In this project, we proposed an architecture consisting of 3 modouls: an encoder, a feature selector, and a classifier to achieve high accuracy and f-measure for magnification-independent multi-category (MIM) classification of microscopic biopsy images.
More details are provided in breakhis_classification_report.pdf. This report is organized as follows: Dataset and data distribution in detail, Metrics of evaluation of the model, Background and previous must-known information and details of the selected pre-trained model and feature selector, Architecture of the model including feature representations, classifier selection, and hyperparameters, Results of the comparison with the other researches and finally a visualization of feature embedding vectors after a dimension reduction.