Fig. 2: Training new models for mycobacterial cell segmentation.
From: Phenotypic drug susceptibility testing for Mycobacterium tuberculosis variant bovis BCG in 12 hours
a Phase contrast image and the corresponding cell membrane image stained by 3HC-3-Tre dye. b, c Segmentation performance of different training parameters using data excluding (b) and including (c) high-density cell microchambers. Mycobact_2* in (b) is the Mycobact_2 model performed using images from one camera. Inset in (c) is a zoom-in of the x and y-axis. d–i Representative micrographs of mycobact_2 model for cells in various conditions–(d) M. smegmatis on agarose gel, (e) and (f) M. smegmatis in microchamber, (g) M. bovis BCG in microchamber, (h) and (i) Relatively and highly dense M. smegmatis clumps. Ground truth is labeled from phase images. Flow field is an intermediate output from the neural network. Average precision at an IoU threshold of 0.5 (AP@0.5) for the entire image is reported on the top left corner of the outline - scale bar 5 µm.
