2017 Winter Project Week/Knee Cartilage Segmentation

From NAMIC Wiki
Jump to: navigation, search
Home < 2017 Winter Project Week < Knee Cartilage Segmentation

Key Investigators

  • Hans Meine (University of Bremen, MEVIS)

Project Description

Objective Approach and Plan Progress and Next Steps
  • Segment knee bones and cartilages for cartilage thickness assessment (osteoarthritis)
  • Train network
  • Save checkpoints for nice learning animation
  • Carefully increase size of training data during training (how much data is needed?)
  • Create nice animation showing how segmentation improves with increasing number of datasets
  • Win Titan X ;-)
  • Code for special kind of training written
  • Network trained (Tue/Wed)
  • Visualization created (Wed/Thu)
  • Code for running visualization on all intermediate states written (Thu)
  • Broke remote TeamViewer connection to machine
    • completely shut out, need to cancel project until return to Bremen

Background and References

Goal / context

Cartilage thickness rendering
  • analysis of cartilage thickness for ostheoarthitis assessment
  • High-resolution data is from the University of Freiburg (T. Lange, K. Izadpanah)
    • patellofemoral joint (prospective motion correction)
    • depicted image is a healthy volunteer
    • cf. CARS 2016 submission

Training worked fine

training log for cartilage segmentation

Planned Visualization

3d rendering of patella with cartilage3d rendering of femur with cartilage2d slice of MRI with segmented structures

Below this should've been the performance over time, but with a rendering of the increasing number of training cases (1..47).

Glimpse at partial animation

This is only the 2d part, I cancelled the rendering because the 3d cameras were not properly set up, and made a mistake, completely shutting myself out of remote access to the machine required for this project. :-(

start of planned animation