Fig. 2
From: Deep learning-based selection of human sperm with high DNA integrity
Results for the test set (20% of all cells, sampled evenly over all donors) show actual DFI versus predicted DFI (red dashed line shows actual = predicted for reference), as well as four highlighted groups: the actual-lowest 10% (green circles), predicted-lowest 10% (green ×’s), actual-highest 10% (magenta circles), and predicted-highest 10% (magenta ×’s). The blue-bordered images represent cells that the model predicted the most accurately. In general, the model performs better with fewer obstructions to the sperm head, fewer background features, and for more regularly shaped sperm heads
