Image classification using Deep Ensemble Learning and Transfer Learning methods is performed on a small, labeled dataset of multimodal nonlinear optical microscopy images coming from Stimulated Raman Scattering, Two Photon Excited Fluorescence and Optical Transmission, to differentiate proliferating cancer cells from senescent ones, a peculiar phenotype following an anti-cancer treatment responsible for tumour relapse. The superior performances of the Deep Ensemble Transfer Learning approach are compared with other less complex neural network architectures. Ultimately, the predictions of the neural network are evaluated using the Grad-CAM visualization approach, which allows highlighting the most important features in the input images responsible for the labels assigned by the network.
|