Figure S5.
Comparison of validation metrics used to evaluate the prediction accuracy of neural models predicting mitochondria, ER, and Golgi apparatus. (A and B) Ground truth annotations from FIB-SEM volume data from the indicated cells at ∼5 nm isotropic resolution prepared by CF (A) or HPFS (B) were used for training to generate models for mitochondria, ER, and Golgi apparatus. The bar plots show F1, precision, and recall metrics obtained using ground truth annotations not used for training. These values are shown as averages from 20 training iterations spaced at 1,000 intervals, with respective error bars representing maximum and minimal values, calculated after ∼100,000 training iterations. The results also show metrics obtained after fine-tuning with a small number of additional training iterations using ground truth annotations from the naïve cell. Details of datasets, ground truth annotations, and models are summarized in Tables S4, S5, and S2.

Comparison of validation metrics used to evaluate the prediction accuracy of neural models predicting mitochondria, ER, and Golgi apparatus. (A and B) Ground truth annotations from FIB-SEM volume data from the indicated cells at ∼5 nm isotropic resolution prepared by CF (A) or HPFS (B) were used for training to generate models for mitochondria, ER, and Golgi apparatus. The bar plots show F1, precision, and recall metrics obtained using ground truth annotations not used for training. These values are shown as averages from 20 training iterations spaced at 1,000 intervals, with respective error bars representing maximum and minimal values, calculated after ∼100,000 training iterations. The results also show metrics obtained after fine-tuning with a small number of additional training iterations using ground truth annotations from the naïve cell. Details of datasets, ground truth annotations, and models are summarized in Tables S4, S5, and S2.

This Feature Is Available To Subscribers Only

or Create an Account

Close Modal
Close Modal