Fig. 2. Internal and external generalization of FIN.

Each output of FIN and MH-PR algorithm is generated using the same raw holograms (M = 3). FIN was trained on human lung tissue samples only, and the internal generalization part uses unseen lung tissue holograms from new patients. The external generalization directly applies the same trained FIN model to new types of samples, i.e., Pap smear, prostate tissue, and salivary gland tissue samples, never seen by the network before. The ground truth for each sample is obtained through the MH-PR algorithm that used M = 8 raw holograms captured at different sample-to-sensor distances