Deep learning for polymer scaffold bioimage analysis: Opportunities and challenges

Significant efforts have been made to advance bioprinted scaffold research in cell biology, tissue engineering, and drug screening studies. Ideal scaffolds should demonstrate suitable mechanical properties, excellent biocompatibility, and bioactivities. However, the design and preparation of such scaffolds are challenging. Imaging modalities including magnetic resonance imaging (MRI), micro-computed tomography (Micro-CT), ultrasound imaging (UI), optical coherence tomography (OCT), and confocal laser scanning microscopy (CLSM) are commonly used to visualize the interior architecture of bioprinted scaffolds as well as the surrounding cells and tissues. The obtained bioimages provide the first-hand information of scaffold biological functionalities, while the interpretation of such images can readily give rise to divergent viewpoints and even arguments. This study is to review and explore deep learning (DL) methods in those image analysis through restoration, segmentation, and classification. Firstly, the current DL methods for biological image processing are summarized, such as convolutional neural network, U-Net and generative adversarial network. The corresponding outcomes of these methods can reveal the cell-scaffold and tissue-scaffold interactions and guide scaffold design in each specific task. Last but not least, the challenges and limitations of DL applications in this area are highlighted, such as building DL models using smaller bioimage datasets, interpreting DL models, vision-language model guided bioimage analysis, and developing intelligent analysis platforms. These studies would mark a paradigm shift in polymer scaffold designs and the associated performance.