This guide explains how to approach AI-powered, semi-automatic, and manual segmentation for CT studies, and how these workflows fit into day-to-day reporting.
CT segmentation is the process of isolating a structure of interest. In day-to-day radiology, that structure might be a tumor, hematoma, organ, cystic lesion, postoperative collection, lung region, or any volume where visual estimation alone is not enough.
The best workflow depends on the anatomy, contrast phase, and how clean the lesion borders are, but most radiologists can think about segmentation in the following order.
Start by identifying the series that gives you the clearest boundary definition. Choose the series that balances boundary clarity and processing time for the structure you want to segment.
Thicker sections usually resolve faster because there are fewer slices to process, while thinner sections are usually more accurate for fine boundaries but require more slices and more review time.
Before placing points, adjust the windowing to create the clearest contrast between the ROI and the surrounding structures. Better window settings usually make both AI-assisted and manual correction steps more reliable.
AI assistance is most useful when the structure has recognizable contrast from surrounding tissue and you want to reduce time spent tracing. A good AI-first workflow gives you an initial region quickly, then lets you correct only what matters.
Place positive points with left click inside the ROI so the model understands what should be included in the mask. And use right click to place negative points outside ROI that needs to be excluded.
Semi-automatic tools are ideal when the first pass is directionally correct but needs local cleanup. This saves time compared with full manual segmentation and helps preserve speed on larger lesions or serial follow-up cases.
When surrounding tissue is being pulled into the mask, place negative points with right click on adjacent structures that should be excluded. This often improves separation without requiring a full redraw.
Manual control is still essential for cases with streak artifact, adjacent soft tissue of similar attenuation, postoperative changes, low-contrast boundaries, or partial volume ambiguity. Even in an AI-assisted workflow, manual review remains the safety layer.
In difficult cases, it is often useful to alternate between point-based correction and manual contour cleanup rather than relying on only one method from start to finish.
Once the mask is acceptable, use the Export to 3D/Volume action to calculate volume. In a slice-based workflow, each accepted slice contributes to the total, so the final volume is built up across the segmented stack.
If you need downstream review or documentation, use Save to Files to preserve the cropped image and polygon mask. That makes it easier to revisit the segmentation later, share the output, or maintain a case record outside the active reading session.
Best when you want a fast first pass on a lesion or structure with readable borders.
Best when the initial region is close and you want rapid correction instead of full redraw.
Best when artifact, anatomy, or pathology makes automation unreliable.
A clinically useful workflow should allow a fast first pass, efficient refinement, and manual override when needed.
3D Slicer is a powerful and respected platform, especially for research, prototyping, teaching, and advanced customization. Radiologists comparing it with tools such as DN Viewer are often weighing a broader research environment against a more focused clinical reading workflow.
Readers who want to evaluate a radiology-focused desktop workflow can explore DN Viewer alongside the considerations discussed in this guide.
