Deep Trainer / U-Net_Bones_d_5_fc_32_org

University of Michigan Department of Orthopedic Surgery

Daniella Patton

This is a FCNN with U-net architecture trained to threshold bone from background in femoral neck nano-CT scans (27 micron isotropic resolution). This network has been shown to outperform the the commonly used automatic Otsu threshold method.
Key collaborators in the making of this network were Rob Goulet, PhD ( and Benjamin Provencher ( who completed the criticl work required to optimze the hyperperamters used in this segmentaiton method.

Acknowledgements: P30 grant (AR069620)

Installing extensions requires ORS Dragonfly 3.1 or later.