Abstract 10444: Building Mycoplasma Checking System Using Image Based Deep-learning in Regeneration Medicine Products

• Published in: Circulation (New York, N.Y.) N.Y.), 2019-11, Vol.140 (Suppl_1 Suppl 1), p.A10444-A10444
• Main Authors: Iseoka, Hiroko, Miyagawa, Shigeru, Sasai, Masao, Date, Satoshi, Ayame, Hiroto, Nishida, Azusa, Sawa, Yoshiki
• Format: Article
• Language: English
• ISSN: 0009-7322; 1524-4539
• DOI: 10.1161/circ.140.suppl_1.10444

BackgroundCell therapy has been developed as promising strategy for treatment of heart failure; however, major concerns in clinical cell therapy may be how to reduce the cost in manufacturing process which mainly depend on quality control as well as its efficacy. Recent progress in technology allow Artificial intelligence to increase the work efficiency in many fields. Therefore, we hypothesize that development of mycoplasma judgement system using convolutional neural network (CNN), one of image based deep-learning, promise shortening of test time and high sensitivity.Methods and ResultsPrototype of mycoplasma judgement program consisting of two layers (prediction layer, cell counting layer) was constructed using DNA stained images of positive and negative control (Vero cells infected with/without mycoplasma, respectively) as educational data. To assess the accuracy of prototype program, test data (images of Vero cells infected with mycoplasma of 0, 10, 100CFU) was input to program. The program calculated the number of mycoplasmas-infected cells, which was increased according to the CFU of infected mycoplasma and mycoplasma positive sample was judged correctly by this prototype program. Furthermore, based on prototype program, improved program consisting of three layers (mycoplasma layer, prediction layer, cell counting layer) was constructed. The improved program has abilities to verify infected or non-infected cells recognized by the program and the result of test data were comparable to those of the prototype program. Finally, we compared test time and sensitivity of judgement by improved program and visual test by human, resulting that the program is more sensitive than visual test and test time of program was 1/20th of that of visual test.ConclusionMycoplasma checking system for regeneration medical products using CNN promise improvement of efficiency and sensitivity of quality test in manufacturing process, suggesting that this system may contribute to cost reduction in quality control as well as prevalence of cell therapy.