2015 Multi atlas based prostate segmentation

• 

  Projects List

• 

  Before project week

• 

  At the and of the project week

Key Investigators

• Paolo Zaffino, ImagEngLab - UMG
• Giampaolo Pileggi, ImagEngLab - UMG
• Salvatore Scaramuzzino, ImagEngLab - UMG
• Peter Behringer, BWH
• Andrey Fedorov, BWH
• Maria Francesca Spadea, ImagEngLab - UMG

Project Description

Objective

• Automatic segmentation of prostate gland on MRI data for preoperative planning and/or guided biopsy.

Approach, Plan

• Use Plastimatch MABS software and the public database hosted by SPL

Progress

• Preliminary results (before project week):

Dice coefficient: median = 0.704

5° percentile = 0.485

95° percentile = 0.850

Average Hausdorff Distance (mm): median = 3.691

5° percentile = 1.790

95° percentile = 7.124

95° percentile Hausdorff Distance (mm): median = 8.903

5° percentile = 4.988

95° percentile = 17.498

Required time:

roughly 10 minutes per case.

• Added in Plastimatch the registration initialization based on the alignment of center of gravity of contours

• Tuned deformable registration parameters

• Tuned atlas selection parameters

• Currently results (at the end of the project week):

Dice coefficient: median = 0.778

5° percentile = 0.482

95° percentile = 0.854

Average Hausdorff Distance (mm): median = 3.005

5° percentile = 1.632

95° percentile = 6.750

95° percentile Hausdorff Distance (mm): median = 8.133

5° percentile = 4.448

95° percentile = 12.895

Required time:

roughly 2 minutes per case.