Table 2.

Comparison table of studies associated with the cell detection and tracking procedure

Paper	AI Architecture	Input data type	Training data size	CNN function	Detection limit	Overall accuracy
Matusevičius et al. (2017) [47]	R-CNN model [48]	Light microscopy image	700 embryo images	Detecting position and size of cells	Only locates cell without stage prediction	Size prediction min error at 11.92% and detection at 5.68%
Rad et al. (2018) [49]	U-Net [25] modified with residual dilation	Light microscopy image	224 embryo images	Locating individual cell(by centroid) and counting total	Up to 5-cells stage	88.2%
Rad et al. (2019) [50]	ResNet model [40] modified with a special encoding & decoding scheme	Light microscopy image	176 embryo images	Locating individual cell(by centroid) and counting total	Up to 8-cells stagel	86.1%
Kutlu and Avci (2019) [51]	Faster R-CNN model [52]	Mouse embryo microscopy image	565 mouse-embryo images	Locating individual cell and counting	Up until 4-cells stage	Averages 95%
Leahy et al. (2020) [37]	Mask R-CNN architecture [53]	Embryoscope time-lapse images	102 embryos labeled at 16,284 times with 8 or fewer cells;	Multiple function pipeline, one of which is locating and segmenting individual cell	Tracking on images that were previously detected to have 1-8 cells	82.8%