Journal: Bioengineering & Translational Medicine
Article Title: An AI ‐assisted integrated, scalable, single‐cell phenomic‐transcriptomic platform to elucidate intratumor heterogeneity against immune response
doi: 10.1002/btm2.10628
Figure Lengend Snippet: Dual deep learning computer vision algorithms deployed for automated trap identification and cytotoxicity analysis. (a) Automated computer vision analysis was divided into four stages: (1) the generation of a microfluidic structure ground truth data set and training of the first regions with convolutional neural network (RCNN‐1) to identify microfluidic cell trapping structures, (2) the deployment of RCNN‐1 to identify and enumerate relevant cell traps in the actual experimental data set, (3) the generation of a cell type (i.e., tumor cell, NK cell) and status (i.e., dead, live) ground truth data set and training of RCNN‐2 to identify cell death events indicated with caspase 3/7 reporter dye, and (4) deployment of RCNN‐2 to track and quantify cell death events in experimental data sets. (b) A flowchart detailing the specific steps taken at each stage in the computer vision analysis process.
Article Snippet: RCNN‐1 was a pretrained CNN from MATLAB with layer architecture based off Resnet‐50.
Techniques: