In the Zeiss AxioVision formats, ROIs (shapes) are contained within Layers. Sets of ROIs are collected in different layers. The UI only uses a single layer, but uses separate layers for acquisition and post-acquisition ROIs. But in the file format one may define arbitrary numbers of layers to act as a grouping mechanism for ROIs.
Adding layers as a top level grouping would permit related ROIs to be grouped together. However, this would also be possible using ROI-ROI links; it could be implemented using Layer-ROI links. Maybe layer could be a ROI type used solely for grouping?