The Evolution and Security of Neural Communication Systems in Healthcare: From Electroencephalography to AI-Driven Brain-Computer-IoT Convergence

Abstract

Neural communication systems, which carry information out of the nervous system without using nerves or muscles, have moved in a century and a half from a curiosity of animal electrophysiology to clinically deployed speech prostheses that operate unsupervised in the home, placing them at the centre of modern healthcare. This survey traces that trajectory across seventy primary sources and organises it into four eras: discovery and instrumentation, paradigm definition, clinical translation with statistical learning, and the present era of deep learning, foundation models and convergence with the Internet of Things. We compare acquisition modalities on a resolution against invasiveness trade space, chart the decoding literature from common spatial patterns and Riemannian geometry to large pretrained brain models, and quantify assistive throughput, which has risen from roughly two characters per minute in 1988 to a sustained fifty-six words per minute in long-term home use. We then argue that the same convergence responsible for these gains has multiplied the attack surface. Drawing on the classical side-channel literature and on brain-specific results, we propose a four-family taxonomy of leakage covering physical emanation, protocol metadata, semantic elicitation and learned-model channels, and map each onto the processing pipeline. Countermeasures are assessed as a five-tier defence-in-depth stack. We conclude that decoding capability now substantially outpaces the security evidence base, and set out a staged research agenda to close that gap before healthcare deployment scales further.