Post processing - nntrongnghia/TDSI21-Shoulder-Muscle-Segmentation GitHub Wiki

Post processing details

Connected component-based post-processing

• FOR EACH IMAGE IN A FOLD
• Treat all foreground classes as one
• **IF suppression of all but the largest region improves the average foreground Dice coefficient and does not reduce the Dice coefficient for any of the classes : **
• Do it and reiterate for individual classes
• ELSE : Do not apply

Ensemble selection

• FOR EACH MODEL
• collect the predicted niftis from the 5-fold cv and evaluate them against the ground truth
• obtain mean foreground dice
• IF NOT disable_ensembling
• select two models
• models are ensembled by averaging sofmax probabilities the two selected models
• END
• chose the best model (ensembled model) based on the mean foreground dice