Fig. 2: CL model workflow.

The original, raw TVUS image input (column 1) is fed to the segmentation model to generate a predicted mask (column 2) of the labeled anatomy, which is then provided as input to the CL extraction algorithm. During an intermediate step, the internal and external os are identified from the segmentation mask (column 3) in order to identify the cervical trace (column 3) and then visualize this cervical trace feature overlaid on the original US image (column 4) to measure CL.