Fig. 6
From: Scalable deep learning artificial intelligence histopathology slide analysis and validation
Illustration of the architecture for the deep learning model with spatial awareness. The diagram represents the model architecture during training since bagging is used to train a model ensemble. There are \(\:N\) datasets as the input to the model along with \(\:N\) classifier blocks that are independently trained in parallel on the data ensembles. During inference, the model only has one input that is processed by each classifier block, and the prediction of each block is averaged to create the final prediction. Each classifier block has 3 inputs, 1 for each of the CNN outputs from the input image pyramid with a depth of 3.
