Figure 3
From: Tens of images can suffice to train neural networks for malignant leukocyte detection
A CNN trained on the small ALL dataset is able to discriminate lymphoblast cells from other blood cells with high sensitivity and specificity. (a) During training one of the ten folds (training set size n = 200) the validation loss nicely follows the training loss. (b) The trained CNNs yield high receiver operating characteristics with an ROC-AUC of 0.97 ± 0.02 (mean ± s.d., n = 10 folds). Shown is the mean curve of the 10 folds.
