Machine learning-enabled optimization of extrusion-based 3D printing.

Machine learning (ML) and three-dimensional (3D) printing are among the fastest-growing branches of science. While ML can enable computers to independently learn from available data to make decisions with minimal human intervention, 3D printing has opened up an avenue for modern, multi-material, manufacture of complex 3D structures with a rapid turn-around ability for users with limited manufacturing experience. However, the determination of optimum printing parameters is still a challenge, increasing pre-printing process time and material wastage. Here, we present the first integration of ML and 3D printing through an easy-to-use graphical user interface (GUI) for printing parameter optimization. Unlike the widely held orthogonal design used in most of the 3D printing research, we, for the first time, used nine different computer-aided design (CAD) images and in order to enable ML algorithms to distinguish the difference between designs, we devised a self-designed method to calculate the "complexity index" of CAD designs. In addition, for the first time, the similarity of the print outcomes and CAD images are measured using four different self-designed labeling methods (both manually and automatically) to figure out the best labeling method for ML purposes. Subsequently, we trained eight ML algorithms on 224 datapoints to identify the best ML model for 3D printing applications. The "gradient boosting regression" model yields the best prediction performance with an R-2 score of 0.954. The ML-embedded GUI developed in this study enables users (either skilled or unskilled in 3D printing and/or ML) to simply upload a design (desired to print) to the GUI along with desired printing temperature and pressure to obtain the approximate similarity in the case of actual 3D printing of the uploaded design. This ultimately can prevent error-and-trial steps prior to printing which in return can speed up overall design-to-end-product time with less material waste and more cost-efficiency.

Glioma-on-a-Chip Models.

Glioma, as an aggressive type of cancer, accounts for virtually 80% of malignant brain tumors. Despite advances in therapeutic approaches, the long-term survival of glioma patients is poor (it is usually fatal within 12-14 months). Glioma-on-chip platforms, with continuous perfusion, mimic in vivo metabolic functions of cancer cells for analytical purposes. This offers an unprecedented opportunity for understanding the underlying reasons that arise glioma, determining the most effective radiotherapy approach, testing different drug combinations, and screening conceivable side effects of drugs on other organs. Glioma-on-chip technologies can ultimately enhance the efficacy of treatments, promote the survival rate of patients, and pave a path for personalized medicine. In this perspective paper, we briefly review the latest developments of glioma-on-chip technologies, such as therapy applications, drug screening, and cell behavior studies, and discuss the current challenges as well as future research directions in this field.