Fig. 2: Plaque classification using deep learning of functional amyloid microscopy data.
From: Chemical imaging delineates Aβ plaque polymorphism across the Alzheimer’s disease spectrum
a Data preparation and preprocessing. Labelled images were manually extracted, categorised, and resized to a standard size (120 × 120). The images were converted to grayscale and adjusted for contrast. (CGP coarse-grained plaques, CP cored plaques, DP diffuse plaques) b Model training and development. The datasets were split into training (n = 307) and test sets (n = 78). Training datasets were split further for training and validation to execute three-fold cross validation and models were trained using various training parameters. The final model was obtaining by averaging the models from three-fold cross validation c Network architecture of the deep learning training model. d Results of the model evaluation using standard classification metrics. The final model achieved an accuracy of 92.3% in the unseen test set (n = 78).
