Figure 4.

A single image of 27 skin cells contains information about an individual’s overall body size.

(A) Panel of palmskin images indicating the different crop sizes used as training inputs. The red boxes highlight the local trunk surface areas of a palmskin larva used for training. Below each cropped area, the size of the trunk surface area, pixel size, and the approximate number of cells contained within the boxed area are labeled. Scale bar, 50 μm. (B) Schematic flowchart of the inputs, Vision Transformer model, and growth size classification. (A, C) Confusion matrix and F-scores for the corresponding crop sizes shown in (A). Scores above 0.7 are shown in black for ease of visualization. (D) The F-scores of Areas #5, #6, and #7. Percent differences are shown above the horizontal lines for intergroup comparisons. Five iterations were conducted, and the data are presented as mean ± s.d.