Fig. 5: Distribution of error in the species identification using a single 3D RI tomogram.

a The confusion matrix visualizes the overall performance and the frequent errors in the blind test dataset. The row and column indices correspond to the ground truth and the prediction, respectively. The indices of the 19 species are ordered to reflect the common bacterial categories. b The distribution of the second and the third most likely species further visualizes the interspecific similarity recognized by the trained ANN. c, d Individual tomograms are categorized under broader groups including gram-stainability and respiratory metabolism using a modified ANN for each task