Figure 2.

Figure 2. Random Forest model, trained on defined cellular features, exhibits moderate performance in fish size classification.

(A) Schematic flowchart of the inputs, Random Forest model, and growth size classification output. (B) Cell segmentation of palmskin images and automated counting of individual SEC cells. The cell numbers are shown in white. Scale bar, 50 μm. (C) Color patch segmentation of palmskin images and automated counting of individual color patches. The color patch numbers are shown either in white or in black, depending on whether the cells merge with neighboring cells of the same color. (B, C) Of note, the same palmskin image is shown in both (B, C) for side-by-side comparison. (D, E) (Top) Schematic illustrations of the extracted features used for machine learning. (i) “Cell coverage” represents the trunk surface area covered by SEC cells. (ii) “Cell count” represents the counting of individually segmented SEC cells. (iii) “Cell average size” represents the average size of each cell, highlighted by the dotted white borders. (iv) “Color patch count” represents the counting of segmented SEC clones. (v) “Color patch count average size” represents the average size of each clone highlighted by the dotted white borders. (Bottom) Confusion matrix and F-scores of the corresponding extracted features listed above. Scores above 0.7 are shown in black for ease of visualization.