There is significant fluctuation of the image quality during a reading session, even if network characteristics are kept stable. This is because streaming visualization typically uses image compression, and some images can be compressed better without visual artifacts than other images.
Also the achieved framerate is fluctuating significantly during a reading session. User actions such as pan-zoom, scrolling and window-level operations have strong influence on achieved framerate.
Type of image (modality, test pattern) also has a strong effect. Technical test patterns are much more robust to artifacts while for typical radiology images >10% of the pixels is changed. As can be seen in figure 3, the chest Xray image suffers from image artifacts, while the technical test pattern included in the same frame does not show any quality degradation.
This poses an important challenge, because this means that merely relying on visual (or automated) inspection of test patterns as part of a QA procedure for streaming visualization cannot guarantee that image quality of clinical images will be sufficient.
It is not possible to define an overall minimum network bandwidth that is required in order to always achieve sufficient (diagnostic grade) image quality. The required minimum bandwidth depends on the type of modality that is being viewed. For example: higher resolution modalities such as mammography will requiring higher network bandwidth, and modalities with significant moving image contents will require higher achieved framerate.
In general one can state that a stable network bandwidth of less than 30 Mbps will regularly result into reduced visual quality. End-to-end latency of more than 30ms will typically result in degraded user interaction.
