An active learning approach to train a deep learning algorithm for tumor segmentation from brain MR images

Insights into Imaging

Table 3 Metric results of the pre-active learning and post-active learning segmentation models

Model	Epoch	Sensitivity	Positive predictive value	Dice similarity coefficient	Jaccard similarity coefficient	Modified Hausdorff distance
Model A; Reference	52	0.912	0.906	0.906	0.834	3.309
Model B; Baseline	56	0.913	0.842	0.865	0.778	3.710
Model B; Post-active learning	63	0.882	0.876	0.870	0.780	3.837
Model C; Baseline	57	0.849	0.825	0.825	0.718	4.317
Model C; Post-active learning	57	0.907	0.845	0.868	0.777	3.761