Table 1 Correlation coefficient between ground truth and estimated mitotic count on all evaluated models, data sets, and all folds of the cross-validation.

From: Deep learning algorithms out-perform veterinary pathologists in detecting the mitotically most active tumor region

Method

Full data set

Ablation study

Cross-val

Fold 1

Fold 2

Fold 3

Fold 1

Fold 2

Fold 3

Segmentation

0.925

0.947

0.963

–

–

–

MC Regression

0.924

0.935

0.948

–

–

–

Object detection, 1 stage

0.916

0.939

0.957

0.868

0.904

0.930

Object detection, 2 stage

0.963

0.971

0.979

0.897

0.924

0.944

  1. We look for the position of the highest mitotic density. Thus correlation is a suitable measure to compare models, as scaling and additional factors have no influence in the subsequent argmax operation.
Back to article page