Figure 3.

Data clean-up using communicators improves network’s performance. Confusion Matrices from retraining the neural network with (a) original test data or (b) with data post-clean-up via communicators are shown. (c) Heatmap of the performance difference between the network trained on data with and without the data clean-up via the communicators is shown. (d) Representative test spots from healthy lymph nodes (HLN), healthy pancreases (HP) and pancreatic ductal adenocarcinoma (PDAC) classified with the neural network are shown (scale bar = 300 µm). HLN (yellow), HP (red), PDAC (blue), background (BG, grey) and adipose tissue (ADI, cyan) were predicted by the retrained CNN.