A Review of Deep Learning-Enhanced Imaging Methods for Microrobots

Microrobots have emerged as effective tools for non-invasive tasks related to biomedical applications, such as drug delivery. Their operation in complex environments, such as in vivo applications, requires accurate real-time imaging techniques. However, imaging methods usually present several challenges, such as low contrast, limited spatiotemporal resolution, and restricted depth penetration. To address these issues, deep learning (DL) offers an effective solution by automatic extraction of features from microrobot images, even with dense backgrounds. This paper reviews recent advancements to enhance biomedical imaging methods, such as optical, magnetic field-based, and ultrasound-based methods, by using DL algorithms for the detection and tracking of microrobots. It also highlights the existing challenges and offers directions for future studies in improving imaging performance and extending the scope of the application of microrobots.