Table 2.

Summary of existing applications for deep learning in organoid images

Application	Reference	Input	Network architecture	Output	Function
Object classification	[100]	Fluorescently labeled retinal organoids	CNN	Classification of retinal organoids	Recognition and prediction of retinal differentiation in organoids
	[102]	Bright-field images of normal organoids	CNN	Score evaluation based on organoid viability	Continuous monitoring of organoid viability after drug treatment
	[29]	Bright-field images of human intestinal organoids	CNN	Localization of human intestinal organoids	Automatic quantification and localization of organoids in bright-field images
	[70]	Bright-field images of human lung epithelial spheroids	CNN	Classification and localization of polarized and non-polarized lung epithelial spheroids	Analysis of morphological changes in 3D spheroid models
Image segmentation	[48]	Bright-field images of tumor spheroids	CNN	Boundaries of tumor spheroids	Analyzing tumor invasion using EPI and MSEI
	[101]	Fluorescent image of disparate ventricle lumens in each organoid	CNN	Segmentation mask of ventricle lumens	Detection and segmentation of SOX2-lined ventricle lumens
Object tracking	[31]	Bright-field images of human alveolar organoids	DNN	Organoids identification and tracking over time	Tracking and monitoring organoids throughout their entire lifetime

CNN: convolutional neural network; DNN: deep neural network; EPI: excess perimeter index; MSEI: multiscale entropy index