FIGURE 3.

Workflow of the proposed EL framework for the classification of TIS (Senescent) and proliferating (Control) cancer cells. We investigate two different TL scenarios, in which the classification probability is given by the average of the probabilities predicted employing seven different pretrained networks. In the first actualization, the weights of the pre-trained networks are left untouched, while in the final model, both the weights of the classification layers and the pre-trained networks are fine-tuned. The 3-channel label-free microscopy images in input to the neural networks consist of optical transmission (TRASM), SRS at the 2850  ${cm}^{-1}$ Raman mode of lipids (SRS), and TPEF of the intrinsic NADH and FAD coenzymes (TPEF).