Quality Improvement of Embryonic Bodies on Integrated Spheroid Culture Chip by Using 3DCNN

Abstract

Embryonic body (EB) formation has become a routine step in differentiation of stem cells. EB size influences the efficiency of differentiation induction. To induce differentiation with high reproducibility in mass-production, it's very important to exclude irregular EBs at an early stage. In this paper, we propose a system that cultivates many EBs at one time and predicts EBs quality by machine learning. EBs were cultured by TASCL, and time-lapse image of each well was took every 30 minutes. Then, we input 6 images up to 3 hours after seeding as one training data to the three-dimensional convolutional neural network (3DCNN) and predicted whether EBs were formed 1 day after seeding. As a result, the accuracy was 96.5% in the test data. Furthermore, we predicted EB diameter 3 days after seeding by inputting 12 images up to 6 hours after seeding into 3DCNN. As a result, the prediction error was ±7.1μm.