Skip to main content
Fig. 1 | Acta Neuropathologica Communications

Fig. 1

From: Validation of machine learning models to detect amyloid pathologies across institutions

Fig. 1

The Tang dataset was broken down into four unique datasets. The Train (29 WSI) and Val (4 WSI) datasets were used for training a CNN model. The Test (10 WSI) was used to test the performance of the trained model (i.e. generated ROC and PRC curves). These three datasets went through the process of tiling, which extracts small images (256 by 256 color images) and labels them for their inclusions of amyloid deposits. The Test dataset with the addition of the Tang Hold-out (20 WSI) dataset where used to generate confidence heatmaps and CNN scores for each of the three amyloid morphologies (for each WSI). The Emory dataset (40 WSI) was used to also generate confidence heatmaps and CNN scores. The CNN model architecture is shown on the right. The architecture includes six convolutional layers with max pooling layers and with two dense layers (512 and 100 nodes respectively) at end. The CNN model inputs are red-green-blue 256 by 256 images and it outputs three class probabilities (one for each amyloid pathology)

Back to article page