Figure 3.

Workflow for the deep learning segmentation. Step 1: New images were fed into our model (detector) for automatic segmentation. Step 2: Our experts corrected errors manually. Step 3: These corrected segmented images were examined by a pathologist (JBH) for quality control. Step 4: Post-processing was performed to remove unwanted dots or pixels as errors. Step 5: The final gold standard labeled data was used to train our model to improve segmentation accuracy.