Figure 6.

Hyperparameter Tuning illustrates the performance of different network architectures trained with variable learning rates and optimizers. (a) Performance of 72 trained and validated neuronal networks were ranked regarding the four-score, highlighting the best 10 network configurations. (b) Correlations were tested between F1-score and four-score via r2-score. (c,d) PCA (linear kernel) of the network metrics from hyperparameter tuning colored by the (c) different optimizers and architectures, (d) learning rates and architectures, and (e) PCA (linear kernel) of the modified four-score parts: HLN-score, HP-score, PDAC-score and LNMP-score (vs. the pathologist annotations over all 29 validation images).