Printing the Future: On the Enhancement of Bioprinting Techniques Through Artificial Intelligence

Abstract

Artificial intelligence, or AI, is a collection of techniques through which computers can learn to recognise complex patterns and draw conclusions from those patterns using machine learning (ML) models. A number of different variations and supporting technologies exist, which are expanded upon in this paper. These machine learning models are often supported through computer vision, which allows information to be extracted from visual data. ML capabilities are growing rapidly with the development of faster and more powerful hardware. This advent of AI technologies has a wide range of applications, including biofabrication. Biofabrication is a branch of biotechnology that deals with the automation of tissue engineering, mainly focusing on bioprinting. The most widely used technology within biofabrication is extrusion bioprinting, a printing technique in which a printing nozzle extrudes the bioink (often a hydrogel) in layers to make a 3D object. Bioprinting, as essential as it is to biofabrication, faces a number of hurdles. Extrusion printing is a difficult process to get right, mainly because it is challenging to find the ideal printing parameters, bioink formulation, and the optimal printing path. This means that the printing process is often very slow, as researchers have to make their best guess at these parameters and often have to make multiple passes, before they achieve a result that is close to what they had in mind. Because AI is so good at finding patterns in data, it is a logical solution to the problems that biofabrication is facing. As such, there have been a fair number of researchers who have investigated to what extent, and in which situations, AI can be integrated into bioprinting. This literature review takes a look at the existing research and aims to draw some conclusions on the applicability of AI within biofabrication overall. Generally, the outlook of AI within biofabrication is quite positive, and the realistic applications are wide. Although the results of the research reviewed were overwhelmingly positive, there are some issues to solve, with the largest being data acquisition. Almost all cases where ML was used, a significant positive effect was found on the aspect of bioprinting that the researchers were working on. There was no singular type of machine learning model which was found to be superior to any other between the articles reviewed, instead the selection of model type is highly specific to the application. Unfortunately, the nature of biofabrication means that gathering data for AI to learn from has to be done manually, which is time consuming and expensive. This is mostly the case for applications such as the development of novel bioink formulations, as well as printing parameter optimalisation, and less so for cases such as optimising print trajectory. Some researchers made do with smaller datasets, expanding them by a prediction method called finite element simulation. A good solution in many cases would be open-source databases, so that researchers can combine their efforts. Although datasets used by researchers are often limited, researchers are clearly enthusiastic about the new possibilities AI brings them, and they are quick to judge their results as successful based on their models successfully finding patterns. However, it is often unclear how transferrable and replicable their results are due to the small dataset. This is often not discussed by researchers. A trend within the papers is that more time is spent on deliberating the biofabrication aspects of the research than on the specifics of the models utilised. This is unfortunate, as it decreases the replicability of the research, and prevents possible insights which may be gained from this information. In many cases, researchers fail to mention important information such as the number of datapoints used, the ratio of traini