Recent Trends in Deep Neural Networks and Their Hardware Implementation for Biomedical Applications

The development of dedicated deep learning (DL) accelerators and neuromorphic processors has opened up new possibilities for implementing neural network algorithms in healthcare and biomedical applications. These specialized hardware systems present unmatched opportunities to improve the effectiveness and usefulness of DL algorithms in medical contexts, including image enhancement, disease diagnosis, and treatment monitoring. Although DL has progressed rapidly in terms of network complexity, effectiveness, and training efficiency, the evolution of specialized hardware tailored for DL algorithms lags behind. Innovative advancements in algorithms and architectures are essential to fulfill the demanding standards of efficient and affordable healthcare devices. Implementing these algorithms at the edge level is imperative to safeguard data privacy, particularly in critical medical applications. By processing data locally on specialized devices, edge computing reduces the risk of sensitive medical information being exposed to external servers, ensuring compliance with privacy regulations and enhancing patient trust. Edge computing not only minimizes the risks of data breaches but also supports applications where low-latency responses are essential, such as in wearable health monitors or portable diagnostic tools. In this study, we evaluate the numerous technological advances in DL approaches used on medical images for classification, recognition, and enhancement, both at the algorithmic and hardware levels that have been reported to address a variety of healthcare problems. Finally, we present our analysis and discuss our views on the benefits, drawbacks, and opportunities that different algorithms and neuromorphic processors bring to the biomedical and healthcare domains. This analysis signifies the importance of a synergistic evolution of algorithms and hardware to revolutionize modern healthcare.