DICOM Segmentation Quality: What “Good Enough” Labels Look Like for Radiology AI

Radiology AI projects often treat segmentation as a pixel problem. Draw the organ, draw the lesion, export the mask, train the model. In practice, most failures that show up late in validation started