Skip to main content
. 2022 Feb 16;11:806603. doi: 10.3389/fonc.2021.806603

Figure 2.

Figure 2

The overall analysis workflow: Training data generation, model training, inference (prediction), and computing threshold values. Training data is generated via a combination of manual annotations and model-generated predictions. A trained model generates predictions in the form of the probability that a patch is TIL-positive. The probability values are mapped to binary classifications by applying an automatically computed threshold.