Fig. 6

Time-course metrics. For the time-course dataset, the model trained on the primary dataset was applied via an ensemble mechanism with the resulting metrics to test for out-of-domain performance. A shows the Dice and Jaccard scores for rods and rings for each image compared with expert annotation. B shows the per-pixel AUC scores for both the rods and rings. C and D show the correlation between derived features from the segmentations for both the expert analysis and the model. As with the previous primary dataset, correlation here shows the model can recover meaningful biological information as well as pixel-wise accuracy