Prototype workflow for automatic whole kidney and compartamental segmentation on GRE data.

1.  Atlas of the kidney was constructed manually from the Patient (id  PP506).  Last TE timepoint (=30.53 ms) was choosen where the contrast between kidney and surrounding tissue is the maximum

2.  Another patient ( RH.41.L)  GRE dataset was chosen for segmentation.  At the moment we ask User to enclose the kidney with the bounding box in the mid-slice. Again the last TE timepoint (39.36 ms) was chosen to maximize the contrast between kidney and surroundings.

 

 

3. Internally full 3D bounding box is constructed which encloses the kidney.

 

4. We activate Large Deformation registration of the Atlas into the rough bounding box of the kidney.

 

 

 

5. After registration runs we get the Whole kidney mask and {Cortex,Medulla} masks over the 2nd patient. Note: the target volume was resliced to isotropic 2.08mm^3  to minimize data loss during the reslicing (since the Z-resolution is very coarse 5mm).

 

 

6.  Same result as above but displayed without Cortex mask.

7. Future improvements:  have better higher resolution Atlas, perhaps even from different protocol.

Have several atlases, match to all and select the best.

Alternatively, only the Whole kidney mask can be registered and then Whole mask can be separated into Cortex and Medulla.  Histogram thresholding is typically not working well for that purpose, so more locally adaptive operator should be used.

8 Below is one such result.  Whole kidney mask was taken from the result above, but then we used Detect Dark Ridges (default parms) to split the Medulla and the remainder is Cortex.