Comprehensive Summary
This paper is focused on using mapping of peripheral nerve fascicles improving targeted electrical stimulation. They developed a deep learning algorithm segmentation model that they trained on histological sections on different peripheral nerves, stained with a variety of different methods. The researchers also used a pre-trained encoder to reduce the need for large dataset training and tested it on different nerve types and stains previously unseen by it. The results showed that the models had minimal error when fascicle-wise recruitment compared to manual segmentation. The models got accustomed to the new nerves and staining methods not originally in its training set. Its auto segmentation also is dependable enough to replace manual tracing for making neuromodulation models. This also should help create large scale use of cadaveric histology to improve stimulation strategies.
Outcomes and Implications
The medical implications of this paper are that its automatic fascicle mapping could save a lot of time and reduce fallible human error. The selective nerve stimulation could also possibly restore motor and sensory function and treat drug resistant conditions but this would require more knowledge on fascicle organization. Clinically, the computer models could lead to the next generation of new devices for bioelectronics and similar fields where this study and its results would be relevant. The researchers also suggest using large datasets to train it more and with a bit more testing it could soon be in clinical use.