Skip to main content
. Author manuscript; available in PMC: 2021 Feb 4.
Published in final edited form as: IEEE Trans Med Imaging. 2021 Feb 2;40(2):585–593. doi: 10.1109/TMI.2020.3031913

Fig. 2.

Fig. 2.

Adaptive training methodologies for the coefficients in the Generalized loss function. (a) Exploratory model where all possible actions (−10%, no change, +10%) are explored at each iteration step for each of the three parameters, and the action triplet yielding the best performance (argminAjLAji+1) is picked. (b) Deterministic model based on rule-based update formula for the three coefficients that can each take one of five possible actions. For example, αP decreases by a factor of 50% if TP/FP is smaller than 1/15. In order to avoid the unbounded divergence of the parameters, the parameters αP and αN are not updated when <FP> ≤5 and <FN> ≤ 1, respectively, where the symbols <·> denote batch-averaged values.