Tuesday, December 2 | 3:10 p.m.-3:20 p.m. | T7-SSIN03-2 | Room S404
This session focuses on PACS image storage and AI visualization in DICOM formats, in particular, the storage and transfer efficiency of DICOM Grayscale Softcopy Presentation State (GSPS).
In clinical AI workflows, outputs such as regions of interest, annotations, and classification labels are typically stored using either secondary capture or GSPS, according to presenter Abhishek Gupta and colleagues. However, in the new AI era, storage demand, image transfer speed, and maintaining high-quality diagnostic images could test the limits of some PACS.
To compare GSPS to secondary capture, Gupta's group simulated AI-generated outputs for 1,000 studies per modality. They calculated storage requirements and estimated transfer times using a 0.5 Gbps network under a sequential upload scenario, assuming one AI output file per study.
"[Secondary capture] generates a new image series by embedding visual overlays into duplicated images, which increases storage load and consumes full display viewports in PACS viewers," the group explained in their abstract. "In contrast, GSPS references the original DICOM images and stores visual overlays as lightweight, toggleable vector layers, preserving the original images and minimizing additional storage demands."
GSPS demonstrated substantial advantages over secondary capture across all modalities, the group found. For example, storage requirements decreased by nearly 50% for x-ray alone. The presentation will include additional modalities and data, including network transfer time outcomes and other outcomes.
Ultimately, GSPS improves clinical usability by preserving the primary diagnostic image and avoids additional viewport use, according to the group. They say it is "ideal for routine and high-volume clinical settings."
If you are concerned about image storage and transmission with your PACS in the new AI era, attend the Tuesday afternoon session to learn more.
