You are here: Segmentation > Multi-ROIs > Creating Multi-ROIs from Regions of Interest

Creating Multi-ROIs from Regions of Interest

In a number of workflows, such as labeling a dataset for training a deep learning model, the required input is a multi-ROI. However, you can still choose to work with multiple regions of interest, from which you can create a multi-ROI or dense multi-ROI.

You can also create multi-ROIs from the results of segmentations made with the Histographic Segmentation tool or from previous results produced with the Segmentation Trainer. You should note that whenever you create a multi-ROI from multiple regions of interest, labeled voxels that intersect will be removed from the multi-ROI, based on the order that the regions of interest were selected in the Data Properties and Settings panel.

As shown in the following illustration, in which the labeled voxels of three regions of interest intersect, the selection order determines how labeled voxels are distributed in the resulting classes of the multi-ROI.

Intersecting ROIs and resulting multi-ROIs

Intersecting ROIs

Dense Multi-ROIs

A dense multi-ROI is a multi-ROI in which the labeled voxels of each selected ROI is assigned to a class and all unlabeled, or 'complementary' voxels, are labeled as an additional class. For example, if you create a dense multi-ROI from five regions of interest, then the multi-ROI will contain six classes and will be fully segmented.

As shown in the following illustration, in which the labeled voxels of three regions of intersect, the selection order determines how labeled voxels are distributed in the resulting classes of the dense multi-ROI.

Intersecting ROIs and resulting dense multi-ROIs

Intersecting ROIs

 

Dragonfly Help Live Version