Recent trends in stem cell-based therapies and applications of artificial intelligence in regenerative medicine

Table 3 Summary of recent artificial intelligence-based stem cells therapies

Study objectives	Applied AI algorithm	Important conclusions	Study group
iPSC-derived endothelial cells Identification without the application of molecular labelling using CNN	CNN	Prediction accuracy was a function of pixel size of the images and network depth. The k-fold cross validation suggested that morphological features alone could be enough for optimizing CNNs and they can deliver a high value prediction	Kusumoto and Yuasa[167] (2019)
Automated identification of the iPSC colony images quality	SVM, k-NN	k-NN yielded 62% of the accuracy which was found to be better than the previous studies of that time	Joutsijoki et al[168] (2016)
Assess automated texture descriptors of segmented colony regions of iPSCs and to check their potential	SVM, RF, MLP, Adaboost, DT	SVM, RF and Adaboost classifiers were concluded to exhibit superior classification ability than MLP and DT	Kavitha et al[169] (2018)
Develop a V-CNN model to distinguish the colony-characteristics on the basis of extracted descriptors of the iPSC colony	CNN	Recall, precision, and F-measure values by CNN were found to be comparatively much higher than the SVM. Colony quality accuracy was found to be 95.5% (morphological), 91.0% (textural) and 93.2% (textural)	Kavitha et al[170] (2017)
Use CNNs with transmitted light microscopy images to find out pluripotent stem cells from initial differentiating cells	CNN	CNN can be trained to distinguish among differentiated and undifferentiated cells with an accuracy of 99%	Waisman et al[172] (2019)
Use machine learning algorithms to analyze drug effects on iPSC cardiomyocytes	NB, KNN, LS-SVM, DT, multinomial logistic regression	Classification accuracy of the algorithm developed was found to be nearly 79%	Juhola et al[173] (2021)
To build an analytical procedure for automatic evaluation of Ca2+ transient abnormality, by applying SVM together with an analytical algorithm	SVM	The training and test accuracies were found to be 88% and 87% respectively	Hwang et al[175] (2020)
To develop a linear classification-learning model to differentiate among somatic cells, iPSCs, ESCs, and ECCs on the basis of their DNA methylation profiles	Jubatus (ML analytical platform)	The accuracy of the ML model in identifying various cell types was found to be 94.23%. Also, component analysis of the learned models identified the distinct epigenetic signatures of the iPSCs	Nishino et al[176] (2021)

AI: Artificial intelligence; CNN: Convolution neural network; DT: Decision tree; ECCs: Embryonal carcinoma cells; ESCs: Embryonic stem cells; iPSC: Induced pluripotent stem cells; k-NN: K-nearest neighbor; LS-SVM: Least-squares support-vector machines; MLP: Multilayer perceptron; RF: Random forest; SVM: Support vector machine.