Predictions of mitochondria. Passing through the FIB-SEM volume with contrast was equalized using CLAHE. The video shows image and predictions from naïve Cell 1 HEK293A (not used for training using the model trained with ground truth annotations for mitochondria from Cell 2 HEK293A. Both cells were prepared by CF and imaged at ∼5 nm isotropic resolution. The model identified all mitochondria; comparison of the ground truth annotations and predictions shows correct voxel assignments (true positives, yellow), missed assignments (false negatives, cyan), incorrect assignments (false positives, magenta). The small fraction of false positive assignments predicted by the model are associated with unidentified tubular and spherical structures of small size (Chou et al., 2021).
