Fig. 2: Gram stain attributes classification results.

a A sample of experimentally derived synthetic hologram showing our deep learning model detects and classifies E. coli (EC) and downscaled B. subtilis (DsBS), as well as the detections below our confidence threshold (unknown). The figure also includes in-focused hologram samples for EC and DsBS accompanied by the corresponding reconstructed images at 10 and 20 µm away from the in-focused plane showing distinct diffraction patterns between these two bacteria. b Receiver operator characteristic (ROC) curves and corresponding area under curve (AUC) values of our deep learning model for EC and DsBS with vertical dashed line marking 0.1% false positive rate (FPR). c The confusion matrix showing the accuracy and prediction errors of our classification for each type of bacteria evaluated at 0.1% FPR.