Fig. 2

NeuriteNet effectively classifies images corresponding to DRGNs grown on unpatterned and topographically micropatterned substrates. A, B Representative images of rDRGNs grown on unpatterned (A) and topographically micropatterned substrates (B). C, D Same images as in A, B that were correctly classified as Unpatterned (C) or Patterned (D). The intensity of the color indicates the relative importance of that area. Green and red indicate areas that were used by NeuriteNet to suggest the image belonged to Unpatterned and Patterned groups, respectively. E, F The representative images of rDRGNs (A, B) with their saliency map overlayed (C, D). G Comparison of performance of 3 machine learning classification approaches. The percentage of total images or traces (n = 2604, 2604, 241) classified correctly as belonging to pattern and unpatterned groups is shown along with kappa statistic. H Fractional distribution of Predicted Pattern Scores. Color represents the actual group (pattern or unpatterned) to which the image corresponds. NeuriteNet classifies an image as “Patterned” if the Predicted Pattern Score is greater than 0.5 and to “Unpatterned” if less than 0.5. NeuriteNet classified vast majority of images correctly (the small red bar at Predicted Pattern Score of 0.4 (appears brown as it is overlaying the green) represents a miniscule fraction of patterned images falsely classified as unpatterned). Scale bar = 100 µm