NA-MIC Project Weeks

Back to Projects List

Kidney and tumor segmentation for surgery planning

Key Investigators

• Steve Pieper (Isomics, USA)
• Sylvia Ladstatter (Children's National, USA)
• Kevin Cleary (Children's National, USA)

Project Description

Our overall project aims to help automate kidney surgery and requires a fast and accurate way to make detailed segmentations of renal structures. Currently we can do this in 3D Slicer in a few hours using existing segmentation techniques.

We would like to test improved methods for this task, and also define a good terminology for it.

Objective

1. Have a supervised segmentation method that works well with standard pre-op clinical images (typically diagnostic CTs with contrast enhancement at 1mm or smaller pixel size).
2. The method should segment the following structures:
   • Aorta
   • Vena cava
   • Renal cortex
   • Renal artery (including accessories, inside and outside the kidneys)
   • Renal vein
   • Renal pyramids / medulla
   • Renal pelvis
   • Ureters
   • Tumors / masses
3. Define a good terminology and map it to SNOMED terms.
4. The method should work well on a wide range of clinically realistic cases, such as noisy images and anatomical variants.
5. Ideally a method should also work on non-contrast CT and MR as well

Approach and Plan

1. Use test data from IDC (KiTS data) as a testbed. See if there are other datasets we could use for testing.
2. Meet with experts to discuss state-of-the-art approaches and find out about any existing kidney segmentation models we can try
3. Experiment with ScribblePrompt, MultiverSeg, VISTA-3D, and Radiology Copilot for this task
4. Get input from the IDC team and others on terminologies for this task

Progress and Next Steps

1. Describe specific steps you have actually done.

Illustrations

Background and References

Example KiTS case from IDC: https://viewer.imaging.datacommons.cancer.gov/viewer/1.3.6.1.4.1.14519.5.2.1.6919.4624.135173370342136417423953641748