Fig. 6.

Automatic masking system of melanosome images by DNN. A) Typical SE-ADM melanosome images in MNT-1 cells (10,000 ×). We manually selected 190 typical images for masking. B) Hand-written masking results from SE-ADM images in (A). C) Schematic diagram of the automatic masking system by DNN. The DNN configuration is the same as in Fig. 5C, with an input range of 50 × 50 pixel. Details of the masking DNN are described in Materials and Methods. D) Recognition accuracy for each learning iteration calculated from unlearned 2000 images. The red line indicates the training accuracy of each cycle and the blue line indicates the smoothed training accuracy. After 1000 trials, the accuracy reached close to 97 %. E) Typical SE-ADM melanosome images in MNT-1 cells (10,000 ×) picked up by the trained DNN system. F) Automatic masking results of melanosomes using the trained DNN system. Scale bars, 500 nm in (B, F).