quantitative comparison of WebRTC and a CVS for synchronized viewing of medical images over adverse network conditions. Image update lag (top row) and error introduced by compression artifacts (bottom row) are shown as a function of increasing network latency (left column) and increasing packet loss (right column). Plotted points indicate means over 100 viewer updates and associated shaded regions span the standard error

Fig. 4

Representative example of image compression artifacts typical of screen sharing with a CVS. The WebRTC synchronization scheme (left) shows the original image by design, while the data compression applied by the videoconferencing server degrades image quality, blurring subtle edges (center). The difference between the two techniques is more clearly visible in the subtraction image (right)

Fig. 5

Comparison of network resources required by the WebRTC and CVS image viewer synchronization methods. Bandwidth utilization is shown for the peer sending the image updates (top row) and a peer receiving the image updates (bottom row) as a function of the total number of meeting participants. Data rates are divided into inbound (left column) and outbound (right column) flows. Plotted points indicate means over 100 sampling windows and associated shaded regions span the standard error