Table 2. Applications of convolutional neural network on pluripotent stem cells studies since 2017.

Reference	Title	Objective	Methodology	Findings
Kavitha et al (2017)	Deep vector-based convolutional neural network approach for automatic recognition of colonies of induced pluripotent stem cells	A V-CNN model is developed to distinguish colony characteristics based on extracted features of the iPSC colony	• A transfer function from the feature vectors to the virtual image was generated at the front of the CNN in order for classification of feature vectors of healthy and unhealthy colonies	• Precision, recall, and F-measure values of the V-CNN model were higher than the SVM classifier with a range of 87–93%
			• The performance of V-CNN model in distinguishing colonies was compared with the competitive SVM classifier based on morphological, textural, and combined features	• For the quality of colonies, the V-CNN model showed higher accuracy values based on morphological (95.5%), textural (91.0%), and combined (93.2%) features than those estimated with the SVM classifier (86.7, 83.3, and 83.4%, respectively)
			• Five-fold cross-validation was used to investigate the performance of the V-CNN model	• The accuracy of the feature sets using five-fold cross-validation was higher than 90% for the V-CNN model as compared to SVM that just yielded around 75–77%
Fan et al (2017)	A Machine Learning Assisted, Label-free, Non-invasive Approach for Somatic Reprogramming in Induced Pluripotent Stem Cell Colony Formation Detection and Prediction	A computer vision system to recognize and assist the classification of induced pluripotent mouse embryonic stem cells colonies from microscopic images	• CNN is used as classifier to recognize colonies	• This algorithm shows no significant differences (Pearson coefficient r > 0.9) in detection and prediction of colonies in terms of biological features compared to manually processed colonies
			• Colonies are located and their boundaries are detected by a semi-supervised segmentation method	• Evaluation was completed by standard immunofluorescence staining, quantitative polymerase chain reaction (QPCR), and RNA-Seq for verification of pluripotency
			• Growth phase and maturation time window of colony formation was estimated with trained Hidden Markov Model (HMM) during the reprogramming procedure
			• This system can predict the best selection time window for iPSC colonies to prevent random differentiation caused by overgrowth using data from colonies traced via time-lapse
Kusumoto et al (2018)	The application of CNN to stem cell biology	Recognition of iPSC-derived endothelial cells based on morphological features using CNN	• Images were collected at day 6 of differentiation	• The results proved that identification of iPSC-derived endothelial cells can be made based on morphology alone
			• 200 images were attained from each of four independent experiments and of these, 640 images were allocated for training and 160 were used for validation
			• Under K-fold validation, three experiments that rendered 600 images were used for training phase and 200 images from one experiment were allocated for validation, in every possible combination
			• CNN based LeNet and AlexNet model used
			• F1 scores determined the performance, which indicated the aggregate of recall and precision, and on accuracy (portion of true predictions)
Waisman et al (2019)	Deep Learning Neural Networks Highly Predict Very Early Onset of Pluripotent Stem Cell Differentiation	Use CNNs to precisely forecast the beginning of PSC differentiation in transmitted light microscopy images	• Mouse ESCs were cultured in distinct conditions to maintain the ground state of pluripotency	• CNN model could be trained to recognize the ESCs from differentiating cells within 24–48 h with an outstanding accuracy higher than 99%.
			• Images were taken randomly through EVOS microscope at consecutive hours post differentiation	• This trained CNN was found to detect differentiating cells only minutes after the cells were stimulated to differentiate
			• Light microscopic images of pluripotent stem cells were used to train a CNN to distinguish pluripotent stem cells from early differentiating cells	• So far, this approach is the best cell assay for verifying differentiation in such a short timeframe, due to its accuracy (which is close to 100%) and low cost
			• CNN based ResNet and DenseNet model used	• The performance of the CNN in accurately identifying cellular morphology in microscopic images will definitely have substantial effects on the ways cell assays are conducted in the future.
			• 2134 images were selected for training and 400 for validation (200 in each group). Independent prediction after training phase was done with 100 images (50 per group)

Notes.

CNN

convolutional neural network

iPSC

induced pluripotent stem cells

ESC

embryonic stem cell

SVM

support vector machine

V-CNN

vector-based convolutional neural network