Table 1.

Accuracy of U-Net based class predictions

Prediction accuracy and errors for all classes of trigeminal neuron in the validation set. Columns are predicted classes (ND, not determined); rows are the manually segmented, clustering based classification. Shades of blue highlight the proportion of predicted neurons that fall into a cell. Overall prediction accuracy was 94.8 % at a pixel level and 84.9 % at the cellular level. It should be noted that when a subset of predicted neurons that did not match the manually segmented cells were examined, many turned out to have the gene expression profile of the predicted class rather than that of the manual segmentation approach. This is not surprising since manual segmentation relies on subjective assessment of cell boundaries and inaccuracies in this process are likely to affect classification for a subset of cells.