Resource-Efficient Neural Network Architectures for Classifying Nerve Cuff Recordings on Implantable Devices

Abstract

Background: Closed-loop control of functional electrical stimulation involves using recorded nerve signals to make decisions regarding nerve stimulation in real-time. Surgically implanted devices that can implement this strategy have significant potential to restore natural movement after paralysis. Previous work demonstrated the use of convolutional neural networks (CNNs) to discriminate between activity from different neural pathways recorded by a high-density multi-contact nerve cuff electrode. Despite state-of-the-art performance, that approach required too much data storage, power and computation time for a practical implementation on surgically implanted hardware.Objective: To reduce resource utilization for an implantable implementation, with a minimal performance loss for CNNs that can discriminate between neural pathways in multi-contact nerve cuff electrode recordings.Methods: Neural network (NN) architectures were evaluated on a dataset of rat sciatic nerve recordings previously collected using 56-channel (7 x 8) spiral nerve cuff electrodes to capture spatiotemporal neural activity patterns. The NNs were trained to classify individual, natural compound action potentials (nCAPs) elicited by sensory stimuli. Three architecture types were explored: the previously reported ESCAPE-NET, a fully convolutional network, and a recurrent neural network. Variations of each architecture yielded NNs with a range in the number of weights and required floating-point operations (FLOPs). Each NN was evaluated based on F1-score and resource requirements.Results: NNs were identified that, when compared to ESCAPE-NET, required 1,132-1,787x fewer weights, 389-995x less memory, and 6-11,073x fewer FLOPs, while maintaining macro F1-scores of 0.70-0.71 compared to a baseline of 0.75. Memory requirements range from 22.69 KB to 58.11 KB, falling within the range of on-chip memory sizes from several published deep learning accelerators fabricated in 65nm ASIC technology. Conclusion:Reduced versions of ESCAPE-NET require significantly fewer resources without significant accuracy loss, thus can be more easily incorporated into a surgically implantable device that performs closed-loop real-time responsive neural stimulation.

Closed-Loop Control of Functional Electrical Stimulation Using a Selectively Recording and Bidirectional Nerve Cuff Interface

“…every 8 rows of the image belong to a line of contacts running along the nerve), which ensures that rows of This version of ESCAPE-NET uses 2,912,963 weights and 150,013,714 floating point operations required for interference on a single input. Possible optimizations of this neural network architecture have recently been investigated to make it appropriate for use in implanted systems [18], [19] however for these in vivo experiments the architecture was kept fixed.…”

“…Less demanding hardware requirements facilitates implementation on implantable devices. Further reductions of the neural network size are expected to be possible with additional optimization [18].…”

“…A CNN's classification ability is often a result of high computational complexity, which translates to larger devices and more computation time due to higher resource requirements. A CNN optimized for implementation on implantable devices should be designed, with a focus on reducing memory and power requirements, while maintaining reasonable classifica-tion accuracy for its intended application [18]. This would reduce battery size, hardware form factor and operational latency.…”

Time series classification of multi-channel nerve cuff recordings using deep learning

“…While the computational requirements of deep learning models may pose challenges for implantable systems, recent work suggests that CNNs for ENG classification can be reduced in size with only a minor impact on performance [39].…”