期刊
JOURNAL OF INTELLIGENT MANUFACTURING
卷 -, 期 -, 页码 -出版社
SPRINGER
DOI: 10.1007/s10845-023-02167-4
关键词
3D bioprinting; 3D printing; Biofabrication; Machine learning; Drop-on-demand bioprinting; Living cells
In this study, a machine learning approach was used to predict the number of cells within printed droplets based on droplet velocity at two different points along the nozzle-substrate distance. A contactless method combining an optical system and machine learning algorithms was employed for cell detection within single droplets and prediction of the total number of printed cells within multiple droplets. Various machine learning algorithms were compared, and the random forest regressor algorithm achieved the highest accuracy in cell prediction within single droplets and the extra tree regressor algorithm demonstrated the lowest mean error in predicting the number of printed cells within multiple droplets. Incorporating these models in a droplet monitoring system enables live assessment of the number of printed cells during an inkjet-based bioprinting process.
In this work, our proof-of-concept study can be used to predict the number of cells within printed droplets based on droplet velocity at two different points along the nozzle-substrate distance using machine learning approaches. A novel high throughput contactless method that combines the use of an optical system and machine learning algorithms was utilized for various applications such as cell detection within single droplets (presence/absence of cells) and prediction of the total number of printed cells within multiple droplets by measuring the droplet deceleration between two positions along the nozzle-substrate distance. The proposed method in this work has demonstrated good accuracy in cell prediction within single droplet (presence/absence of cells) and low prediction error in determining number of cells within multiple droplets by reducing the error by a factor of vN for N droplets measured in a batch. The performance of five different machine learning algorithms such as linear regression, support vector regression, decision tree regressor, random forest regression, and extra tree regression were compared to determine the best algorithm for each type of application. The random forest regressor algorithm demonstrated the highest accuracy at 80% in cell prediction (presence/absence of cells) within single droplets, while the extra tree regressor demonstrated the lowest mean error of 12% in predicting the number of printed cells within multiple droplets (e.g., 20 droplets on same spot). By incorporating these models in a droplet monitoring system, live assessment of the number of printed cells during an inkjet-based bioprinting process can be achieved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据